This paper derives two spatio-temporal extensions of the well-known FastICA algorithm of Hyvarinen and Oja that are applicable to the convolutive blind source separation task. Our time-domain algorithms combine multichannel spatio-temporal prewhitening via multistage least-squares linear prediction with novel adaptive procedures that impose paraunitary, constraints on the multichannel separation filter. The techniques converge quickly to a separation solution without any step size selection or divergence difficulties, and unlike other methods, ours do not require special coefficient initialization procedures to obtain good separation performance. They also allow for the efficient reconstruction of individual signals as observed in the sensor measurements directly from the system parameters for single-input multiple-output blind source separation tasks. An analysis of one of the adaptive constraint procedures shows its fast convergence to a paraunitary filter bank solution. Numerical evaluations of the proposed algorithms and comparisons with several existing convolutive blind source separation techniques indicate the excellent relative performance of the proposed methods.

Acoustical signals are often corrupted by other speeches, sources, and background noise. This makes it necessary to use some form of preprocessing so that signal processing systems such as a speech recognizer or machine diagnosis can be effectively employed. In this contribution, we introduce and evaluate a new algorithm that uses independent component analysis (ICA) with a geometrical constraint [constrained ICA (CICA)]. It is based on the fundamental similarity between an adaptive beamformer and blind source separation with ICA, and does not suffer the permutation problem of ICA-algorithms. Unlike conventional ICA algorithms, CICA needs prior knowledge about the rough direction of the target signal. However, it is more robust against an erroneous estimation of the target direction than adaptive beamformers: CICA converges to the right solution as long as its look direction is closer to the target signal than to the jammer signal. A high degree of robustness is very important since the geometrical prior of an adaptive beamformer is always roughly estimated in a reverberant environment, even when the look direction is precise. The effectiveness and robustness of the new algorithms is proven theoretically, and shown experimentally for three sources and three microphones with several sets of real-world data.

This paper presents a method for enhancing target sources of interest and suppressing other interference sources. The target sources are assumed to be close to sensors, to have dominant powers at these sensors, and to have non-Gaussianity. The enhancement is performed blindly, i.e., without knowing the position and active time of each source. We consider a general case where the total number of sources is larger than the number of sensors, and neither the number of target sources nor the total number of sources is known. The method is based on a two-stage process where independent component analysis (ICA) is first employed in each frequency bin and then time-frequency masking is used to improve the performance further. We propose a new sophisticated method for deciding the number of target sources and then selecting their frequency components. We also propose a new criterion for specifying time-frequency masks. Experimental results for simulated cocktail party situations in a room, whose reverberation time was 130 ms, are presented to show the effectiveness and characteristics of the proposed method.

Natural gradient adaptation is an especially convenient method for adapting the coefficients of a linear system in inverse filtering tasks such as convolutive blind source separation and multichannel blind deconvolution. When developing practical implementations of such methods, however, it is not clear how best to window the signals and truncate the filter impulse responses within the filtered gradient updates. In this paper, we show how inadequate use of truncation of the filter impulse responses and signal windowing within a well-known natural gradient algorithm for multichannel blind deconvolution and source separation can introduce a bias into its steady-state solution. We then provide modifications of this algorithm that effectively mitigate these effects for estimating causal FIR solutions to single- and multichannel equalization and source separation tasks. The new multichannel blind deconvolution algorithm requires approximately 6.5 multiply/adds per adaptive filter coefficient, making its computational complexity about 63% greater than the originally-proposed version. Numerical experiments verify the robust convergence performance of the new method both in multichannel blind deconvolution tasks for i.i.d. sources and in convolutive BSS tasks for real-world acoustic sources, even for extremely-short separation filters.

Blind source separation (BSS) for convolutive mixtures can be solved efficiently in the frequency domain, where independent component analysis (ICA) is performed separately in each frequency bin. However, frequency-domain BSS involves a permutation problem: the permutation ambiguity of ICA in each frequency bin should be aligned so that a separated signal in the time-domain contains frequency components of the same source signal. This paper presents a robust and precise method for solving the permutation problem. It is based on two approaches: direction of arrival (DOA) estimation for sources and the interfrequency correlation of signal envelopes. We discuss the advantages and disadvantages of the two approaches, and integrate them to exploit their respective advantages. Furthermore, by utilizing the harmonics of signals, we make the new method robust even for low frequencies where DOA estimation is inaccurate. We also present a new closed-form formula for estimating DOAs from a separation matrix obtained by ICA. Experimental results show that our method provided an almost perfect solution to the permutation problem for a case where two sources were mixed in a room whose reverberation time was 300 ms.

Despite several recent proposals to achieve blind source separation (BSS) for realistic acoustic signals, the separation performance is still not good. enough. In particular, when the impulse responses are long, performance is highly limited. In this paper, we consider a two-input, two-output convolutive BSS problem. First, we show that it is not good to be constrained by the condition T > P, where T is the frame length of the DFT and P is the length of the room impulse responses. We show that there is an optimum frame size. that is determined by the trade-off between maintaining the number of samples in each frequency bin to estimate statistics and covering the whole reverberation. We also clarify the reason for. the poor performance of BSS in long reverberant environments, highlighting that the framework of BSS works as two sets of frequency-domain adaptive beamformers. Although BSS can reduce reverberant sounds to some extent like adaptive beamformers, they mainly remove the sounds from the jammer direction. This is the reason for the difficulty of BSS in reverberant environments.

Use of a common-acoustical-pole and zero model is proposed for modeling head-related transfer functions (HRTF's) for various directions of sound incidence. The HRTF's are expressed using the common acoustical poles, which do not depend on the source directions, and the zeros, which do, The common acoustical poles are estimated as they are common to HRTF's for various source directions; the estimated values of the poles agree well with the resonance frequencies of the ear canal. Because this model uses only the zeros to express the HRTF variations due to changes in source direction, it requires fewer parameters (the order of the zeros) that depend on the source direction than do the conventional all zero or pole/zero models. Furthermore, the proposed model can extract the zeros that are missed in the conventional models because of pole-zero cancellation. As a result, the directional dependence of the zeros can be traced well. Analysis of the zeros for HRTF's on the horizontal plane showed that the nonminimum-phase zero variation was well formulated using a simple pinna-reflection model, The common-acoustical-pole and zero (CAPZ) model is thus effective for modeling and analyzing HRTF's.

This paper describes a block (affine) projection algorithm that has exactly the same convergence rate as the original sample-by-sample algorithm and smaller computational complexity than the fast affine projection algorithm. This is achieved by 1) introducing a correction term that compensates for the filter output difference between the sample-by-sample projection algorithm and the straightforward block projection algorithm, and 2) applying a fast finite impulse response (FIR) filtering technique to compute filter outputs and to update the filter.
We describe how to choose a pair of block lengths that gives the longest filter length under a constraint on the total computational complexity and processing delay. An example shows that the filter length can be doubled if a delay of a few hundred samples is permissible.

A new model for a room transfer function (RTF) by using common acoustical poles that correspond to resonance properties of a room is proposed. These poles are estimated as the common values of many RTF's corresponding to different source and receiver positions. Since there is one-to-one correspondence between poles and AR coefficients, these poles are calculated as common AR coefficients by two methods: i) using the least squares method, assuming all the given multiple RTF's have the same AR coefficients and ii) averaging each set of AR coefficients estimated from each RTF. The estimated poles agree well with the theoretical poles when estimated with the same order as the theoretical pole order. When estimated with a lower order than the theoretical pole order, the estimated poles correspond to the major resonance frequencies, which have high Q factors. Using the estimated common AR coefficients, the proposed method models the RTF's with different MA coefficients. This model is called the common-acoustical-pole and zero (CAPZ) model, and it requires far fewer variable parameters to represent RTF's than the conventional all-zero or pole/zero model. This model was used for an acoustic echo canceller at low frequencies, as one example. The acoustic echo canceller based on the proposed model requires half the variable parameters and converges 1.5 times faster than one based on the all-zero model, confirming the efficiency of the proposed model.

This paper proposes a new normalized least-mean-squares (NLMS) adaptive algorithm with double the convergence speed, at the same computational load, of the conventional NLMS for an acoustic echo canceller. This algorithm, called the ES (exponentially weighted stepsize) algorithm, uses a different stepsize (feedback constant) for each weight of an adaptive transversal filter. These stepsizes are time-invariant and weighted proportional to the expected variation of a room impulse response. The algorithm is based on the fact that the expected variation of a room impulse response becomes progressively smaller along the series by the same exponential ratio as the impulse response energy decay. As a result, the algorithm adjusts coefficients with large errors in large steps, and coefficients with small errors in small steps. A transition formula is derived for the mean-squared coefficient error of the proposed algorithm. The mean stepsize determines the convergence condition, the convergence speed, and the final excess mean-squared error. The algorithm is modified for a practical multiple DSP structure, so that it requires only the same amount of computation as the conventional NLMS. The algorithm is implemented in a commercial acoustic echo canceller and its fast convergence is demonstrated.

We previously proposed the virtual microphone technique to improve speech enhancement performance in underdetermined situations, in which the number of channels is virtually increased by estimating extra microphone signals at arbitrary positions along the straight line formed by real microphones. The effectiveness of the interpolation of virtual microphone signals for speech enhancement was experimentally confirmed. In this work, we apply the extrapolation of a virtual microphone as preprocessing of the maximum signal-to-noise ratio (SNR) beamformer and compare its speech enhancement performance (the signal-to-distortion ratio (SDR) and signal-to-interference ratio (SIR)) with that of using the interpolation of a virtual microphone. Furthermore, we aim to improve speech enhancement performance by solving a trade-off relationship between performance at low and high frequencies, which can be controlled by adjusting the virtual microphone interval. We propose a new arrangement where a virtual microphone is placed at a distance from the reference real microphone proportional to the wavelength at each frequency. From the results of our experiment in an underdetermined situation, we confirmed speech enhancement performance using the extrapolation of a virtual microphone is higher than that of using the interpolation of a virtual microphone. Moreover, the proposed wavelength-proportional interpolation and extrapolation method improves speech enhancement performance compared with the interpolation and extrapolation. Furthermore, we present the directivity patterns of a spatial filter and confirmed the behavior that improves speech enhancement performance.

This paper presents a new low-latency online blind source separation (BSS) algorithm. Although algorithmic delay of a frequency domain online BSS can be reduced simply by shortening the short-time Fourier transform (STFT) frame length, it degrades the source separation performance in the presence of reverberation. This paper proposes a method to solve this problem by integrating BSS with Weighted Prediction Error (WPE) based dereverberation. Although a simple cascade of online BSS after online WPE upgrades the separation performance, the overall optimality is not guaranteed. Instead, this paper extends a recently proposed batch processing algorithm that can jointly optimize dereverberation and separation so that it can perform online processing with low computational cost and little processing delay (&lt
12 ms). The results of a source separation experiment in a noisy car environment suggest that the proposed online method has better separation performance than the simple cascaded methods.

In this paper, we propose SepNet, a deep neural network (DNN) designed to predict separation matrices from multichannel observations. One well-known approach to blind source separation (BSS) involves independent component analysis (ICA). A recently developed method called independent low-rank matrix analysis (ILRMA) is one of its powerful variants. These methods allow the estimation of separation matrices based on deterministic iterative algorithms. Specifically, ILRMA is designed to update the separation matrix according to an update rule derived based on the majorization-minimization principle. Although ILRMA performs reasonably well under some conditions, there is still room for improvement in terms of both separation accuracy and computation time, especially for large-scale microphone arrays. The existence of a deterministic iterative algorithm that can find one of the stationary points of the BSS problem implies that a DNN can also play that role if designed and trained properly. Motivated by this, we propose introducing a DNN that learns to convert a predefined input (e.g., an identity matrix) into a true separation matrix in accordance with a multichannel observation. To enable it to find one of the multiple solutions corresponding to different permutations of the source indices, we further propose adopting a permutation invariant training strategy to train the network. By using a fully convolutional architecture, we can design the network so that the forward propagation can be computed efficiently. The experimental results revealed that SepNet was able to find separation matrices faster and with better separation accuracy than ILRMA for mixtures of two sources.

Emotion recognition is essential for human behavior analysis and possible through various inputs such as speech and images. However, in practical situations, such as in call center analysis, the available information is limited to speech. This leads to the study of speech emotion recognition (SER). Considering the complexity of emotions, SER is a challenging task. Recently, automatic speech recognition (ASR) has played a role in obtaining text information from speech. The combination of speech and ASR results has improved the SER performance. However, ASR results are highly affected by speech recognition errors. Although there is a method to improve ASR performance on emotional speech, it requires the fine-tuning of ASR, which is costly. To mitigate the errors in SER using ASR systems, we propose the use of the combination of a self-attention mechanism and a word-level confidence measure (CM), which indicates the reliability of ASR results, to reduce the importance of words with a high chance of error. Experimental results confirmed that the combination of self-attention mechanism and CM reduced the effects of incorrectly recognized words in ASR results, providing a better focus on words that determine emotion recognition. Our proposed method outperformed the stateof- the-art methods on the IEMOCAP dataset.

This paper proposes a multichannel source separation approach, which uses a star generative adversarial network (StarGAN) to model power spectrograms of sources. Various studies have shown the significant contributions of a precise source model to the performance improvement in audio source separation, which indicates the importance of developing a better source model. In this paper, we explore the potential of StarGAN for modeling source spectrograms and investigate the effectiveness of the StarGAN source model in determined multichannel source separation by incorporating it into a frequency-domain independent component analysis (ICA) framework. The experimental results reveal that the proposed StarGAN-based method outperformed conventional methods that use non-negative matrix factorization (NMF) or a variational autoencoder (VAE) for source spectrogram modeling.

Acoustic scene classification (ASC) classifies the place or situation where an acoustic sound was recorded. The Detection and Classification of Acoustic Scenes and Events (DCASE) 2017 Challenge prepared a task involving ASC. Some methods using convolutional neural networks (CNNs) were proposed in the DCASE 2017 Challenge. The best method independently performed convolution operations for the left, right, mid (addition of left and right channels), and side (subtraction of left and right channels) input channels to capture spatial features. On the other hand, we propose a new method of spatial feature extraction using CNNs. In the proposed method, convolutions are performed for the time-space (channel) domain and frequency-space domain in addition to the time-frequency domain to capture spatial features. We evaluate the effectiveness of the proposed method using the task in the DCASE 2017 Challenge. The experimental results confirmed that convolution operations for the frequency-space domain are effective for capturing spatial features. Furthermore, by using a combination of the three domains, the classification accuracy was improved by 2.19% compared with that obtained using the tim

<p>ビームフォーミングなどの従来のアレー信号処理による雑音抑圧手法は，位相情報を活用した指向性制御に基づいており，特定方向から到来する雑音に対しては指向性の零点を向けることで高い効果が得られる。しかし，到来方向が特定できないような，いわゆる背景雑音の抑圧は，一般に難しかった。本論文では，伝達関数ゲイン基底NMFにより，遠方から到来する雑音を複数マイクを用いて効果的に抑圧する手法を提案する。提案手法では，背景雑音が遠方から到来することを仮定し，時間周波数領域における振幅情報のみに着目することで，様々な方向から到来する遠方音源を一つの混合音源としてモデル化する。次にこの振幅の混合モデルを従来提案されている制約付き伝達関数ゲイン基底NMFに適用し，遠方音源の抑圧を行う。更に，半教師あり伝達関数ゲイン基底NMFを適用し，遠方音源の抑圧を行う。本手法は振幅情報のみを用いているため，非同期録音機器を用いることができ

<p>In this paper, we propose a novel blind source separation method for the hose-shaped rescue robot based on independent low-rank matrix analysis and statistical speech enhancement. The rescue robot is aimed to detect victims'speech in a disaster area, wearing multiple microphones around the body. Different from the common microphone array, the positions of microphones are unknown, and the conventional beamformer cannot be utilized. In addition, the vibration noise (ego-noise) is generated when the robot moves, yielding the serious contamination in the observed signals. Therefore, it is important to eliminate the ego-noise in this system. This paper describes our newly developed software and hardware system of blind source separation for the robot noise reduction. Also, we report objective and subjective evaluation results showing that the proposed system outperforms the conventional methods in the source separation accuracy and perceptual sound quality via experiments with actual sounds observed in the rescue robot.</p>

This paper presents the design and implementation of a two-stage human-voice enhancement system for a hose-shaped rescue robot. When a microphoneequipped hose-shaped robot is used to search for a victim under a collapsed building, human-voice enhancement is crucial because the sound captured by a microphone array is contaminated by the ego-noise of the robot. For achieving both low latency and high quality, our system combines online and offline human-voice enhancement, providing an overview first and then details on demand. The online enhancement is used for searching for a victim in real time, while the offline one facilitates scrutiny by listening to highly enhanced human voices. Our online enhancement is based on an online robust principal component analysis, and our offline enhancement is based on an independent lowrank matrix analysis. The two enhancement methods are integrated with Robot Operating System (ROS). Experimental results showed that both the online and offline enhancement methods outperformed conventional methods.

Non-negative matrix factorization (NMF) is a powerful approach to single channel audio source separation. In a supervised setting, NMF is first applied to train the basis spectra of each sound source. At test time, NMF is applied to the spectrogram of a mixture signal using the pretrained spectra. The source signals can then be separated out using a Wiener filter. A typical way to train the basis spectra of each source is to minimize the objective function of NMF. However, the basis spectra obtained in this way do not ensure that the separated signal will be optimal at test time due to the inconsistency between the objective functions for training and separation (Wiener filtering). To address this, a framework called discriminative NMF (DNMF) has recently been proposed. In in this work a multiplicative update algorithm was proposed for the basis training, however one drawback is that the convergence is not guaranteed. To overcome this drawback, this paper proposes using a majorization-minimization principle to develop a convergence-guaranteed algorithm for DNMF. Experimental results showed that the proposed algorithm outperformed standard NMF and DNMF using a multiplicative update algorithm as regards both the signal-to-distortion and signal-to-interference ratios.

Spectral domain speech enhancement algorithms based on nonnegative spectrogram models such as non-negative matrix factorization (NMF) and non-negative matrix factor deconvolution are powerful in terms of signal recovery accuracy, however they do not directly lead to an enhancement in the feature domain (e.g., cepstral domain) or in terms of perceived quality. We have previously proposed a method that makes it possible to enhance speech in the spectral and cepstral domains simultaneously. Although this method was shown to be effective, the devised algorithm was computationally demanding. This paper proposes yet another formulation that allows for a fast implementation by replacing the regularization term with a divergence measure between the NMF model and the mel-generalized cepstral (MGC) representation of the target spectrum. Since the MGC is an auditory-motivated representation of an audio signal widely used in parametric speech synthesis, we also expect the proposed method to have an effect in enhancing the perceived quality. Experimental results revealed the effectiveness of the proposed method in terms of both the signal-To-distortion ratio and the cepstral distance.

<p>本論文では，時間チャネル領域の非負値行列因子分解（NMF）による，非同期分散型録音の目的音強調手法について述べる。複数の録音機器による多チャネル信号は，機器ごとのサンプリング周波数の微小なずれが引き起こす位相差のドリフトのため，位相情報を用いるアレー信号処理は適さない。位相に比べると振幅の分析はドリフトの影響を大きく受けないことに着目し，戸上らが提案した時間チャネル領域のNMFによるチャネル間ゲイン差の分析（伝達関数ゲイン基底NMF）に基づく時間周波数マスクを用いる。また，基底数よりも十分大きなチャネル数が得られない条件の音声強調のための，基底を事前に学習する教師ありNMFについて議論する。</p>

MUltiple Signal Classification (MUSIC) is a standard technique for direction of arrival (DOA) estimation with high resolution. However, MUSIC cannot estimate DOAs accurately in the case of underdetermined conditions, where the number of sources exceeds the number of microphones. To overcome this drawback, an extension of MUSIC using cumulants called 2q-MUSIC has been proposed, but this method greatly suffers from the variance of the statistics, given as the temporal mean of the observation process, and requires long observation. In this paper, we propose a new approach for extending MUSIC that exploits higher-order moments of the signal for the underdetermined DOA estimation with smaller variance. We propose an estimation algorithm that nonlinearly maps the observed signal onto a space with expanded dimensionality and conducts MUSIC-based correlation analysis in the expanded space. Since the dimensionality of the noise subspace is increased by the mapping, the proposed method enables the estimation of DOAs in the case of underdetermined conditions. Furthermore, we describe the class of mapping that allows us to analyze the higher-order moments of the observed signal in the original space. We compare 2q-MUSIC and the proposed method through an experiment assuming that the true number of sources is known as prior information to evaluate in terms of the bias-variance tradeoff of the statistics and computational complexity. The results clarify that the proposed method has advantages for both computational complexity and estimation accuracy in short-time analysis, i.e., the time duration of the analyzed data is short.

MUltiple SIgnal Classification (MUSIC) is a standard technique for direction of arrival (DOA) estimation with high resolution. However, MUSIC cannot estimate DOAs accurately in the case of underdetermined conditions, where the number of sources exceeds the number of microphones. To overcome this drawback, an extension of MUSIC using cumulants called 2q-MUSIC has been proposed, but this method greatly suffers from the variance of the statistics, given as the temporal mean of the observation process, and requires long observation. In this paper, we propose a new approach for extending MUSIC that exploits higher-order moments of the signal for the underdetermined DOA estimation with smaller variance. We propose an estimation algorithm that nonlinearly maps the observed signal onto a space with expanded dimensionality and conducts MUSIC-based correlation analysis in the expanded space. Since the dimensionality of the noise subspace is increased by the mapping, the proposed method enables the estimation of DOAs in the case of underdetermined conditions. Furthermore, we describe the class of mapping that allows us to analyze the higher-order moments of the observed signal in the original space. We compare 2q-MUSIC and the proposed method through an experiment assuming that the true number of sources is known as prior information to evaluate in terms of the bias-variance tradeoff of the statistics and computational complexity. The results clarify that the proposed method has advantages for both computational complexity and estimation accuracy in short-time analysis, i.e., the time duration of the analyzed data is short.

<p>This paper presents a noise reduction on a hose-shaped rescue robot. The hose-shaped rescue robot is one of rescue robots developed on Tough Robotics Challenge, and it is used for searching for victims by getting one's voice with its microphone-array. However, the ego noise, caused by its vibration motors, makes it difficult to get the human voice. We propose a noise reduction method using a blind source separation technique based on Independent Vector Analysis (IVA) and noise cancellation. Our method consists of two steps: (1) estimating a speech signal and an ego-noise signal from observed multi-channel signals using the IVA-based blind source separation technique, and (2) applying the noise cancellation to the estimated speech signal using the estimated ego-noise signal as a noise reference.</p>

<p>A hose-shaped rescue robot is one of the robots that are developed for disaster response in case of a large-scale disasters such as a great earthquake. The robot is suitable for entering narrow and dark places covered with rubble in the disaster site, and for finding inside it. This robot can transmit the ambient sound to its operator by using the built-in microphones. However, there is a serious problem that the inherent noise of this robot, such as the vibration sound or the fricative sound, is mixed into the transmitting voice, therefore disturbing the operator's hearing for a call of help from the victim of the disaster. In this paper, we apply the multichannel NMF (nonnegative matrix factorization) with the rank-1 spatial constraint (Rank-1 MNMF), which was proposed by Kitamura et al., to the reduction of the inherent noise.</p>

Previously, methods for estimating the performance of noisy speech recognition based on a spectral distortion measure have been proposed. Although they give an estimate of recognition performance without actually performing speech recognition, no consideration is given to any change in the components of a speech recognition system. To solve this problem, we propose a novel method for estimating the performance of noisy speech recognition, a major feature of which is the ability to accommodate the use of different noise reduction algorithms and recognition tasks by using two cepstral distances (CDs) and the square mean root perplexity (SMR-perplexity). First, we verified the effectiveness of the proposed distortion measure, i.e., the two CDs. The experimental results showed that the use of the proposed distortion measure achieves estimation accuracy equivalent to the use of the conventional distortion measures, the perceptual evaluation of speech quality (PESQ) and the signal-to-noise ratio (SNR) of noise-reduced speech, and has the advantage of being applicable to noise reduction algorithms that directly output the mel-frequency cepstral coefficient (MFCC) feature. We then evaluated the proposed method by performing a closed test and an open test (10-fold crossvalidation test). The results confirmed that the proposed method gives better estimates without being dependent on the differences among the noise reduction algorithms or the recognition tasks.

To ensure a satisfactory QoE (Quality of Experience), it is essential to establish a method that can be used to efficiently investigate recognition performance for spontaneous speech. By using this method, it is allowed to monitor the recognition performance in providing speech recognition services. It can be also used as a reliability measure in speech dialogue systems. Previously, methods for estimating the performance of noisy speech recognition based on spectral distortion measures have been proposed. Although they give an estimate of recognition performance without actually performing speech recognition, the methods cannot be applied to spontaneous speech because they require the reference speech to obtain the distortion values. To solve this problem, we propose a novel method for estimating the recognition performance of spontaneous speech with various speaking styles. The main feature is to use non-reference acoustic features that do not require the reference speech. The proposed method extracts non-reference features by openSMILE (open-Source Media Interpretation by Large feature-space Extraction) and then estimates the recognition performance by using SVR (Support Vector Regression). We confirmed the effectiveness of the proposed method by experiments using spontaneous speech data from the OGVC (On-line Gaming Voice Chat) corpus.

In this paper, we present noise reduction for a hose-shaped rescue robot. The robot is used for searching for disaster victims by capturing their voice with its microphone array. However, the ego noise generated by its vibration motors makes it difficult to distinguish human voices. To solve this problem, we propose a noise reduction method using a blind source separation technique based on independent vector analysis (IVA) and noise cancellation. Our method consists of two steps: (1) estimating a speech signal and an ego noise signal from observed multichannel signals using the IVA-based blind source separation technique, and (2) applying noise cancellation to the estimated speech signal using the estimated ego noise signal as a noise reference. The experimental evaluations show that this approach is effective for suppressing the ego noise.

In this paper, we propose a new microphone array signal processing technique, which increases the number of microphones virtually by generating extra signal channels from real microphone signals. Microphone array signal processing methods such as speech enhancement are effective for improving the quality of various speech applications such as speech recognition and voice communication systems. However, the performance of speech enhancement and other signal processing methods depends on the number of microphones. Thus, special equipment such as a multichannel A/D converter or a microphone array is needed to achieve high processing performance. Therefore, our aim was to establish a technique for improving the performance of array signal processing with a small number of microphones and, in particular, to increase the number of channels virtually by synthesizing virtual microphone signals, or extra signal channels, from two channels of microphone signals. Each virtual microphone signal is generated by interpolating a short-time Fourier transform (STFT) representation of the microphone signals. The phase and amplitude of the signal are interpolated individually. The phase is linearly interpolated on the basis of a sound propagation model, and the amplitude is nonlinearly interpolated on the basis of beta divergence. We also performed speech enhancement experiments using a maximum signal-to-noise ratio (SNR) beamformer equipped with virtual microphones and evaluated the improvement in performance upon introducing virtual microphones.

A hose-shaped rescue robot is one of the robots that have been developed for disaster response in times of large-scale disasters such as a massive earthquake. This robot is suitable for entering narrow and dark places covered with rubble in a disaster site and for finding victims inside it. It can transmit ambient sound captured by its built-in microphones to its operator. However, there is a serious problem, that is, the inherent noise of this robot, such as vibration sound or fricative sound, is mixed with the transmitted voice, thereby disturbing the operator's perception of a call for help from a disaster victim. In this paper, we apply the multichannel nonnegative matrix factorization (NMF) with the rank-1 spatial constraint (determined rank-1 MNMF), which was proposed by Kitamura et al., to the reduction of the inherent noise.

In this paper, we present a multi-talker speech recognition system based on blind source separation with an ad hoc microphone array, which consists of smartphones and cloud storage. In this system, a mixture of voices from multiple speakers is recorded by each speaker's smartphone, which is automatically transferred to online cloud storage. Our prototype system is realized using iPhone and Dropbox. Although the signals recorded by different iPhones are not synchronized, the blind synchronization technique compensates both the differences in the time offset and the sampling frequency mismatch. Then, auxiliary-function-based independent vector analysis separates the synchronized mixture into each speaker's voice. Finally, automatic speech recognition is applied to transcribe the speech. By experimental evaluation of the multi-talker speech recognition system using Julius, we confirm that it effectively reduces the speech overlap and improves the speech recognition performance.

In this study we propose a novel stimulus-driven brain-computer interface (BCI) paradigm, which generates control commands based on classification of somatosensory modality P300 responses. Six spatial vibrotactile stimulus patterns are applied to entire back and limbs of a user. The aim of the current project is to validate an effectiveness of the vibrotactile stimulus patterns for BCI purposes and to establish a novel concept of tactile modality communication link, which shall help locked-in syndrome (LIS) patients, who lose their sight and hearing due to sensory disabilities. We define this approach as a full-body BCI (fbBCI) and we conduct psychophysical stimulus evaluation and realtime EEG response classification experiments with ten healthy body-able users. The grand mean averaged psychophysical stimulus pattern recognition accuracy have resulted at 9 8 : 1 8 %, whereas the realtime EEG accuracy at 5 3 : 6 7 %. An information-transfer-rate (ITR) scores of all the tested users have ranged from 0 : 0 4 2 to 4 : 1 5 4 bit/minute.

In this paper, we present an ego noise reduction method for a hose-shaped rescue robot developed for search and rescue operations in large-scale disasters such as a massive earthquake. It can enter narrow and dark places covered with rubble in a disaster site and is used to search for disaster victims by capturing their voices with its microphone array. However, ego noises, such as vibration or fricative sounds, are mixed with the voices, and it is difficult to differentiate them from a call for help from a disaster victim. To solve this problem, we here propose a two-step noise reduction method as follows: (1) the estimation of both speech and ego noise signals from an observed multichannel signal by multichannel nonnegative matrix factorization (NMF) with the rank-1 spatial constraint, which was proposed by Kitamura et al., and (2) the application of noise cancellation to the estimated speech signal using the noise reference. Our evaluations show that this approach is effective for suppressing ego noise.

We propose a method for separating speech signals when sources outnumber the sensors. In this paper we mainly concentrate on the case of three sources and two sensors. Some existing methods employ binary masks to extract the signals, and therefore, the extracted signals contain loud musical noise. To overcome this problem, we propose the utilization of a directivity pattern based continuous mask, which removes a single source from the observations, and independent component analysis (ICA) to separate the remaining mixtures. Experimental results show that our proposed method can separate signals with little distortion even in a real reverberant environment of T R =130 ms.

In this paper, we propose a novel method for the blind compensation of drift for the asynchronous recording of an ad hoc microphone array. Digital signals simultaneously observed by different recording devices have drift of the time differences between the observation channels because of the sampling frequency mismatch among the devices. On the basis of a model in which the time difference is constant within each short time frame but varies in proportion to the central time of the frame, the effect of the sampling frequency mismatch can be compensated in the short-time Fourier transform (STFT) domain by a linear phase shift. By assuming that the sources are motionless and have stationary amplitudes, the observation is regarded as being stationary when drift does not occur. Thus, we formulate a likelihood to evaluate the stationarity in the STFT domain to evaluate the compensation of drift. The maximum likelihood estimation is obtained effectively by a golden section search. Using the estimated parameters, we compensate the drift by STFT analysis with a noninteger frame shift. The effectiveness of the proposed blind drift compensation method is evaluated in an experiment in which artificial drift is generated. (C) 2014 The Authors. Published by Elsevier B.V.

The developed acoustic echo canceller uses an exponentially weighted step-size projection algorithm and a duo-filter control system to achieve fast convergence and high speech quality. The duo-filter control system has an adaptive filter and a fixed filter, and uses variable-loss insertion. Evaluation of this system with multi-channel A/D and D/A converters showed that (1) the convergence speed is under 1.5 seconds for speech input when the adaptive filter length is 125 ms, (2) the residual echo level is nearly as low as the ambient noise level (average: Under -20 dB
maximum: Under -35 dB), and (3) near-end speech is sent with no disturbance during double talk.

© 2013, Springer Science+Business Media New York. We propose a novel method for the extraction of discriminative features in electroencephalography (EEG) evoked potential latency. Based on our offline results, we present evidence indicating that a full surround sound auditory brain–computer interface (BCI) paradigm has potential for an online application. The auditory spatial BCI concept is based on an eight-directional audio stimuli delivery technique, developed by our group, which employs a loudspeaker array in an octagonal horizontal plane. The stimuli presented to the subjects vary in frequency and timbre. To capture brain responses, we utilize an eight-channel EEG system. We propose a methodology for finding and optimizing evoked response latencies in the P300 range in order later to classify them correctly and to elucidate the subject’s chosen targets or ignored non-targets. To accomplish the above, we propose an approach based on an analysis of variance for feature selection. Finally, we identify the subjects’ intended commands with a Naive Bayesian classifier for sorting the final responses. The results obtained with ten subjects in offline BCI experiments support our research hypothesis by providing higher classification results and an improved information transfer rate compared with state-of-the-art solutions.

We present results of an approach to a code-modulated visual evoked potential (cVEP) based braincomputer interface (BCI) paradigm using four high-frequency flashing stimuli. To generate higher frequency stimulation compared to the state-of-the-art cVEP-based BCIs, we propose to use the light-emitting diodes (LEDs) driven from a small micro-controller board hardware generator designed by our team. The high-frequency and green-blue chromatic flashing stimuli are used in the study in order to minimize a danger of a photosensitive epilepsy (PSE). We compare the the green-blue chromatic cVEP-based BCI accuracies with the conventional white-black flicker based interface. The high-frequency cVEP responses are identified using a canonical correlation analysis (CCA) method.

© Springer International Publishing Switzerland 2015. We present results of a classification improvement approach for a code–modulated visual evoked potential (cVEP) based brain– computer interface (BCI) paradigm using four high–frequency flashing stimuli. Previously published research reports presented successful BCI applications of canonical correlation analysis (CCA) to steady–state visual evoked potential (SSVEP) BCIs. Our team already previously proposed the combined CCA and cVEP techniques’ BCI paradigm. The currently reported study presents the further enhanced results using a support vector machine (SVM) method in application to the cVEP–based BCI.

The reported project aims to confirm whether a tactile glove fingertips' stimulator is effective for a brain-computer interface (BCI) paradigm using somatosensory event potential (SEP) responses with possible attentional modulation. The proposed simplified stimulator device is presented in detail together with psychophysical and EEG BCI experiment protocols. Results supporting the proposed simple tactile glove device are presented in form of online BCI classification accuracy results using shrinkage linear discriminant analysis (sLDA) technique. Finally, we discuss future possible paradigm improvement steps.

In this paper, we propose a method to estimate a correlation coefficient of two correlated complex signals on the condition that only the amplitudes are observed and the phases are missing. Our proposed method is based on a maximum likelihood estimation. We assume that the original complex random variables are generated from a zero-mean bivariate complex normal distribution. The likelihood of the correlation coefficient is formulated as a bivariate Rayleigh distribution by marginalization over the phases. Although the maximum likelihood estimator has no analytical form, an expectation-maximization (EM) algorithm can be formulated by treating the phases as hidden variables. We evaluate the accuracy of the estimation using artificial signal, and demonstrate the estimation of narrow-band correlation of a two-channel audio signal.

The paper presents a study of an inter-stimulus interval (ISI) influence on a tactile point-pressure stimulus-based brain-computer interface's (tpBCI) classification accuracy. A novel tactile pressure generating tpBCI stimulator is also discussed, which is based on a three-by-three pins' matrix prototype. The six pin-linear patterns are presented to the user's palm during the online tpBCI experiments in an oddball style paradigm allowing for "the aha-responses" elucidation, within the event related potential (ERP). A subsequent classification accuracies' comparison is discussed based on two ISI settings in an online tpBCI application. A research hypothesis of classification accuracies' non-significant differences with various ISIs is confirmed based on the two settings of 120 ms and 300 ms, as well as with various numbers of ERP response averaging scenarios.

In this paper, we propose a method for suppressing a large number of interferences by using multichannel amplitude analysis based on nonnegative matrix factorization (NMF) and its effective semi-supervised training. For the point-source interference reduction of an asynchronous microphone array, we propose amplitude-based speech enhancement in the time-channel domain, which we call transfer-function-gain NMF. Transfer-function-gain NMF is a robust method against drift, which disrupts an inter-channel phase analysis. We use this method to suppress a large number of sources. We show that a mass of interferences can be modeled by a single basis assuming that the noise sources are sufficiently far from the microphones and the spatial characteristics become similar to each other. Since the blind optimization of the NMF parameters does not work well with merely sparse observation contaminated by the constant heavy noise, we train the diffuse noise basis in advance of the noise suppression using a speech absent observation, which can be obtained easily using a simple voice activity detection technique. We confirmed the effectiveness of our proposed model and semi-supervised transfer-function-gain NMF in an experiment simulating a target source that was surrounded by a diffuse noise.

This paper presents a study of classification and EEG feature improvement for a spatial auditory brain-computer interface (saBCI). This study provides a comprehensive test of a head-related impulse response (HRIR) cues for the saBCI speller paradigm. We present a comparison with previously developed HRIR-based spatial auditory modalities. We propose and optimize the three types of sound spatialization settings using a variable elevation in order to evaluate the HRIR efficacy for the saBCI. Three experienced and seven naive BCI users participate in the three experimental setups based on ten presented Japanese syllables. The obtained EEG auditory evoked potentials (AEPs) result with encouragingly good and stable P300 responses in online saBCI experiments. We analyze the differences and dispersions of saBCI command accuracies, as well as the individual user accuracies for various spatial sound locations. Our case study indicates that the participating users could perceive elevation in the saBCI experiments using the HRIR measured from a general head model.

This study provides a comprehensive test of a head-related impulse response (HRIR) cues for a spatial auditory brain-computer interface (saBCI) speller paradigm. We present a comparison with the conventional virtual sound headphone-based spatial auditory modality. We propose and optimize the three types of sound spatialization settings using a variable elevation in order to evaluate the HRIR efficacy for the saBCI. Three experienced and seven naive BCI users participated in the three experimental setups based on ten presented Japanese syllables. The obtained EEG auditory evoked potentials (AEP) resulted with encouragingly good and stable P300 responses in online BCI experiments. Our case study indicated that users could perceive elevation in the saBCI experiments generated using the HRIR measured from a general head model. The saBCI accuracy and information transfer rate (ITR) scores have been improved comparing to the classical horizontal plane-based virtual spatial sound reproduction modality, as far as the healthy users in the current pilot study are concerned.

© Springer International Publishing Switzerland 2015. We present visual BCI classification accuracy improved results after application of high- and low-pass filters to an electroen- cephalogram (EEG) containing code-modulated visual evoked poten- tials (cVEPs). The cVEP responses are applied for the brain-computer interface (BCI) in four commands paradigm mode. The purpose of this project is to enhance BCI accuracy using only the single trial cVEP response. We also aim at identification of the most discriminable EEG bands suitable for the broadband visual stimuli. We report results from a pilot study optimizing the EEG filtering using infinite impulse response filters in application to feature extraction for a linear support vector machine (SVM) classification method. The goal of the presented study is to develop a faster and more reliable BCI to further enhance the sym- biotic relationships between humans and computers.

We present a study of a support vector machine (SVM) application to brain-computer interface (BCI) paradigm. Four SVM kernel functions are evaluated in order to maximize classification accuracy of a four classes-based BCI paradigm utilizing a code-modulated visual evoked potential (cVEP) response within the captured EEG signals. Our previously published reports applied only the linear SVM, which already outperformed a more classical technique of a canonical correlation analysis (CCA). In the current study we additionally test and compare classification accuracies of polynomial, radial basis and sigmoid kernels, together with the classical linear (non-kernel-based) SVMs in application to the cVEP BCI.

We present a concept and a pilot study of a two-step input speller application combined with a spatial auditory brain-computer interface (BCI) for locked-in syndrome (LIS) users. The application has been developed for 25 Japanese syllabary (hiragana) characters using a two-step input procedure, in order to create an easy-to-use BCI-speller interface. In the proposed procedure, the user first selects the representative letter of a subset, defining the second step. In the second step, the final choice is made. At each interfacing step, the user's intentional choices are classified based on the P300 event related potential (ERP) responses captured in the EEG, as in the classic oddball paradigm. The BCI experiment and EEG results of the pilot study confirm the effectiveness of the proposed spelling method. (C) 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

本論文ではEMアルゴリズムを用いたスパース性に基づく劣決定ブラインド音源分離(BSS)の計算を高速化する手法を提案する。Izumi et al.は,雑音・残響下でも頑健な劣決定BSSを提案したが,到来時間差パラメータをイタレーションごとに離散全探索で求める更新則のために計算量に問題があった。そこで,到来時間差パラメータが周波数に依存すると捉えた,時間差パラメータが解析的に更新される計算量の少ない更新則を提案する。また,帯域重み付け平均による帯域非依存到来時間差推定によってパラメータ数を削減し,収束性を向上させる。実験により,提案手法が計算時間を1/10程度に削減することを確認した。

Microphone array signal processing is a framework for source localization, source enhancement and source separation with processing multichannel observations achieved using multiple microphones, which are difficult using a single microphone. In microphone array signal processing, the small time difference between channels is a very important cue to obtain spatial information. Therefore, a multichannel A-D converter has been conventionally essential for synchronized observation. On the other hand, if microphone array signal processing can be performed using asynchronous recording devices such as laptop PCs, voice recorders, or smart phones, which are easily available in daily life, it would enhance convenience and increase the number of possible applications markedly. In this review, focusing on a new trend of microphone array signal processing using asynchronously recording devices, we survey existing works and also introduce our approach.

In this paper, we present a generalization of the virtual microphone array we previously proposed to increase the microphone elements by nonlinear interpolation. In the previous work, we generated a virtual observation from two actual microphones by an interpolation in the logarithmic domain. This corresponds to a linear interpolation of the phase and the geometric mean of the amplitude. In this paper, we generalize this interpolation using a linear interpolation of the phase and a nonlinear interpolation of the amplitude with adjustable nonlinearity based on beta-divergence. Improvement of the array signal processing performance is obtained by appropriate tuning of the parameter beta. We evaluate the improvement in speech enhancement using a maximum SNR beamformer.

In this paper, we investigate amplitude-based speech enhancement for asynchronous distributed recording. In an ad-hoc microphone array context, it is supposed that different asynchronous devices record speech. As a result, the phase information is unreliable due to sampling frequency mismatch. For speech enhancement based on the amplitude information instead of the phase information, supervised nonnegative matrix factorization (NMF) is introduced in the time-channel domain. The basis vectors, which represents the gain of the transfer function from a source to each microphone, are trained in advance by using single source observation. The experimental evaluations show that this approach is well robust against the sampling frequency mismatch.

We present a concept and a pilot study of a two-step input speller application combined with a spatial auditory brain-computer interface (BCI) for locked-in syndrome (LIS) users. The application has been developed for 25 Japanese syllabary (hiragana) characters using a two-step input procedure, in order to create an easy-to-use BCI-speller interface. In the proposed procedure, the user first selects the representative letter of a subset, defining the second step. In the second step, the final choice is made. At each interfacing step, the user's intentional choices are classified based on the P300 event related potential (ERP) responses captured in the EEG, as in the classic oddball paradigm. The BCI experiment and EEG results of the pilot study confirm the effectiveness of the proposed spelling method. (C) 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

A novel approach to steady-state visual evoked potential (SSVEP) based brain-computer interface (BCI) is presented in the paper. To minimize possible side effects of the monochromatic light SSVEP-based BCI we propose to utilize chromatic green blue flicker stimuli in higher, comparing to the traditionally used, frequencies. The developed safer SSVEP responses are processed an classified with features drawn from EEG power spectra. Results obtained from healthy users support the research hypothesis of the chromatic and higher frequency SSVEP. The feasibility of proposed method is evaluated in a comparison of monochromatic versus chromatic SSVEP responses. We also present preliminary results with empirical mode decomposition (EMD) adaptive filtering which resulted with improved classification accuracies.

This paper reports on a project aiming to confirm whether a tactile stimulator "touch sense glove" is effective for a novel brain computer interface (BCI) paradigm and whether the tactile stimulus delivered to the fingers could be utilized to evoke event related potential (ERP) responses with possible attentional modulation. The tactile ERPs are expected to improve the BCI accuracy. The proposed new stimulator device is presented in detail together with psychophysical and EEG BCI experiment protocols. Results supporting the proposed "touch sense glove" device are presented in form of online BCI classification accuracy results. Finally, we outline the future possible paradigm improvements.

In this paper, we propose a new estimation method for the time difference of arrival (TDOA) between two microphones with improved accuracy by exploiting higher-order moments. In the proposed method analyzes the steered response power (SRP) of the observed signals after nonlinearly mapped onto a higher-dimensional space. Since the mapping operation enhances the linear independence between different vectors by increasing the dimensionality of the observed signals, the TDOA analysis achieves higher resolution. The results of an experiment comparing the TDOA estimation performance of the proposed method with that of the conventional methods reveal the robustness of the proposed method against noise and reverberation.

In this paper, we investigate the relationship between the way microphones are arranged and the degree to which speech is enhanced using the transfer-function-gain non-negative matrix factorization (NMF), which is an amplitude-based speech enhancement method that is suitable for use with an asynchronous distributed microphone array. In an asynchronous distributed microphone array, recording devices can be placed freely and the number of devices can be easily increased. Therefore, it is important that to determine the optimum microphone arrangement and the degree to which the performance is improved by using many microphones. We understood experimental evaluations to show that the performance by supervised NMF can achieve close to the ideal time-frequency masking with a sufficient number of microphones. We also show that the performance is better when more microphones are placed close to each source.

This paper aims to improve tactile and bone-conduction brain computer interface (tbaBCI) classification accuracy based on a new stimulus pattern search in order to trigger more separable P300 responses. We propose and investigate three approaches to stimulus spatial and frequency content modification. As result of the online tbaBCI classification accuracy tests with six subjects we conclude that frequency modification in the previously reported single vibrotactile exciter-based patterns leads to border of significance statistical improvements.

The paper presents a tactile pressure stimulus-based brain-computer interface (BCI) paradigm. 3 x 3 pressure pins matrix stimulus patterns are presented to the subjects in an oddball paradigm allowing for "aha-responses" generation to attended targets. A research hypothesis is confirmed with the results with five subjects performing online BCI experiments. One of the users could score with 100% accuracy in online ten averages based BCI test. Three users scored above chance levels, while one remained on the chance level border. The presented pilot study experiments and EEG results confirm the effectiveness of the proposed tactile pressure stimulus based BCI.

In this paper, we propose an easy and convenient method for traffic monitoring based on acoustic sensing with vehicle sound recorded by an ad-hoc microphone array. Since signals recorded by an ad-hoc microphone array are asynchronous, we perform channel synchronization by compensating for the difference between the start and the end of the recording and the sampling frequency mismatch. To monitor traffic, we estimate the number of the vehicles by employing the peak detection of the power envelopes, and classify the traffic lane from the difference between the propagation times of the microphones. We also demonstrate the effectiveness of our proposed method using the results of an experiment in which we estimated the number of vehicles and classified the lane in which the vehicles were traveling, according to F-measure.

Most speech codecs utilize a post-filter that emphasizes pitch structures to enhance perceptual quality at the decoder. Particularly, the bass post-filter used in ITU-T G.718 performs an adaptive pitch enhancement technique for a lower fixed frequency band. This paper describes a new post-filtering method in which the bass the frequency band and the gain are adaptively controlled frame-by-frame depending on the pitch frequency of decoded signal to improve bass post-filter performance. We have confirmed the improvement of the speech quality with the developed method through objective and subjective evaluations.

In this paper, we investigate the relationship between the way microphones are arranged and the degree to which speech is enhanced using the transfer-function-gain non-negative matrix factorization (NMF), which is an amplitude-based speech enhancement method that is suitable for use with an asynchronous distributed microphone array. In an asynchronous distributed microphone array, recording devices can be placed freely and the number of devices can be easily increased. Therefore, it is important that to determine the optimum microphone arrangement and the degree to which the performance is improved by using many microphones. We understood experimental evaluations to show that the performance by supervised NMF can achieve close to the ideal time-frequency masking with a sufficient number of microphones. We also show that the performance is better when more microphones are placed close to each source.

In this paper, we propose an automatic scoring method for the open answer task of the Japanese speaking test SJ-CAT. The proposed method first extracts a set of features from an input answer utterance and then estimates a vocabulary richness score by human raters, which ranges from 0 to 4, by employing SVR (support vector regression). We devised a novel set of features, namely text statistics weighted by word reliability, to assess the abundance of vocabulary and expression, and degree of word relevance based on the hierarchical distance in a thesaurus to evaluate the suitability of vocabulary. We confirmed experimentally that the proposed method provides good estimates of the human richness score, with a correlation coefficient of 0.92 and an RMSE (root mean square error) of 0.56. We also showed that the proposed method is relatively robust to differences among examinees and among questions used for training and testing.

The aim of this study is to provide a comprehensive test of head related
impulse response (HRIR) for an auditory spatial speller brain-computer
interface (BCI) paradigm. The study is conducted with six users in an
experimental set up based on five Japanese hiragana vowels. Auditory evoked
potentials resulted with encouragingly good and stable "aha-" or P300-responses
in real-world online BCI experiments. Our case study indicated that the
auditory HRIR spatial sound reproduction paradigm could be a viable alternative
to the established multi-loudspeaker surround sound BCI-speller applications,
as far as healthy pilot study users are concerned.

本稿では,Web 上で公開されている 「一つの曲を様々な歌手が歌った歌声」 を活用する二つのシステムを提案する.一つは,それらの歌声を重ね合わせる合唱生成支援システム,もう一つは,それらの歌声同士や白分の歌声を比較できる歌唱力向上支援システムである.従来,復数の楽曲を用いた鑑賞や創作支援,自分が歌うだけの歌唱力向上支援は研究されてきたが,同一曲を複数人が歌った歌声を活用した合唱生成や歌唱力向上支援はなかった.合唱生成支援システムでは,歌声の出現時刻と左右チャネルの音量をマウスで直感的に調整できる.直感的な操作と,それぞれの歌が完成された作品であることを利用することで,創作と同時に鑑賞を楽しむ 「創作鑑賞」 も可能となる.また,歌唱力向上支援システムでは,声質 (MFCC) と歌い回し (F0軌跡) が近い歌声同士を比較表示できる.Web 上で公開されていて再生数・マイリスト数があるため,それらの情報を活用しながら歌唱力向上に取り組める.これらのシス

This paper proposes a novel blind compensation of sampling frequency mismatch for asynchronous microphone array. Digital signals simultaneously observed by different recording devices have drift of the time differences between the observation channels because of the sampling frequency mismatch among the devices. Based on the model that such the time difference is constant within each time frame, but varies proportional to the time frame index, the effect of the sampling frequency mismatch can be compensated in the short-time Fourier transform domain by the linear phase shift. By assuming the sources are motionless and stationary, a likelihood of the sampling frequency mismatch is formulated. The maximum likelihood estimation is obtained effectively by a golden section search.

With independent component analysis (ICA), one promising source separation framework, it is difficult to separate desired signal components from the EEG observation, where vast number of sources are mixed. In this paper, we define the change of magnitude caused by each phenomenon inside brain as EEG event, and we formulate the probability model of the EEG event assuming the observation of each EEG event follows multivariate normal distribution locally in every short period. By regarding that each EEG event distirubtes sparsely in the time-frequency domain, the likelihood of the observation is given by Gaussian mixture model (GMM), and the parameters of the EEG events are estimated by an expectation-maximization (EM)algorithm. Also, by introducing Dirichlet prior probability with an appropriate hyperparameter to the activation of each Gaussian components, the EM algorithm achieves the ability to estimate both the number of significant EEG events and their parameters. An EEG separation experiment reveals that the proposed method can separate an appropriate number of EEG event.

Music induces a wide range of emotions. However, the influence of physiological<br> functions on musical emotions needs further theoretical considerations. This paper<br> summarizes the physical and physiological functions that are related to musical emo-<br>tions, and proposes a model for the embodied communication of musical emotions based<br> on a discussion on the transmission of musical emotions across people by sharing move-<br>ments and gestures. In this model, human with musical emotion is represented with<br> (1) the interfaces of perception and expression (senses, movements, facial and vocal<br> expressions), (2) an internal system of neural activities including the mirror system<br> and the hormonal secretion system that handles responses to musical activities, and<br> (3) the musical emotion that is enclosed in the internal system. Using this model, mu-<br>sic is the medium for transmitting emotions, and communication of musical emotions<br> is the communication of internal emotions through music and perception/expression<br> interfaces. Finally, we will discuss which aspect in music functions to encourage the<br> communication of musical emotions by humans.

Music induces a wide range of emotions. However, the influence of physiological functions on musical emotions needs further theoretical considerations. This paper summarizes the physical and physiological functions that are related to musical emo-tions, and proposes a model for the embodied communication of musical emotions based on a discussion on the transmission of musical emotions across people by sharing move-ments and gestures. In this model, human with musical emotion is represented with (1) the interfaces of perception and expression (senses, movements, facial and vocal expressions), (2) an internal system of neural activities including the mirror system and the hormonal secretion system that handles responses to musical activities, and (3) the musical emotion that is enclosed in the internal system. Using this model, mu-sic is the medium for transmitting emotions, and communication of musical emotions is the communication of internal emotions through music and perception/expression interfaces. Finally, we will discuss which aspect in music functions to encourage the communication of musical emotions by humans.

In this paper, we propose a method for synchronizing asynchronous channels in an ad-hoc microphone array based on single source activity for speech enhancement. An ad-hoc microphone array can include multiple recording devices, which do not communicate with each other. Therefore, their synchronization is a significant issue when using the conventional microphone array technique. We here assume that we know two or more segments (typically the beginning and the end of the recording) where only the sound source is active. Based on this situation, we compensate for the difference between the start and end of the recording and the sampling frequency mismatch. We also describe experimental results for speech enhancement with a maximum SNR beamformer.

To ensure a satisfactory QoE (Quality of Experience) and facilitate system design in speech recognition services, it is essential to establish a method that can be used to efficiently investigate recognition performance in different noise environments. Previously, we proposed a performance estimation method using the PESQ (Perceptual Evaluation of Speech Quality) as a spectral distortion measure. However, there is the problem that the relationship between the recognition performance and the distortion value differs depending on the noise reduction algorithm used. To solve this problem, we propose a novel performance estimation method that uses an estimator defined as a function of the distortion value and the SNR (Signal to Noise Ratio) of noise-reduced speech. The estimator is applicable to different noise reduction algorithms without any modification. We confirmed the effectiveness of the proposed method by experiments using the AURORA-2J connected digit recognition task and four different noise reduction algorithms. © 2013 IEEE.

The aim of the presented study is to provide a comprehensive test of the EEG evoked response potential (ERP) feature selection techniques for the spatial auditory BCI-speller paradigm, which creates a novel communication option for paralyzed subjects or body-able individuals requiring a direct brain-computer interfacing application. For rigor, the study is conducted with 16 BCI-naive healthy subjects in an experimental setup based on five Japanese hiragana characters in an offline processing mode. In our previous studies the spatial auditory stimuli related P300 responses resulted with encouragingly separable target vs. non-target latencies in averaged responses, yet that finding was not well reproduced in the online BCI single trial based settings. We present the case study indicating that the auditory spatial unimodal paradigm classification accuracy can be enhanced with an AUC based feature selection approach, as far as BCI-naive healthy subjects are concerned. © 2013 IEEE.

The paper presents a novel bone-conduction based brain-computer interface paradigm. Four sub-threshold acoustic frequency stimulus patterns are presented to the subjects in an oddball paradigm allowing for 'aha-responses' generation to the attended targets. This allows for successful implementation of the bone-conduction based brain-computer interface (BCI) paradigm. The concept is confirmed with seven subjects in online bone-conducted auditory Morse-code patterns spelling BCI paradigm. We report also brain electrophysiological signal processing and classification steps taken to achieve the successful BCI paradigm. We also present a finding of the response latency variability in a function of stimulus difficulty. © 2013 IEEE.

In this paper, we propose a new array signal processing technique for an underdetermined condition by increasing the number of observation channels. We introduce virtual observation as an estimate of the observed signals at positions where real microphones are not placed. Such signals at virtual observation channels are generated by the complex logarithmic interpolation of real observed signals. With the increased number of observation channels, conventional linear array signal processing methods can be applied to underdetermined conditions. As an example of the proposed array signal processing framework, we show experimental results of speech enhancement obtained with maximum SNR beamformers modified using the virtual observation.

The presented study explores the extent to which tactile stimuli delivered to five chest positions of a healthy user can serve as a platform for a brain computer interface (BCI) that could be used in an interactive application such as robotic vehicle operation. The five chest locations are used to evoke tactile brain potential responses, thus defining a tactile brain computer interface (tBCI). Experimental results with five subjects performing online tBCI provide a validation of the chest location tBCI paradigm, while the feasibility of the concept is illuminated through information-transfer rates. Additionally an offline classification improvement with a linear SVM classifier is presented through the case study. © Springer International Publishing 2013.

A brain-computer interface (BCI) is a technology for operating computerized devices based on brain activity and without muscle movement. BCI technology is expected to become a communication solution for amyotrophic lateral sclerosis (ALS) patients. Recently the BCI2000 package application has been commonly used by BCI researchers. The P300 speller included in the BCI2000 is an application allowing the calculation of a classifier necessary for the user to spell letters or sentences in a BCI-speller paradigm. The BCI-speller is based on visual cues, and requires muscle activities such as eye movements, impossible to execute by patients in a totally locked-in state (TLS), which is a terminal stage of the ALS illness. The purpose of our project is to solve this problem, and we aim to develop an auditory BCI as a solution. However, contemporary auditory BCI-spellers are much weaker compared with a visual modality. Therefore there is a necessity for improvement before practical application. In this paper, we focus on an approach related to the differences in responses evoked by various acoustic BCI-speller related stimulus types. In spite of various event related potential waveform shapes, typically a classifier in the BCI speller discriminates only between targets and non-targets, and hence it ignores valuable and possibly discriminative features. Therefore, we expect that the classification accuracy could be improved by using an independent classifier for each of the stimulus cue categories. In this paper, we propose two classifier training methods. The first one uses the data of the five stimulus cues independently. The second method incorporates weighting for each stimulus cue feature in relation to all of them. The results of the experiments reported show the effectiveness of the second method for classification improvement. © 2013 APSIPA.

Several extensions of the MUltiple SIgnal Classification (MUSIC) algorithm exploiting high order statistics were proposed to estimate directions of arrival (DOAs) with high resolution in underdetermined conditions. However, these methods entail a trade-off between two performance goals, namely, robustness and resolution, in the choice of orders because use of high-ordered statistics increases not only the resolution but also the statistical bias. To overcome this problem, this paper proposes a new extension of MUSIC using a nonlinear high-dimensional map, which corresponds to the joint analysis of moments of multiple orders and helps to realize the both advantages of robustness and high resolution of low-ordered and high-ordered statistics. Experimental results show that the proposed method can estimate DOAs more accurately than the conventional MUSIC extensions exploiting moments of a single high order.

This paper proposes a blind synchronization of ad-hoc microphone array in the short-time Fourier transform (STFT) domain with the optimized frame analysis centered at non-integer discrete time. We show that the drift caused by sampling frequency mismatch of asynchronous observation channels can be disregarded in a short interval. Utilizing this property, the sampling frequency mismatch and the recording start offset are estimated roughly by finding two pairs of the short intervals corresponding to the same continuous time. Using the estimate, STFT analysis is synchronized roughly between channels with optimized frame central. Since the optimized frame central is generally non-integer, we approximate the frame analysis by the linear phase filtering of the frame centered at the nearest integer sample. Maximum likelihood estimation refines the compensation of sampling frequency mismatch.

This paper presents recent results obtained with a new auditory spatial localization based BCI paradigm in which ERP shape differences at early latencies are employed to enhance classification accuracy in an oddball experimental setting. The concept relies on recent results in auditory neuroscience showing the possibility to differentiate early anterior contralateral responses to the spatial sources attended to. We also find that early brain responses indicate which direction, front or rear loudspeaker source, the subject attended to. Contemporary stimuli-driven BCI paradigms benefit most from the P300 ERP latencies in a so-called 'aha-response' setting. We show the further enhancement of the classification results in a spatial auditory paradigm, in which we incorporate N200 latencies. The results reveal that these early spatial auditory ERPs boost offline classification results of the BCI application. The offline BCI experiments with the multi-command BCI prototype support our research hypothesis with higher classification results and improved information transfer rates. © 2013 APSIPA.

This paper overviews a series of recent advances in adaptive processing and learning for audio source separation. In real world, speech and audio signal mixtures are observed in reverberant environments. Sources are usually more than mixtures. The mixing condition is occasionally changed due to the moving sources or when the sources are changed or abruptly present or absent. In this survey article, we investigate different issues in audio source separation including overdetermined/underdetermined problems, permutation alignment, convolutive mixtures, contrast functions, nonstationary conditions and system robustness. We provide a systematic and comprehensive view for these issues and address new approaches to overdetermined/underdetermined convolutive separation, sparse learning, nonnegative matrix factorization, information-theoretic learning, online learning and Bayesian approaches. © 2013 APSIPA.

This paper presents a novel concept of spatial auditory brain-computer interface utilizing real and virtual sound images. We report results obtained from psychophysical and EEG experiments with nine subjects utilizing a novel method of spatial real or virtual sound images as spatial auditory brain computer interface (BCI) cues. Real spatial sound sources result in better behavioral and BCI response classification accuracies, yet a direct comparison of partial results in a mixed experiment confirms the usability of the virtual sound images for the spatial auditory BCI. Additionally, we compare stepwise linear discriminant analysis (SWLDA) and support vector machine (SVM) classifiers in a single sequence BCI experiment. The interesting point of the mixed usage of real and virtual spatial sound images in a single experiment is that both stimuli types generate distinct event related potential (ERP) response patterns allowing for their separate classification. This discovery is the strongest point of the reported research and it brings the possibility to create new spatial auditory BCI paradigms. © 2013 APSIPA.

The study presented explores the extent to which tactile stimuli delivered to the ten digits of a BCI-naive subject can serve as a platform for a brain computer interface (BCI) that could be used in an interactive application such as robotic vehicle operation. The ten fingertips are used to evoke somatosensory brain responses, thus defining a tactile brain computer interface (tBCI). Experimental results on subjects performing online (real-time) tBCI, using stimuli with a moderately fast inter-stimulus-interval (ISI), provide a validation of the tBCI prototype, while the feasibility of the concept is illuminated through information-transfer rates obtained through the case study. © 2013 Springer-Verlag.

The presented study explores the extent to which tactile-force stimulus delivered to a hand holding a force-feedback joystick can serve as a platform for a brain-computer interface (BCI). The four pressure directions are used to evoke tactile brain potential responses, thus defining a tactile-force brain computer interface (tfBCI). We present brain signal processing and classification procedures leading to successful online interfacing results. Experimental results with seven subjects performing online BCI experiments provide a validation of the hand location tfBCI paradigm, while the feasibility of the concept is illuminated through remarkable information-transfer rates. © 2013 IEEE.

Technical ear training aims to improve the listening of sound engineers so they can skillfully modify and edit the structure of sound. Despite recent increasing interest in listening ability and subjective evaluation in the field of audio- and acoustic-related fields and the subsequent appearance of various technical ear-training methods, the subject of how to provide efficient training for a self-trainee has not yet been studied. This paper investigated trainees' performances and showed that an (inherent or learned) ability to correctly describe spectral differences using the terms of a parametric equalizer (center frequency, Q, and gain) was different for each person. To cope with such individual differences in spectral identification, the authors proposed a novel method that adaptively controls the training task based on a trainee's prior performances. In detail, the method estimates the weakness of the trainee, and generates a training routine that focuses on that weakness. Subsequently, we tried to determine whether the proposed method-adaptive feedback-helps self-learners improve their performance in technical listening that involves identifying spectral differences. The results showed that the proposed method could assist trainees in improving their ability to identify differences more effectively than the counterpart group. Together with other features required for effective selftraining, this adaptive feedback would assist a trainee in acquisition of timbre-identification ability. © 2013 The Acoustical Society of Japan.

This paper presents a quantitative metric to describe the multidimensionality of spectral envelope perception, that is, the perception specifically related to the spectral element of timbre. Mel-cepstrum (Mel-frequency cepstral coefficients or MFCCs) is chosen as a hypothetical metric for spectral envelope perception due to its desirable properties of linearity, orthogonality, and multidimensionality. The experimental results confirmed the relevance of Mel-cepstrum to the perceived timbre dissimilarity when the spectral envelopes of complex-tone synthetic sounds were systematically controlled. The first experiment measured the perceived dissimilarity when the stimuli were synthesized by varying only a single coefficient from MFCC. Linear regression analysis proved that each of the 12 MFCCs has a linear correlation with spectral envelope perception. The second experiment measured the perceived dissimilarity when the stimuli were synthesized by varying two of the MFCCs. Multiple regression analysis showed that the perceived dissimilarity can be explained in terms of the Euclidean distance of the MFCC values of the synthetic sounds. The quantitative and perceptual relevance between the MFCCs and spectral centroids is also discussed. These results suggest that MFCCs can be a metric representation of spectral envelope perception, where each of its orthogonal basis functions provides a linear match with human perception.

本論文では,音源信号のスパース性に基づき,時間周波数バイナリマスク(BM)を用いる音源分離手法におけるミュージカルノイズの低減を目的とした,分離音声のケプストラムスムージング(CSS)を提案する。CSSは,近年提案されたスペクトルマスクのケプストラムスムージング(CSM)で用いられるケプストラム領域でスムージングする考え方と,ケプストラム表現による音声特性の保持の制御という観点では,マスクではなくBMによって得られた分離音声を直接スムージングする方が好ましいという仮説とに基づいている。また,従来法(CSM)や提案法(CSS)と他のミュージカルノイズ低減手法の性能を実験により比較する。CSSでは,CSMと同程度のミュージカルノイズ低減性能を有し,更に目的音声の歪の小さい分離信号が得られた。

In this paper, we propose a new technique for sparseness-based underdetermined BSS that is based on the clustering of the frequency-dependent time difference of arrival (TDOA) information and that can cope with diffused noise environments. Such a method with an EM algorithm has already been proposed, however, it required a time-consuming exhaust search for TDOA inference. To remove the need for such an exhaust search, we propose a new technique by focusing on a stereo case. We derive an update rule for analytical TDOA estimation. This update rule eliminates the need for the exhaustive TDOA search, and therefore reduces the computational load. We show experimental results for separation performance and calculation time in comparison with those obtained with the conventional approach. Our reported results validate our proposed method, that is, our proposed method achieves high performance without a high computational cost.

The paper presents our recent results obtained with a new auditory spatial localization based BCI paradigm in which the ERP shape differences at early latencies are employed to enhance the traditional P300 responses in an oddball experimental setting. The concept relies on the recent results in auditory neuroscience showing a possibility to differentiate early anterior contralateral responses to attended spatial sources. Contemporary stimuli-driven BCI paradigms benefit mostly from the P300 ERP latencies in so called "aha-response" settings. We show the further enhancement of the classification results in spatial auditory paradigms by incorporating the N200 latencies, which differentiate the brain responses to lateral, in relation to the subject head, sound locations in the auditory space. The results reveal that those early spatial auditory ERPs boost online classification results of the BCI application. The online BCI experiments with the multi-command BCI prototype support our research hypothesis with the higher classification results and the improved information-transfer-rates. © 2012 APSIPA.

Biofeedback is currently considered as an effective method for medical rehabilitation. It aims to increase the awareness and recognition of the body's motion by feeding back the physiological information to the patients in real time. Our goal is to create an auditory biofeedback that aids understanding of the dynamic motion involving multiple muscular parts, with the ultimate aim of clinical rehabilitation use. In this paper, we report the development of a real-time sonification system using EMG, and we propose three sonification methods that represent the data in pitch, timbre, and the combination of polyphonic timbre and loudness. Our user evaluation test involves the task of timing and order identification and a questionnaire about the subjective comprehensibility and the preferences, leading to a discussion of the task performance and usability. The results show that the subjects can understand the order of the muscular activities at 63.7% accuracy on average. And the sonification method with polyphonic timbre and loudness provides an 85.2% accuracy score on average, showing its effectiveness. Regarding the preference of the sound design, we found that there is not a direct relationship between the task performance accuracy and the preference of sound in the proposed implementations.

The paper presents results from a psychophysical study conducted to optimize vibrotactile stimuli delivered to subject finger tips in order to evoke the somatosensory responses to be utilized next in a haptic brain computer interface (hBCI) paradigm. We also present the preliminary EEG evoked responses for the chosen stimulating frequency. The obtained results confirm our hypothesis that the hBCI paradigm concept is valid and it will allow for rapid stimuli presentation in order to improve information-transfer-rate (ITR) of the BCI. © 2012 IEEE.

We are now developing a Japanese speaking test called SCAT, which is part of J-CAT (Japanese Computerized Adaptive Test), a free online proficiency test for Japanese language learners. In this paper, we focus on the sentence-reading-aloud task and the sentence generation task in SCAT, and propose an automatic scoring method for estimating the overall score of answer speech, which is holistically determined by language teachers according to a rating standard. In that process, teachers carefully consider different factors but do not rate the scores of them. We therefore analyze how each factor contributes to the overall score. The factors are divided into two categories: the quality of speech and the content of speech. The former includes pronunciation and intonation, and the latter representation and vocabulary. We then propose an automatic scoring method based on the analysis. Experimental results confirm that the proposed method gives relatively accurate estimates of the overall score.

We propose a method for an improvement of auditory BCI (aBCI) paradigm based on a combination of ASSR stimuli optimization by choosing the subjects' best responses to AM-, flutter-, AM/FM and click-envelope modulated sounds. As the ASSR response features we propose pairwise phase-locking-values calculated from the EEG and next classified using binary classifier to detect attended and ignored stimuli. We also report on a possibility to use the stimuli as short as half a second, which is a step forward in ASSR based aBCI. The presented results are helpful for optimization of the aBCI stimuli for each subject. © 2012 APSIPA.

This paper describes an application of a multichannel EEG sonification approach. We present results obtained with a multichannel-sonification method tested with steady-state EEG responses. We elucidate brain synchrony patterns in an auditory domain with utilization of the EEG coherence measure. The transitions in the synchrony patterns are represented as timbre (i.e., spectro-temporal) deviation and as spatial movement of the sound cluster. Our final sonification evaluation experiment with six subjects confirms the validity of the proposed brain synchrony-elucidation approach. © 2012 APSIPA.

This paper describes control of distance attenuation using spherical loudspeaker array. Fisher et al. proposed radial filtering with spherical microphone to control the sensitivity to distance from a sound source by modeling the propagation of waves in spherical harmonic domain. Since transfer functions are not changed by swapping their inputs and outputs, we can use the same theory of radial filtering for microphone arrays to the filter design of distance attenuation control with loudspeaker arrays. Experimental results confirmed that the proposed method is effective in low frequencies.

The paper presents results from a project aiming to create horizontally distributed surround sound sources and virtual sound images as auditory BCI (aBCI) stimuli. The purpose is to create evoked brain wave response patterns depending on attended or ignored sound directions. We propose to use a modified version of the vector based amplitude panning (VBAP) approach to achieve the goal. The so created spatial sound stimulus system for the novel oddball aBCI paradigm allows us to create a multi-command experimental environment with very encouraging results reported in this paper.We also present results showing that a modulation of the sound image depth changes also the subject responses. Finally, we also compare the proposed virtual sound approach with the traditional one based on real sound sources generated from the real loudspeaker directions. The so obtained results confirm the hypothesis of the possibility to modulate independently the brain responses to spatial types and depths of sound sources which allows for the development of the novel multi-command aBCI. © 2012 APSIPA.

The paper presents a pilot study conducted with spatial visual, audiovisual and auditory brain-computer-interface (BCI) based speller paradigms. The psychophysical experiments are conducted with healthy subjects in order to evaluate a difficulty and a possible response accuracy variability. We also present preliminary EEG results in offline BCI mode. The obtained results validate a thesis, that spatial auditory only paradigm performs as good as the traditional visual and audiovisual speller BCI tasks. © 2012 IEEE.

Multichannel Wiener filter proposed by Duong et al. can conduct underdetermined blind source separation (BSS) with low distortion. This method assumes that the observed signal is the superimposition of the multichannel source images generated from multivariate normal distributions. The covariance matrix in each time-frequency slot is estimated by an EM algorithm which treats the source images as the hidden variables. Using the estimated parameters, the source images are separated as the maximum a posteriori estimate. It is worth nothing that this method does not assume the sparseness of sources, which is usually assumed in underdetermined BSS. In this paper we investigate the effectiveness of the three attributes of Duong's method, i.e., the source image model with multivariate normal distribution, the observation model without sparseness assumption, and the source separation by multichannel Wiener filter. We newly formulate three BSS methods with the similar source image model and the different observation model assuming sparseness, and we compare them with Duong's method and the conventional binary masking. Experimental results confirmed the effectiveness of all the three attributes of Duong's method.

We have proposed a new algorithm for sparseness-based underdetermined blind source separation (ESS) that can cope with diffused noise environments. This algorithm includes a technique for estimating the time-difference-of-arrival (TDOA) parameter separately in individual frequency bins for each source. In this paper, we propose methods that integrate the frequency-bin-wise TDOA parameter to estimate the TDOA of each source. The accuracy of TDOA estimation with the proposed approach is shown experimentally in comparison with a conventional approach. The separation performance and calculation time of the proposed approach is also examined.

In this research, we propose a method to update estimation of time difference of arrival (TDOA) analytically in sparseness-based underdetermined blind source separation (BSS) with an EM algorithm. Izumi et at. proposed underdetermined BSS that can cope with diffuse noise environments. However, Izumi's method requires discrete exhaustive search to update TDOA parameter every iteration, thereby takes high computational cost. In this paper, focusing on the stereo case, we obtain analytical update of TDOA parameters in each frequency bin using frequency-dependent TDOA modeling. This update rule eliminates the exhaustive TDOA search, and therefore reduces the computational load. We show experimental results of separation performance and calculation time in comparison with those obtained with the conventional approach.

This paper proposes a method for estimating the direction of arrival (DOA) of multiple source signals for an underdetermined situation, where the number of sources N exceeds the number of sensors M (M &lt; N). Some DOA estimation methods have already been proposed for underdetermined cases. However, since most of them restrict their microphone array arrangements, their DOA estimation ability is limited to a 2-dimensional plane. To deal with an underdetermined case where sources are distributed arbitrarily, we propose a method that can employ a 2- or 3-dimensional sensor array. Our new method employs the source sparseness assumption to handle an underdetermined case. Our formulation with the sensor coordinate vectors allows us to employ arbitrarily arranged sensors easily. We obtained promising experimental results for 2-dimensionally distributed sensors and sources 3x4, 3x5 (#sensors x #speech sources), and for 3-dimensional case with 4x5 in a room (reverberation time (RT) of 120 ms). We also investigate the DOA estimation performance under several reverberant conditions.

インタレース映像を符号化したMPEG-2ストリームをH.264へトランスコードする際に,初段符号化情報を利用して,混入する量子化雑音を低減する手法を提案する.本手法では,MPEG-2のDCT種別と動き補償種別をH.264へ極力継承し,更にフレームベクトルからフィールドベクトルに変換すれば継承可能となるペアMBをDCT種別と動き補償種別の組合せから判別し,ベクトル変換することで継承率を向上させる.実験の結果,符号化情報を利用しない従来手法に比べ,0.19〜0.31dBのPSNR向上が確認できた.

Blind source separation (BSS) is a technique for estimating and separating individual source signals from a mixed signal using only information observed by each sensor. BSS is still being developed for mixed signals that are affected by reverberation. In this paper, we propose combining the BSS method that considers reverberation proposed by Duong et al. with the BSS method reported by Sawada et al., which does not consider reverberation, for the initial setting of the EM algorithm. This proposed method assumes the underdetermined case. In the experiment, we compare the proposed method with the conventional method reported by Duong et al. and that reported by Sawada et al., and demonstrate the effectiveness of the proposed method. © 2011 IEEE.

The paper introduces a novel auditory BCI/BMI paradigm based on combined sound timbre and horizontal plane spatial locations as informative cues. The presented concept is based on responses to eight-directional audio stimuli with various tonal and environmental sound stimuli. The approach is based on a monitoring of brain electrical activity by means of the electroencephalogram (EEG). The previously developed by the authors spatial auditory stimulus is extended to varying in timbre sound stimuli which feature helps the subjects to attend to the targets. The main achievement discussed in the paper is an offline BCI analysis based on an optimization of electrode locations on the scalp and evoked response latency for further classification results improvement. The so developed new BCI paradigm is more user-friendly and it leads to better results comparing to previously utilized simple tonal or steady-state stimuli.

This paper proposes signal restoration from clipping effect without prior knowledge. First, an interval of signal including clipped samples is analyzed by recursive vector projection. By analyzing the neighboring samples of the clipped interval and excluding the clipped interval in the analysis of similarity, signal estimation in the clipped interval is estimated as a by-product of the analysis. Since the estimation holds consistency with the neighboring samples, the restored signal does not suffer from click noise. Evaluation of the clipping restoration with various audio signal ascertained that the proposed method improves signal-to-noise ratio.

This paper presents a new method for underdetermined Blind Source Separation (BSS), based on a concept called multichannel complex non-negative matrix factorization (NMF). The method assumes (1) that the time-frequency representations of sources have disjoint support (W-disjoint orthogonality of sources), and (2) that each source is modeled as a superposition of components whose amplitudes vary over time coherently across all frequencies (amplitude coherence of frequency components) in order to jointly solve the indeterminacy involved in the frequency domain underdetermined BSS problem. We confirmed experimentally that the present method performed reasonably well in terms of the signal-to-interference ratio when the mixing process was known.

To provide users with natural and intelligible speech in noisy environments, the use of a noise reduction algorithm, which reduces the noise component in the noisy input speech, can be effective. It is, however, well-known that any noise reduction algorithm unavoidably produces speech distortion and residual noise. Here, the critical issue is that the characteristics of these undesired byproducts vary according to the noise reduction algorithm used and the type of noise to be reduced. It is therefore essential to establish methods that can be used to evaluate the quality of noise-reduced speech. In this paper, we describe subjective and objective quality evaluation methods for noise-reduced speech.

It is important for efficiently reviewing meeting speech archives to preliminarily and automatically detect "when, how and who talked". In this paper, we propose a method for automatically detecting a short and scattered signal such as agreements by using only acoustical information. The proposed method has two steps: 1) extract a spatial power spectrum frame-by-frame from the meeting speech archive recorded by a microphone array, and 2) detect the target signal by using an outlier detection algorithm based on principal component analysis. To evaluate the effectiveness of the proposed method, we conducted an experiment using the meeting speech archive recorded in a real room. The experimental results imply that we can detect a long utterance, a short utterance, no utterance from only a few principal components.

We propose an MPEG-2 to H.264 transcoding method for interlace streams intermingled with frame and field macroblocks. This method uses the encoding information from an MPEG-2 stream and keeps as many DCT coefficients of the original MPEG-2 bitstream as possible. Experimental results show that the proposed method improves PSNR by about 0.19-0.31 dB compared with a conventional method. © 2010 IEEE.

To ensure a satisfactory QoE (Quality of Experience) and facilitate system design in speech recognition services, it is essential to establish a method that can be used to efficiently investigate recognition performance in different noise environments. Previously, we proposed a performance estimation method using a spectral distortion measure. However, there is the problem that recognition task complexity affects the relationship between the recognition performance and the distortion value. To solve this problem, this paper proposes a novel performance estimation method considering the recognition task complexity. We confirmed that the proposed method gives accurate estimates of the recognition performance for various recognition tasks by an experiment using noisy speech data recorded in a real room.

Communication quality in IP telephony is rated in terms of the Mean Opinion Score (MOS), which is an Absolute Category Rating (ACR) scale. There is a problem when comparing subjectively evaluated MOSs in that the evaluation results are strongly affected by differences in language, the instruction words used for the evaluation, and the nationality of the evaluator. To solve these problems, ITU-T SG12 has started to investigate the cultural and language dependencies of subjective quality evaluations undertaken with the MOS method for speech/video/multimedia. In this paper, we present the results of a comparison of the MOS evaluation characteristics for Chinese, Japanese, and English. ©2010 IEEE.

In this paper, we propose a novel sparse source separation method that can estimate the number of sources and time-frequency masks simultaneously, even when the spatial aliasing problem exists. Recently, many sparse Source separation approaches with time-frequency masks have been proposed. However, most of these approaches require information on the number of sources in advance. In our proposed method, we model the phase difference of arrival (PDOA) between microphones with a Gaussian mixture model (GMM) with a Dirichlet prior. Then we estimate the model parameters by using the maximum a posteriori (MAP) estimation based on the EM algorithm. In order to avoid one cluster being modeled by two or more Gaussians, we utilize a sparse distribution modeled by the Dirichlet distributions as the prior of the GMM mixture weight. Moreover, to handle wide microphone spacing cases where the spatial aliasing problem occurs, the indeterminacy of modulus 2 pi k in the phase is also included in our model. Experimental results show good performance of our proposed method.

In this paper, we propose a novel sparse source separation method that can be applied even if the number of sources is unknown. Recently, many sparse source separation approaches with time-frequency masks have been proposed. However, most of these approaches require information on the number of sources in advance. In our proposed method, we model the histogram of the estimated direction of arrival (DOA) with a Gaussian mixture model (GMM) with a Dirichlet prior. Then we estimate the model parameters by using the maximum a posteriori estimation based on the EM algorithm. In order to avoid one cluster being modeled by two or more Gaussians, we utilize a sparse distribution modeled by the Dirichlet distributions as the prior of the GMM mixture weight. By using this prior, without any specific model selection process, our proposed method can estimate the number of sources and time-frequency masks simultaneously. Experimental results show the performance of our proposed method.

This paper presents a speaker diarization system that estimates who spoke when in a meeting. Our proposed system is realized by using a noise robust voice activity detector (VAD), a direction of arrival (DOA) estimator, and a DOA classifier. Our previous system utilized the generalized cross correlation method with the phase transform (GCC-PHAT) approach for the DOA estimation. Because the GCC-PHAT can estimate just one DOA per frame, it was difficult to handle speaker overlaps. This paper tries to deal with this issue by employing a DOA at each time-frequency slot (TFDOA), and reports how it improves diarization performance for real meetings / conversations recorded in a room with a reverberation time of 350 ms.

This paper presents a speaker indexing method that uses a small number of microphones to estimate who spoke when. Our proposed speaker indexing is realized by using a noise robust voice activity detector (VAD), a GCC-PHAT based direction of arrival (DOA) estimator, and a DOA classifier. Using the estimated speaker indexing information, we can also enhance the utterances of each speaker with a maximum signal-to-noise-ratio (MaxSNR) beamformer. This paper applies our system to real recorded meetings/conversations recorded in a room with a reverberation time of 350 ms, and evaluates the performance by a standard measure: the diarization error rate (DER). Even for the real conversations, which have many speaker turn-takings and overlaps, the speaker error time was very small with our proposed system. We are planning to demonstrate a real-time speaker indexing system at ICASSP2008.

Especially for tasks like automatic meeting transcription, it would be useful to automatically recognize speech also while multiple speakers are talking simultaneously. For this purpose, speech separation can be performed, for example by using maximum SNR beamforming. However, even when good interferer suppression is attained, the interfering speech will still be recognizable during those intervals, where the target speaker is silent. In order to avoid the consequential insertion errors, a new soft masking scheme is proposed, which works in the time domain by inducing a large damping on those temporal periods, where the observed direction of arrival does not correspond to that of the target speaker. Even though the masking scheme is aggressive, by means of missing feature recognition the recognition accuracy can be improved significantly, with relative error reductions in the order of 60% compared to maximum SNR beamforming alone, and it is successful also for three simultaneously active speakers. Results are reported based on the SOLON speech recognizer, NTT's large vocabulary system [1], which is applied here for the recognition of artificially mixed data using real-room impulse responses and the entire clean test set of the Aurora 2 database.

This paper presents a new method for blind sparse source separation. Some sparse source separation methods, which. rely on source sparseness and an anechoic mixing model, have already been proposed. These methods utilize level ratios and phase differences between sensor observations as their features, and they separate signals by classifying them. However, some of the features cannot form clusters with a well-known clustering algorithm, e.g., the k-means. Moreover, most previous methods utilize a linear sensor array (or only two sensors), and therefore they cannot separate symmetrically positioned sources. To overcome such problems, we propose a new feature that can be clustered by the k-means algorithm and that can be easily applied to more than three sensors arranged non-linearly. We have obtained promising results for two- and three-dimensionally distributed speech separation with non-linear/non-uniform sensor arrays in a real room even in underdetermined situations. We also investigate the way in which the performance of such methods is affected by room reverberation, which may cause the sparseness and anechoic assumptions to collapse. (C) 2007 Elsevier B.V. All rights reserved.

We address the problem of underdetermined BSS. While most previous approaches are designed for instantaneous mixtures,we propose a time-frequency-domain algorithm for convolutive mixtures. We adopt a two-step method based on a general maximum a posteriori (MAP) approach. In the first step, we estimate the mixing matrix based on hierarchical clustering, assuming that the source signals are sufficiently sparse. The algorithm works directly on the complex-valued data in the time-frequency domain and shows better convergence than algorithms based on self-organizing maps. The assumption of Laplacian priors for the source signals in the second step leads to an algorithm for estimating the source signals. It involves the ℓ1-norm minimization of complex numbers because of the use of the time-frequency-domain approach. We compare a combinatorial approach initially designed for real numbers with a second-order cone programming (SOCP) approach designed for complex numbers. We found that although the former approach is not theoretically justified for complex numbers, its results are comparable to, or even better than, the SOCP solution. The advantage is a lower computational cost for problems with low input/output dimensions.

This paper deals with a new technique for blind source separation (BSS) from convolutive mixtures. We present a three-stage separation system employing time-frequency binary masking, beamforming and a non-linear post processing technique. The experiments show that this system outperforms conventional time-frequency binary masking (TFBM) in both (over-)determined and underdetermined cases. Moreover it removes the musical noise and reduces interference in time-frequency slots extracted by TFBM.

This paper describes the frequency-domain approach to the blind source separation of speech/audio signals that are convolutively mixed in a real room environment. With the application of shorttime Fourier transforms, convolutive mixtures in the time domain can be approximated as multiple instantaneous mixtures in the frequency domain. We employ complex-valued independent component analysis (ICA) to separate the mixtures in each frequency bin. Then, the permutation ambiguity of the ICA solutions should be aligned so that the separated signals are constructed properly in the time domain. We propose a permutation alignment method based on clustering the activity sequences of the frequency bin-wise separated signals. We achieved the overall winner status of MLSP 2007 Data Analysis Competition based on the presented method. ©2007 IEEE.

This paper proposes a two-stage method for the blind separation of convolutively mixed sources. We employ time-frequency masking, which can be applied even to an underdetermined case where the number of sensors is insufficient for the number of sources. In the first stage of the method, frequency bin-wise mixtures are classified based on Gaussian mixture model fitting. In the second stage, the permutation ambiguities of the bin-wise classified signals are aligned by clustering the posterior probability sequences calculated in the first stage. Experimental results for separating four speeches with three microphones under reverberant conditions show the superiority of the proposed method over existing methods based on time-difference-of-arrival estimations or signal envelope clustering.

This paper presents a new method for grouping bin-wise separated signals for individual sources, i.e., solving the permutation problem, in the process of frequency-domain blind source separation. Conventionally, the correlation coefficient of separated signal envelopes is calculated to judge whether or not the separated signals originate from the same source. In this paper, we propose a new measure that represents the dominance of the separated signal in the mixtures, and use it for calculating the correlation coefficient, instead of a signal envelope. Such dominance measures exhibit dependence/independence more clearly than traditionally used signal envelopes. Consequently, a simple clustering algorithm with centroids works well for grouping separated signals. Experimental results were very appealing, as three sources including two coming from the same direction were separated properly with the new method.

We propose a speech separation method for a meeting situation, where each speaker sometimes speaks and the number of speakers changes every moment. Many source separation methods have already been proposed, however, they consider a case where all the speakers keep speaking: this is not always true in a real meeting. In such cases, in addition to separation, speech detection and the classification of the detected speech according to speaker become important issues. For that purpose, we propose a method that employs a maximum signal-to-noise (MaxSNR) beamformer combined with a voice activity detector and online clustering. We also discuss the scaling ambiguity problem as regards the MaxSNR beamformer, and provide their solutions. We report some encouraging results for a real meeting in a room with a reverberation time of about 350 ins.

This article provides an overview of the first stereo audio source separation evaluation campaign, organized by the authors. Fifteen underdetermined stereo source separation algorithms have been applied to various audio data, including instantaneous, convolutive and real mixtures of speech or music sources. The data and the algorithms are presented and the estimated source signals are compared to reference signals using several objective performance criteria.

We address the problem of underdetermined BSS. While most previous approaches are designed for instantaneous mixtures, we propose a time-frequency-domain algorithm for convolutive mixtures. We adopt a two-step method based on a general maximum a posteriori ( MAP) approach. In the first step, we estimate the mixing matrix based on hierarchical clustering, assuming that the source signals are sufficiently sparse. The algorithm works directly on the complex-valued data in the time-frequency domain and shows better convergence than algorithms based on self-organizing maps. The assumption of Laplacian priors for the source signals in the second step leads to an algorithm for estimating the source signals. It involves the l(1)-norm minimization of complex numbers because of the use of the time-frequency-domain approach. We compare a combinatorial approach initially designed for real numbers with a second-order cone programming (SOCP) approach designed for complex numbers. We found that although the former approach is not theoretically justified for complex numbers, its results are comparable to, or even better than, the SOCP solution. The advantage is a lower computational cost for problems with low input/output dimensions. Copyright (C) 2007 Stefan Winter et al.

Stereo echo cancellation requires a fast converging adaptive algorithm because the stereo input signals are highly cross correlated and the convergence rate of the misalignment is slow even after preprocessing for unique identification of stereo echo paths. To speed up the convergence, we propose enhancing the contribution of the decorrelated components in the preprocessed input-signal vector to adaptive updates. The adaptive filter coefficients are updated on the basis of either a single or multiple past enhanced input-signal vectors.
For a single-vector update, we show how this enhancement improves the convergence rate by analyzing the behavior of the filter coefficient error in the mean. For a two-past-vector update, simulation showed that the proposed enhancement leads to a faster decrease in misalignment than the corresponding conventional second-order affine projection algorithm while computational complexities are almost the same. (c) 2005 Elsevier B.V. All rights reserved.

This paper describes a method for solving the permutation problem of frequency-domain blind source separation (BSS). The method analyzes the mixing system information estimated with independent component analysis (ICA). When we use widely spaced sensors or increase the sampling rate, spatial aliasing may occur for high frequencies due to the possibility of multiple cycles in the sensor spacing. In such cases, the estimated information would imply multiple possibilities for a source location. This causes some difficulty when analyzing the information. We propose a new method designed to overcome this difficulty. This method first estimates the model parameters for the mixing system at low frequencies where spatial aliasing does not occur, and then refines the estimations by using data at all frequencies. This refinement leads to precise parameter estimation and therefore precise permutation alignment. Experimental results show the effectiveness of the new method.

This paper presents a new method for estimating the direction of arrival (DOA) of source signals whose number N can exceed the number of sensors M. Subspace based methods, e.g., the MUSIC algorithm, have been widely studied, however, they are only applicable when M &gt; N. Another conventional independent component analysis based method allows AY &gt; N, however, it cannot be applied when A,1 &lt; N. By contrast, our new method can be applied where the sources outnumber the sensors (i.e., an underdeterinined case M &lt; N) by assuming source sparseness. Our method can cope with 2- or 3-dimensionally distributed sources with a 2- or 3-dimensional sensor array. We obtained promising experimental results for 3 x 4, 3 x 5 and 4 x 5 (#sensors x #speech sources) in a room (RT60 = 120 ms).

This paper describes the frequency-domain blind source separation (BSS) of convolutively mixed acoustic signals using independent component analysis (ICA). The most critical issue related to frequency domain BSS is the permutation problem. This paper presents two methods for solving this problem. Both methods are based on the clustering of information derived from a separation matrix obtained by ICA. The first method is based on direction of arrival (DOA) clustering. This approach is intuitive and easy to understand. The second method is based on normalized basis vector clustering. This method is less intuitive than the DOA based method, but it has several advantages. First, it does not need sensor ar-ray geometry information. Secondly, it can fully utilize the information contained in the separation matrix, since the clustering is performed in high-dimensional space. Experimental results show that our methods realize BSS in various situations such as the separation of many speech signals located in a 3-dimensional space, and the extraction of primary sound sources surrounded by many background interferences.

The problem of blind source separation (BSS) from convolutive mixtures is often addressed using independent component analysis in the frequency domain. The separation performance with this approach degrades significantly when only a short amount of data is available, since the estimation of the separation system becomes inaccurate. In this paper we present a novel approach to the frequency domain BSS using frequency normalization. Under the conditions of almost sparse sources and of dominant direct path in the mixing systems, we show that the new approach provides better performance than the conventional one when the amount of available data is small.

This paper introduces the blind source separation (BSS) of convolutive mixtures of acoustic signals, especially speech. A statistical and computational technique, called independent component analysis (ICA), is examined. By achieving, nonlinear decorrelation, nonstationary decorrelation, or time-delayed decorrelation, we can find source signals only from observed mixed signals. Particular attention is paid to the physical interpretation of BSS from the acoustical signal processing point of view. Frequency-domain BSS is shown to be equivalent to two sets of frequency, domain adaptive microphone arrays, i.e., adaptive beamformers (ABFs). Although BSS can reduce reverberant sounds to some extent in the same way as ABF, it mainly removes the sounds from the jammer direction. This is why BSS has difficulties with long reverberation in the real world. If sources are not "independent," the dependence results in bias noise when obtaining the correct separation filter coefficients. Therefore, the performance of BSS is limited by that of ABF. Although BSS is upper bounded by ABF, BSS has a strong advantage over ABF. BSS can be regarded as an intelligent version of ABF in the sense that it can adapt without any information on the array manifold or the target direction., and sources can be simultaneously active in BSS.

We discuss approaches for blind source separation where we can use more sensors than sources to obtain a better performance. The discussion focuses mainly on reducing the dimensions of mixed signals before applying independent component analysis. We compare two previously proposed methods. The first is based on principal component analysis, where noise reduction is achieved. The second is based on geometric considerations and selects a subset of sensors in accordance with the fact that a low frequency prefers a wide spacing, and a high frequency prefers a narrow spacing. We found that the PCA-based method behaves similarly to the geometry-based method for low frequencies in the way that it emphasizes the outer sensors and yields superior results for high frequencies. These results provide a better understanding of the former method.

This paper describes a method for solving the permutation problem of frequency-domain blind source separation (BSS). The method analyzes the mixing system information estimated with independent component analysis (ICA). When we use widely spaced sensors or increase the sampling rate, spatial aliasing may occur for high frequencies due to the possibility of multiple cycles in the sensor spacing. In such cases, the estimated information would imply multiple possibilities for a source location. This causes some difficulty when analyzing the information. We propose a new method designed to overcome this difficulty. This method first estimates the model parameters for the mixing system at low frequencies where spatial aliasing does not occur, and then refines the estimations by using data at all frequencies. This refinement leads to precise parameter estimation and therefore precise permutation alignment. Experimental results show the effectiveness of the new method.

The problem of blind source separation (BSS) from convolutive mixtures is often addressed using independent component analysis in the frequency domain. The separation performance with this approach degrades significantly when only a short amount of data is available, since the estimation of the separation system becomes inaccurate. In this paper we present a novel approach to the frequency domain BSS using frequency normalization. Under the conditions of almost sparse sources and of dominant direct path in the mixing systems. we show that the new approach provides better performance than the conventional one when the amount of available data is small.

We propose a new method for solving the underdetermined sparse signal separation problem. Some sparseness based methods have already been proposed. However, most of these methods utilized a linear sensor array (or only two sensors), and therefore they have certain limitations; e.g., they cannot separate symmetrically positioned sources. To allow the use of more than three sensors that can be arranged in a non-linear/non-uniform way, we propose a new method that includes the normalization and clustering of the observation vectors. Our proposed method can handle both underdetermined case and (over-)determined cases. We show practical results for speech separation with nonlinear/non-uniform sensor arrangements. We obtained promising experimental results for the cases of 3 x 4, 4 x 5 (#sensors x #sources) in a room (RT60 = 120 ms).

This paper presents a new method for estimating the direction of arrival (DOA) of source signals whose number N can exceed the number of sensors M. Subspace based methods, e.g., the MUSIC algorithm, have been widely studied, however, they are only applicable when M &gt; N. Another conventional independent component analysis based method allows M &gt; N, however, it cannot be applied when M &lt; N. By contrast, our new method can be applied where the sources outnumber the sensors (i.e., an underdetermined case M &lt; N) by assuming source sparseness. Our method can cope with 2- or 3-dimensionally distributed sources with a 2- or 3-dimensional sensor array. We obtained promising experimental results for 3 x 4, 3 x 5 and 4 x 5 (#sensors x #speech sources) in a room (RT60 = 120 ms).

For blind source separation (BSS) of convolutive mixtures, the frequency-domain approach is efficient and practical, because the convolutive mixtures are modeled with instantaneous mixtures at each frequency bin and simple instantaneous independent component analysis (ICA) can be employed to separate the mixtures. However, the permutation and scaling ambiguities of ICA solutions need to be aligned to obtain proper time-domain separated signals. This paper discusses the idea that calculating the inverses of separation matrices obtained by ICA is very important as regards aligning these ambiguities. This paper also shows the relationship between the ICA-based method and the time-frequency masking method for BSS, which becomes clear by calculating the inverses.

This paper describes the frequency-domain blind source separation (BSS) of convolutively mixed acoustic signals using independent component analysis (ICA). The most critical issue related to frequency domain BSS is the permutation problem. This paper presents two methods for solving this problem. Both methods are based on the clustering of information derived from a separation matrix obtained by ICA. The first method is based on direction of arrival (DOA) clustering. This approach is intuitive and easy to understand. The second method is based on normalized basis vector clustering. This method is less intuitive than the DOA based method, but it has several advantages. First, it does not need sensor array geometry information. Secondly, it can fully utilize the information contained in the separation matrix, since the clustering is performed in high-dimensional space. Experimental results show that our methods realize BSS in various situations such as the separation of many speech signals located in a 3-dimensional space. and the extraction of primary sound sources surrounded by many background interferences.

One challenging problem for robust speech recognition is the cocktail party effect, where multiple speaker signals are active simultaneously in an overlapping frequency range. In that case, independent component analysis (ICA) can separate the signals in reverberant environments, also. However, incurred feature distortions prove detrimental for speech recognition. To reduce consequential recognition errors, we describe the use of ICA for the additional estimation of uncertainty information. This information is subsequently used in missing feature speech recognition, which leads to far more correct and accurate recognition also in reverberant situations at RT60 = 300ms.

The technique of blind source separation (BSS) has been well studied. In this paper, we apply the BSS technique, particularly based on independent component analysis (ICA), to a meeting situation. The goal is to enhance the spoken utterances and to estimate the location of each speaker by means of multiple microphones. The technique may help us to take the minutes of a meeting.

We discuss the frequency-domain blind source separation (BSS) of convolutive mixtures when the number of source signals is large, and the potential source locations are omnidirectional. The most critical problem related to the frequency-domain BSS is the permutation problem, and geometric information is helpful as regards solving it. In this paper, we propose a method for obtaining proper geometric information with which to solve the permutation problem when the number of source signals is large and some of the signals come from the same or a similar direction. First, we describe a method for estimating the absolute DOA by using relative DOAs obtained by the solution provided by independent component analysis (ICA) and the far-field model. Next, we propose a method for estimating the spheres on which source signals exist by using ICA solution and the near-field model. We also address another problem with regard to frequency-domain BSS that arises from the circularity of discrete-frequency representation. We discuss the characteristics of the problem and present a solution for solving it. Experimental results using eight microphones in a room show that the proposed method can separate a mixture of six speech signals arriving from various directions, even when two of them come from the same direction. Copyright (C) 2006 Ryo Mukai et al.

We propose utilizing subband-based blind source separation (BSS) for convolutive mixtures of speech. This is motivated by the drawback of frequency-domain BSS, i.e., when a long frame with a fixed long frame-shift is used to cover reverberation, the number of samples in each frequency decreases and the separation performance is degraded. In subband BSS, (1) by using a moderate number of subbands, a sufficient number of samples can be held in each subband, and (2) by using FIR filters in each subband, we can manage long reverberation. We confirm that subband BSS achieves better performance than frequency-domain BSS. Moreover, subband BSS allows us to select a separation method suited to each subband. Using this advantage, we propose efficient separation procedures that consider the frequency characteristics of room reverberation and speech signals (3) by using longer unmixing filters in low frequency bands and (4) by adopting an overlap-blockshift in BSS's batch adaptation in low frequency bands. Consequently, frequency-dependent subband processing is successfully realized with the proposed subband BSS.

This paper focuses on the underdetermined blind source separation (BSS) of three speech signals mixed in a real environment from measurements provided by two sensors. To date, solutions to the underdetermined BSS problem have mainly been based on the assumption that the speech signals are sufficiently sparse. They involve designing binary masks that extract signals at time-frequency points where only one signal was assumed to exist. The major issue encountered in previous work relates to the occurrence of distortion, which affects a separated signal with loud musical noise. To overcome this problem, we propose combining sparseness with the use of an estimated mixing matrix. First, we use a geometrical approach to detect when only one source is active and to perform a preliminary separation with a time-frequency mask. This information is then used to estimate the mixing matrix, which allows us to improve our separation. Experimental results show that this combination of time-frequency mask and mixing matrix estimation provides separated signals of better quality (less distortion, less musical noise) than those extracted without using the estimated mixing matrix in reverberant conditions where the reverberant time (TR) was 130 ms and 200 ms. Furthermore, informal listening tests clearly show that musical noise is deeply lowered by the proposed method comparatively to the classical approaches.

This paper overviews a total solution for frequency-domain blind source separation (BSS) of convolutive mixtures of audio signals, especially speech. Frequency-domain BSS performs independent component analysis (ICA) in each frequency bin, and this is more efficient than time-domain BSS. We describe a sophisticated total solution for frequency-domain BSS, including permutation, scaling, circularity, and complex activation function solutions. Experimental results of 2 x 2, 3 x 3, 4 x 4, 6 x 8, and 2 x 2 (moving sources), (#sources x #microphones) in a room are promising.

This paper presents a method for enhancing a target source of interest and suppressing other interference sources. The target source is assumed to be close to sensors, to have dominant power at these sensors, and to have non-Gaussianity. The enhancement is performed blindly, i.e. without knowing the total number of sources or information about each source, such as position and active time. We consider a general case where the number of sources is larger than the number of sensors. We employ a two-stage process where independent component analysis (ICA) is first employed in each frequency bin and time-frequency masking is then used to improve the performance further. We propose a new sophisticated method for selecting the target source frequency components, and also a new criterion for specifying time-frequency masks. Experimental results for simulated cocktail party situations in a room (reverberation time was 130 ms) are presented to show the effectiveness and characteristics of the proposed method. © 2005 IEEE.

This paper presents a method for enhancing a dominant target source that is close to sensors, and suppressing other interferences. The enhancement is performed blindly, i.e. without knowing the number of total sources or information about each source, such as position and active time. We consider a general case where the number of sources is larger than the number of sensors. We employ a two-stage processing technique where a spatial filter is first employed in each frequency bin and time-frequency masking is then used to improve the performance further. To obtain the spatial filter we employ independent component analysis and then select the component of the target source. Time-frequency masks in the second stage are obtained by calculating the angle between the basis vector corresponding to the target source and a sample vector. The experimental results for a simulated cocktail party situation were very encouraging. ©2005 IEEE.

This paper presents a method for estimating location information about multiple sources. The proposed method uses independent component analysis (ICA) as a main statistical tool. The nearfield model as well as the farfield model can be assumed in this method. As an application of the method, we show experimental results for the direction-of-arrival (DOA) estimation of three sources that were positioned 3-dimensionally. © 2005 IEEE.

Musical noise is a typical problem with blind source separation using a time-frequency mask. In this paper, we report that a fine-shift and overlap-add method reduces the musical noise without degrading the separation performance. The effectiveness was confirmed by results of a the listening test undertaken in a room with a reverberation time of RT60 =130 ms.

A new approach for estimating the number of sources that employs independent component analysis (ICA) is discussed. Estimating the number of sources provides information for signal processing applications such as blind source separation (BSS) in the frequency domain. The new method can identify a noise component that includes reverberations by calculating the correlation of the envelopes. The results show that the characteristics of the proposed approach compare with the conventional eigenvalue-based method.

This paper presents a method for enhancing a target source of interest and suppressing other interference sources. The target source is assumed to be close to sensors, to have dominant power at these sensors, and to have non-Gaussianity. The enhancement is performed blindly, i.e. without knowing the total number of sources or information about each source, such as position and active time. We consider a general case where the number of sources is larger than the number of sensors. We employ a two-stage process where independent component analysis (ICA) is first employed in each frequency bin and time-frequency masking is then used to improve the performance further. We propose a new sophisticated method for selecting the target source frequency components, and also a new criterion for specifying time-frequency masks. Experimental results for simulated cocktail party situations in a room (reverberation time was 130 ms) are presented to show the effectiveness and characteristics of the proposed method.

This paper presents a spatio-temporal extension of the well-known fastICA algorithm of Hyvarinen and Oja that is applicable to both convolutive blind source separation and multichannel blind deconvolution tasks. Our time-domain algorithm combines multichannel spatio-temporal prewhitening via multi-stage least-squares linear prediction with a fixed-point iteration involving a new adaptive technique for imposing paraunitary constraints on the multichannel separation filter. Our technique also allows for efficient reconstruction of individual signals as observed in the sensor measurements for single-input, multiple-output (SIMO) BSS tasks. Analysis and simulations verify the utility of the proposed methods.

This paper presents if prototype system for Blind Source Separation (BSS) of many speech signals and describes the techniques used in the system. Our System uses 8 microphones located at the vertexes of a 4cmx4cmx4cm cube and has the ability to separate signals distributed in three-dimensional space. The mixed signals observed by the microphone array are processed by Independent Component Analysis (ICA) in the frequency domain and separated into a given number of signals (LIP to 8). We carried Out experiments in all ordinary office and obtained more than 20 dB of SIR improvement.

A maximum it-posteriori approach for overcomplete blind source separation based on Laplacian priors usually involves l(1)-norm minimization. It requires different approaches for real and complex numbers it,; they appear for example in the frequency domain. In this paper we compare a combinatorial approach for real numbers with it second order cone programming approach for complex numbers.
Although the combinatorial solution with a proven minimum number of zeros is not theoretically justified for complex numbers, its performance quality is comparable to the performance of the second order cone programming (SOCP) solution. However, it has the advantage that it is faster for complex overcomplete BSS problems with low input/output dimensions.

This paper describes a real-time blind source separation (BSS) method for moving speech signals in a room. Our method employs frequency domain independent component analysis (ICA) using a blockwise batch algorithm in the first stage, and the separated signals are refined by postprocessing using crosstalk component estimation and non-stationary spectral subtraction in the second stage. The blockwise batch algorithm achieves better performance than an online algorithm when sources are fixed, and the postprocessing compensates for performance degradation caused by source movement. Experimental results using speech signals recorded in a real room show that the proposed method realizes robust real-time separation for moving sources. Our method is implemented on a standard PC and works in realtime.

The use of blind source separation (BSS) technique for the recovery of more than two sources inthe frequency domain was iprensented. It was found that frequency-domain BSS method was practically applicable for more than two sources by overcoming problem of permutation and circularity. The minimization error could be done by adjusting the scaling ambiguity of the independent component analysis (ICA) solution before windowing. The result shows that the effectiveness and efficiency of the BSS method and the separation of six sources with a planar array of eight sensors.

In this paper, we propose a new method for evaluating the separation and dereverberation performance of a convolutive blind source separation (BSS) system, and investigate a separating system obtained by employing frequency domain BSS based on independent component analysis (ICA). As a result, we reveal the acoustical characteristics of the frequency domain BSS for convolutive mixture of speech signals. We show that the separating system removes the direct sound of a jammer signal even when the frame length is relatively short, and it also reduces the reverberation of the jammer according to the frame length. We also confirm that the reverberation of the target is not reduced. Moreover, we propose a technique, suggested by the experimental results, for improving the quality of the separated signals by removing pre-echo noise.

In this paper, we propose a method for separating speech signals when there are more signals than sensors. Several methods have already been proposed for solving the underdetermined problem, and some of these utilize the sparseness of speech signals. These methods employ binary masks to extract the signals, and therefore, their extracted signals contain loud musical noise. To overcome this problem, we propose combining a sparseness approach and independent component analysis (ICA). First, using sparseness, we estimate the time points when only one source is active. Then, we remove this single source from the observations and apply ICA to the remaining mixtures. Experimental results show that our proposed sparseness and ICA (SPICA) method can separate signals with little distortion even in reverberant conditions of T-R=130 and 200 ms.

This paper presents a method for solving the permutation problem of frequency domain blind source separation (BSS) when the number of source signals is large, and the potential source locations are onmidirectional. We propose a combination of small and large spacing sensor pairs with various axis directions in order to obtain proper geometrical information for solving the permutation problem. Experimental results show that the proposed method can separate a mixture of six speech signals that come from various directions, even when two of them come from the same direction.

Blind source separation (BSS) for convolutive mixtures can be efficiently achieved in the frequency domain, where independent component analysis is performed separately in each frequency bin. However, frequency-domain BSS involves a permutation problem, which is well known as a difficult problem, especially when the number of sources is large. This paper presents a method for solving the permutation problem, which works well even for many sources. The successful solution for the permutation problem highlights another problem with frequency-domain BSS that arises from the circularity of discrete frequency representation. This paper discusses the phenomena of the problem and presents a method for solving it. With these two methods, we can separate many sources with a practical execution time. Moreover, real-time processing is currently possible for up to three sources with our implementation.

This paper introduces the blind source separation (BSS) of convolutive mixtures of acoustic signals, especially speech. A statistical and computational technique, called independent component analysis (ICA), is examined. By achieving nonlinear decorrelation,.nonstationary decorrelation, or time-delayed decorrelation, we can find source signals only from observed mixed signals. Particular attention is paid to the physical interpretation of BSS from the acoustical signal processing point of view. Frequency-domain BSS is shown to be equivalent to two sets of frequency domain adaptive microphone arrays, i.e., adaptive beamformers (ABFs). Although BSS can reduce reverberant sounds to some extent in the same way as ABF, it mainly removes the sounds from the jammer direction. This is why BSS has difficulties with long reverberation in the real world. If sources are not "independent," the dependence results in bias noise when obtaining the correct unmixing filter coefficients. Therefore, the performance of BSS is limited by that of ABF. Although BSS is upper bounded by ABF, BSS has a strong advantage over ABF. BSS can be regarded as an intelligent version of ABF in the sense that it can adapt without any information on the array manifold or the target direction, and sources can be simultaneously active in BSS.

This paper presents a method for solving the permutation problem of frequency domain blind source separation (BSS) when source signals come from the same or similar directions. Geometric information such as the direction of arrival (DOA) is helpful for solving the permutation problem, and a combination of the DOA based and correlation based methods provides a robust and precise solution. However when signals come from similar directions, the DOA based approach fails, and we have to use only the correlation based method whose performance is unstable. In this paper, we show that an interpretation of the ICA solution by a near-field model yields information about spheres on which source signals exist, which can be used as an alternative to the DOA. Experimental results show that the proposed method can robustly separate a mixture of signals arriving from the same direction.

In this paper, we study the performance effects caused by coefficient delays in natural gradient blind deconvolution and source separation algorithms. We present a statistical analysis of the effect of coefficient delays within such algorithms, quantifying the relative loss in performance caused by such coefficient delays with respect to delayless algorithm updates. We then propose a simple change to one such algorithm to improve its convergence performance.

In this paper we address the problem of overcomplete BSS for convolutive mixtures following a two-step approach. In the first step the mixing matrix is estimated, which is then used to separate the signals in the second step. For estimating the mixing matrix we propose an algorithm based on hierarchical clustering, assuming that the source signals are sufficiently sparse. It has the advantage of working directly on the complex valued sample data in the frequency-domain. It also shows better convergence than algorithms based on self-organizing maps. The results are improved by reducing the variance of direction of arrival. Experiments show accurate estimations of the mixing matrix and very low musical tone noise.

Practical gradient-based adaptive algorithms for multichannel blind deconvolution and convolutive blind source separation typically employ FIR filters for the separation system. Inadequate use of signal truncation within these algorithms can introduce steady-state biases into their converged solutions that lead to degraded separation and deconvolution performances. In this paper, we derive a natural gradient multichannel blind deconvolution and source separation algorithm that mitigates these effects for estimating causal FIR solutions to these tasks. Numerical experiments verify the robust convergence performance of the new method both in multichannel blind deconvolution tasks for i.i.d. sources and in convolutive BSS tasks for acoustic sources, even for extremely-short separation filters.

Blind source separation (BSS) for convolutive mixtures can be efficiently achieved in the frequency domain, where independent component analysis is performed separately in each frequency bin. However, frequency-domain BSS involves a permutation problem, which is well known as a difficult problem, especially when the number of sources is large. This paper presents a method for solving the permutation problem, which works well even for many sources. The successful solution for the permutation problem highlights another problem with frequency-domain BSS that arises from the circularity of discrete frequency representation. This paper discusses the phenomena of the problem and presents a method for solving it. With these two methods, we can separate many sources with a practical execution time. Moreover, real-time processing is currently possible for up to three sources with our implementation.

We propose a method for separating N speech signals with M sensors where N &gt; M. Some existing methods employ binary masks to extract the signals, and therefore, the extracted signals contain loud musical noise. To overcome this problem, we propose using a directivity pattern based continuous mask, which masks N - M sources in the observations, and independent component analysis (ICA) to separate the remaining mixtures. We conducted experiments for N = 3 with M = 2 and N = 4 with M = 2, and obtained separated signals with little distortion.

Practical gradient-based adaptive algorithms for multichannel blind deconvolution and convolutive blind source separation typically employ FIR filters for the separation system. Inadequate use of signal truncation within these algorithms can introduce steady-state biases into their converged solutions that lead to degraded separation and deconvolution performances. In this paper, we derive a natural gradient multichannel blind deconvolution and source separation algorithm that mitigates these effects for estimating causal FIR solutions to these tasks. Numerical experiments verify the robust convergence performance of the new method both in multichannel blind deconvolution tasks for i.i.d. sources and in convolutive BSS tasks for acoustic sources, even for extremely-short separation filters.

We propose a method for blindly separating real environment speech signals with as less distortion as possible in the special case where speech signals outnumber sensors. Our idea consists in combining sparseness with the use of an estimated mixing matrix. First, we use a geometrical approach to perform a preliminary separation and to detect when only one source is active. This information is then used to estimate the mixing matrix. Then we remove one source from the observations and separate the residual signals with the inverse of the estimated mixing matrix. Experimental results in a real environment (T-R=130ms and 200ms) show that our proposed method, which we call Sparseness Mixing Matrix Estimation (SMME), provides separated signals of better quality than those extracted by only using the sparseness property of the speech signal.

This paper presents a method for solving the permutation problem of frequency domain blind source separation (BSS) when the number of source signals is large, and the potential source locations are omnidirectional. We propose a combination of small and large spacing sensor pairs with various axis directions in order to obtain proper geometric information for solving the permutation problem. Experimental results in a room (reverberation time T-R = 130 ms) with eight microphones show that the proposed method can separate a mixture of six speech signals that come from various directions, even when two of them come from the same direction.

Blind source separation (BSS) for convolutive mixtures can be performed efficiently in the frequency domain, where independent component analysis (ICA) is applied separately in each frequency bin. To solve the permutation problem of frequency-domain BSS robustly, information regarding the number of sources is very important. This paper presents a method for estimating the number of sources from convolutive mixtures of sources. The new method estimates the power of each source or noise component by using ICA and a scaling technique to distinguish sources and noises. Also, a reverberant component can be identified by calculating the correlation of component envelopes. Experimental results for up to three sources show that the proposed method worked well in a reverberant condition whose reverberation time was 200 ms.

This paper discusses a nonlinear function for independent component analysis to process complex-valued signals in frequency-domain blind source separation. Conventionally, nonlinear functions based on the Cartesian coordinates are widely used. However, such functions have a convergence problem. In this paper, we propose a more appropriate nonlinear function that is based on the polar coordinates of a complex number. In addition, we show that the difference between the two types of functions arises from the assumed densities of independent components. Our discussion is supported by several experimental results for separating speech signals, which show that the polar type nonlinear functions behave better than the Cartesian type.

This paper presents a robust and precise method for solving the permutation problem of frequency-domain blind source separation. It is based on two previous approaches: the direction of arrival estimation approach and the inter-frequency correlation approach. We discuss the advantages and disadvantages of the two approaches, and integrate them to exploit the both advantages. We also present a closed form formula to calculate a null direction, which is used in estimating the directions of source signals. Experimental results show that our method solved permutation problems almost perfectly for a situation that two sources were mixed in a room whose reverberation time was 300 ms.

This paper describes a robust real-time blind source separation (BSS) method for moving speech signals in a room. Our method employs frequency domain independent component analysis (ICA) using a blockwise batch algorithm in the first stage, and the separated signals are refined by postprocessing using crosstalk component estimation and non-stationary spectral subtraction in the second stage. The blockwise batch algorithm achieves better performance than an online algorithm when sources are fixed, and the postprocessing compensates for performance degradation caused by source movement. Experimental results using speech signals recorded in a real room show that the proposed method realizes robust real-time separation for moving sources. Our method is implemented on a standard PC and works in realtime.

The goal of this contribution is a new algorithm using independent component analysis with a geometrical constraint. The new algorithm solves the permutation problem of blind source separation of acoustic mixtures, and it is significantly less sensitive to the precision of the geometrical constraint than an adaptive beamformer. A high degree of robustness is very important since the steering vector is always roughly estimated in the reverberant environment, even when the look direction is precise. The new algorithm is based on FastICA and constrained optimization. It is theoretically and experimentally analyzed with respect to the roughness of the steering vector estimation by using impulse responses of real room. The effectiveness of the algorithms for real-world mixtures is also shown in the case of three sources and three microphones.

This paper presents a new method for estimating the directions of source signals. We assume a situation in which multiple source signals are mixed in a reverberant condition and observed at several sensors. The new method is based on independent component analysis, which separates mixed signals into original source signals. It can be applied where the number of sources is equal to the number of sensors, whereas the conventional methods based on sub-space analysis, such as the MUSIC algorithm, are applicable where there are fewer sources than sensors. Even in cases where the MUSIC algorithm can be applied, the new method is better at estimating the directions of sources if they are closely placed.

Subband processing is applied to blind source, separation (BSS) for convolutive mixtures of speech. This is motivated by the drawback of frequency-domain BSS, i.e., when a long frame with a fixed frame-shift is used to cover reverberation, the number of samples in each frequency decreases and the separation performance is degraded. In our proposed subband BSS, (1) by using a moderate number of subbands, a sufficient number of samples can be held in each subband, and (2) by using FIR filters in each subband, we can handle long reverberation. Subband BSS achieves better performance than frequency-domain BSS. Moreover, we propose efficient separation procedures that take into consideration the frequency characteristics of room reverberation and speech signals. We achieve this (3) by using longer unmixing filters in low frequency bands, and (4) by adopting overlap-blockshift in BSS's batch adaptation in low frequency bands. Consequently, frequency-dependent subband processing is successfully realized in the proposed subband BSS.

In this paper, we discuss approaches for blind source separation where we can use more sensors than the number of sources for a better performance. The discussion focuses mainly on reducing the dimension of mixed signals before applying independent component analysis. We compare two previously proposed methods. The first is based on principal component analysis, where noise reduction is achieved. The second involves selecting a subset of sensors based on the fact that a low frequency prefers a wide spacing and a high frequency prefers a narrow spacing. We found that the PCA-based method behaves similarly to the geometry-based method for low frequencies in the way that it emphasizes the outer sensors and yields superior results for high frequencies, which provides a better understanding of the former method.

Natural gradient adaptation is an especially convenient method for adapting the coefficients of a linear system in inverse filtering tasks such as blind deconvolution and equalization. Practical implementations of such methods require truncation of the filter impulse responses within the gradient updates. In this paper, we show how truncation of these filter impulse responses can create convergence problems and introduces a bias into the steady-state solution of one such algorithm. We then show how this algorithm can be modified to effectively mitigate these effects for estimating causal FIR approximations to doubly-infinite IIR equalizers. Simulations indicate that the modified algorithm provides the convergence benefits of the natural gradient while still attaining good steady-state performance.

Frequency domain Blind Source Separation (BSS) is shown to be equivalent to two sets of frequency domain adaptive microphone arrays, i.e., Adaptive Beamformers (ABFs). The minimization of the off-diagonal components in the BSS update equation can be viewed as the minimization of the mean square error in the ABF. The unmixing matrix of the BSS and the filter coefficients of the ABF converge to the same solution in the mean square error sense if the two source signals are ideally independent. Therefore, the performance of the BSS is limited by that of the ABF. This understanding gives an interpretation of BSS from physical point of view.

We propose a time-domain BSS algorithm that utilizes geometric information such as sensor positions and assumed locations of sources. The algorithm tackles the problem of convolved mixtures by explicitly exploiting the non-stationarity of the acoustic sources. The learning rule is based on secondorder statistics and is derived by natural gradient minimization. The proposed initialization of the algorithm is based on the null beamforming principle. This method leads to improved separation performance, and the algorithm is able to estimate long unmixing FIR filters in the time domain due to the geometric initialization. We also propose a post-filtering method for dewhitening which is based on the scaling technique in frequency-domain BSS. The validity of the proposed method is shown by computer simulations. Our experimental results confirm that the algorithm is capable of separating real-world speech mixtures and can be applied to short learning data sets down to a few seconds. Our results also confirm that the proposed dewhitening post-filtering method maintains the spectral content of the original speech in the separated output.

Highly cross-correlated input signals create the problem of slow convergence of misalignment in stereo echo cancellation even after undergoing non-linear preprocessing. We propose a new frequency-domain adaptive algorithm that improves the convergence rate by increasing the contribution of non-linearity in the adjustment vector. Computer simulation showed that it is effective when the non-linearity gain is small.

Frequency domain Blind Source Separation (BSS) is shown to be equivalent to two sets of frequency domain adaptive microphone arrays, i.e., Adaptive Beamformers (ABFs). The minimization of the off-diagonal components in the BSS update equation can be viewed as the minimization of the mean square error in the ABF. The unmixing matrix of the BSS and the filter coefficients of the ABF converge to the same solution in the mean square error sense if the two source signals are ideally independent. Therefore, the performance of the BSS is limited by that of the ABF. This understanding gives an interpretation of BSS from physical point of view.

This paper presents a new type of nonlinear function for independent component analysis to process complex-valued signals, which is used in frequency-domain blind source separation. The new function is based on the polar coordinates of a complex number, whereas the conventional one is based on the Cartesian coordinates. The new function is derived from the probability density function of frequency-domain signals that are assumed to be independent of the phase. We show that the difference between the two types of functions is in the assumed densities of independent components. Experimental results for separating speech signals show that the new nonlinear function behaves better than the conventional one.

This paper describes a post processing method to refine output signals obtained by Blind Source Separation (BSS). The performance of BSS using Independent Component Analysis (ICA) declines significantly in a reverberant environment. The degradation is mainly caused by the cross-talk components derived from the reverberation of the jammer signal. Utilizing this knowledge, we propose a new method, time-delayed non-stationary spectral subtraction, which removes the residual components from the separated signals precisely. The proposed method compensates for the weakness of BSS in a reverberant environment. Experimental results using speech signals show that the proposed method improves the signal-to-noise ratio by 3 to 5 dB.

The performance of Blind Source Separation (BSS) using Independent Component Analysis (ICA) declines significantly in a reverberant environment. The degradation is mainly caused by the residual crosstalk components derived from the reverberation of the jammer signal. This paper describes a post-processing method designed to refine output signals obtained by BSS.
We propose a new method which uses LMS filters in the frequency domain to estimate the residual crosstalk components in separated signals. The estimated components are removed by nonstational spectral subtraction. The proposed method removes the residual components precisely, thus it compensates for the weakness of BSS in a reverberant environment.
Experimental results using speech signals show that the proposed method improves the signal-to-interference ratio by 3 to 5 dB.

This paper presents a method for blind source separation using several separating subsystems whose sensor spacing and filter length can be configured individually. Each subsystem is responsible for source separation of an allocated frequency range. With this mechanism, we can use appropriate sensor spacing as well as filter length for each frequency range. We obtained better separation performance than with the conventional method by using a wide sensor spacing and a long filter for a low frequency range, and a narrow sensor spacing and a short filter for a high frequency range.

Despite several recent proposals to achieve Blind Source Separation (BSS) for realistic acoustic signal, separation performance is still not enough. In particular, when the length of impulse response is long, performance is highly limited. In this paper, we show it is useless to be constrained by the condition, Pmuch less than T, where T is the frame size of FFT and P is the length of room impulse response. From our experiments, a frame size of 256 or 512 (32 or 64 ms at a sampling frequency of 8 kHz) is best even for the long room reverberation of T-R = 150 and 300 ms. We also clarified the reason for poor performance of BSS in long reverberant environment, finding that separation is achieved chiefly for the sound from the direction of jammer because BSS cannot calculate the inverse of the room transfer function both for the target and jammer signals.

The fundamental problems of stereophonic acoustic echo cancellation were discussed and the recent solutions were reviewed. The stereo echo cancellation was achieved by linearly combining two monoaural echo cancellers. A duofilter control system including a continually running adaptive filter and a fixed filter was used for double talk control. A second order stereo projection algorithm was used in the adaptive filter and a stereo switch was also implemented.

We investigated the dependence of the desired echo return loss on frequency for various hands-free telecommunication conditions by subjective assessment The desired echo return loss as a function of frequency (DERLf) is all important factor in the design and performance evaluation of a subband echo canceller, and it is a measure of what is considered all acceptable echo caused by electrical loss in the transmission line. The DERLf during single talk was obtained as attenuated band-limited echo levels that subjects did not find objectionable when listening to the near-end speech and its band-limited echo under various hands-free telecommunication conditions. When we investigated the DERLf during double-talk, subjects also heard the speech in the far-end room from a loudspeaker. The echo was limited to a 250-Hz bandwidth assuming the use of a subband echo canceller. The test results showed that: (1) when the transmission delay was short (30 ms), the echo component around 2 to 3 kHz was the most objectionable to listeners. (2) as the transmission delay rose to 300 ms, the echo component around 1 kHz became the most objectionable; (3) when the room reverberation time was relatively long (about 500 ms). the echo cumyonent around 1 kHz was the most objectionable even if the transmission delay was short; and ( 1) the DERLf during double-talk was about 5 to 10dB lower than that during single-talk. Use of these DERLf values will enable the design of mure efficient subband echo cancellers.

Sound localization is important to make conversation easy between local and remote sites in a teleconference. This requires a multi-channel sound system having a multi-channel acoustic echo canceller (MAEC). The appropriate number of channels is determined from a trade-off between high presence and MAEC performance, so it is not possible to increase the channel number by much.
We propose a channel-number-compressed MAEC to provide teleconferencing systems that exhibit high presence. The channel number of the MAEC inputs is compressed and that of its outputs is expanded.

This paper proposes a novel adaptive algorithm for an echo canceler. In this algorithm, the number of operations and memory capacity are equivalent to those of the conventional NLMS algorithm but the convergence speed is twice that using the conventional algorithm. This adaptive algorithm is referred to as subband ES (exponentially weighted stepsize). In the algorithm, the frequency bands of the received input signal and echo signal are divided into multiple subbands, and echo is independently canceled in each subband. Each adaptive filter in each subband has independent coefficients with an independent stepsize. The stepsize is time-independent and its weight is exponentially proportional to the change of the impulse response within the frequency region, such as the expected value of the difference between the waveforms of two impulse responses. As a result, the characteristic of the acoustic echo path in each frequency band is analyzed using the adaptive algorithm to improve the convergence characteristic. Using the results of computer simulation and experimental results obtained via an experimental setup with DSP, it is shown that the convergence speed with respect to input voice signal can be about 4 times faster when using echo cancellation based on the new algorithm than in conventional full-band echo cancellation based on the NLMS algorithm. (C) 1998 Scripta Technica, Electron Comm Jpn Pt 3, 82(3): 49-57, 1999.

Stereo echo cancellation has been achieved and used in daily teleconferencing. To overcome the non-uniqueness problem, a stereo shaker is introduced in eight frequency bands and adjusted so as to be inaudible and not affect stereo perception. A duo-filter control system including a continually running adaptive filter and a fixed filter is used for double-talk control. A second-order stereo projection algorithm is used in the adaptive filter. A stereo voice switch is also included. This stereo echo canceller was tested in two-way conversation in a conference room, and the strength of the stereo shaker was subjectively adjusted. A misalignment of 20 dB was obtained in the teleconferencing environment, and changing the talker's position in the transmission room did not affect the cancellation. This echo canceller is now used daily in a high-presence teleconferencing system and has been demonstrated to more than 300 attendees.

A new configuration for a stereo echo canceller with nonlinear pre-processing is proposed. The pre-processor which adds uncorrelated components to the original received stereo signals improves the adaptive filter convergence even in the conventional configuration. However, because of the inaudibility restriction, the preprocessed signals still include a large amount of the original stereo signals which are often highly cross-correlated. Therefore, the improvement is limited. To overcome this, our new stereo echo canceller includes exclusive adaptive filters whose inputs are the uncorrelated signals generated in the pre-processor. These exclusive adaptive filters converge to true solutions without suffering from cross-correlation between the original stereo signals. This is demonstrated through computer simulation results.

A multiple-point equalization filter using the common acoustical poles of room transfer functions is proposed, The common acoustical poles correspond to the resonance frequencies, which are independent of source and receiver positions. They are estimated as common autoregressive (AR) coefficients from multiple room transfer functions. The equalization is achieved with a finite impulse response (FIR) filter, which has the inverse characteristics of the common acoustical pole function. Although the proposed filter cannot recover the frequency response dips of the multiple room transfer functions, it can suppress their common peaks due to resonance; it is also less sensitive to changes in receiver position, Evaluation of the proposed equalization filter using measured room transfer functions shows that it can reduce the deviations in the frequency characteristics of multiple room transfer functions better than a conventional multiple-point inverse filter, Experiments show that the proposed filter enables 1-5 dB additional amplifier gain in a public address system without acoustic feedback at multiple receiver positions, Furthermore, the proposed filter reduces the reflected sound in room impulse responses without the pre-echo that occurs with a multiple-point inverse filter. A multiple-point equalization filter using common acoustical poles can thus equalize multiple room transfer functions by suppressing their common peaks.

This paper proposes a new subband stereo echo canceller that converges to the true echo path impulse response much faster than conventional stereo echo cancellers. Since signals are bandlimited and downsampled in the subband structure, the time interval between the subband signals become longer, so the variation of the crosscorrelation between the stereo input signals becomes large. Consequently, convergence to the true solution is improved. Furthermore, the projection algorithm, or affine projection algorithm, is applied to further speed up the convergence. Computer simulations using stereo signals recorded in a conference room demonstrate that this method significantly improves convergence speed and almost solves the problem of stereo echo cancellation with low computational load.

This paper proposes a new algorithm called the fast Projection algorithm, which reduces the computational complexity of the Projection algorithm from (p+1)L+O(p(3)) to 2L+20p (where L is the length of the estimation filter and p is the projection order.) This algorithm has properties that lie between those of NLMS and RLS, i.e. less computational complexity than RLS but much faster convergence than NLMS for input signals like speech. The reduction of computation consists of two parts. One concerns calculating the pre-filtering vector which originally took O(p(3)) operations. Our new algorithm computes the pre-filtering vector recursively with about 15p operations. The other reduction is accomplished by introducing an approximation vector of the estimation filter. Experimental results for speech input show that the convergence speed of the Projection algorithm approaches that of RLS as the projection order increases with only a slight extra calculation complexity beyond that of NLMS, which indicates the efficiency of the proposed fast Projection algorithm.

A new model(Common-Acoustical-Pole and Zero model:CAPZ model)is proposed for a room transfer function(RTF)by using common acoustical poles that correspond to resonance properties of a room. These poles are estimated as the common AR coefficients of many RTFs corresponding to different source and receiver positions. Using the estimated common AR coefficients,the proposed method models the RTFs with different MA coefficients.This new model requires far fewer variable parameters to represent RTFs than the conventional ab-zero or pole, zero model.The acoustic echo canceller based on the proposed model requires half the variable parameters and converges 1.5 times faster than one based on the all-zero model,confmning the efficiency of the proposed model.

This paper proposes a new adaptive algorithm for acoustic echo cancellers with four times the convergence speed for a speech input, at almost the same computational load, of the normalized LMS (NLMS). This algorithm reflects both the statistics of the variation of a room impulse response and the whitening of the received input signal. This algorithm, called the ESP (exponentially weighted step-size projection) algorithm, uses a different step size for each coefficient of an adaptive transversal filter. These step sizes are time-invariant and weighted proportional to the expected variation of a room impulse response. As a result, the algorithm adjusts coefficients with large errors in large steps, and coefficients with small errors in small steps. The algorithm is based on the fact that the expected variation of a room impulse response becomes progressively smaller along the series by the same exponential ratio as the impulse response energy decay. This algorithm also reflects the whitening of the received input signal, ie., it removes the correlation between consecutive received input vectors. This process is effective for speech, which has a highly non-white spectrum. A geometric interpretation of the proposed algorithm is derived and the convergence condition is proved. A fast projection algorithm is introduced to reduce the computational complexity and modified for a practical multiple DSP structure so that it requires almost the same computational load, 2L multiply-add operations, as the conventional NLMS. The algorithm is implemented in an acoustic echo canceller constructed with multiple DSP chips, and its fast convergence is demonstrated.

This paper reviews echo control techniques for telecommunications, emphasizing the principles and applications of both circuit and acoustic echo cancellers. First, echo generating mechanisms and echo problems are described for circuit and acoustic echoes. Circuit echo is caused by impedance mismatching in a hybrid coil. Acoustic echo is caused by acoustic coupling between loudspeakers and microphones in a room. The echo problem is severe when the round-trip propagation delay is long. In this case, the echo must be removed. Next, the basic principle of the echo canceller, adaptive filter structure and adaptive algorithm are discussed. Emphasis is focused on the construction and operation of an adaptive transversal filter using the NLMS (Normalized Least Mean Square) algorithm, which is the most popular for the echo canceller. Then, applications of circuit and acoustic echo cancellers are described. Circuit echo cancellers have been well studied and implemented in LSIs for many applications. Although acoustic echo cancellers have been introduced into audio teleconference systems, they still have some problems which must be solved. Therefore, they are now being studied intensely. Finally, this paper mentions the problems of echo cancellers and the direction of future work on them. The main targets for acoustic echo cancellers are improving the convergence speed, reducing the amount of hardware and bettering the double-talk control technique. © 1990, Acoustical Society of Japan. All rights reserved.

A new configuration of acoustic echo canceller for multiple microphone teleconferencing systems is proposed. It is designed for use with microphones whose gains switch or vary during teleconferencing according to the talker. This system requires memory for multiple echo paths, which enables the updating of filter coefficients when an echo path is changed due to the switching of the actuated microphone during talker alternation. In comparison to the single echo path model which uses only adaptation, this method maintains echo cancellation during abrupt changes of the echo path when the microphone alternates between talkers. Also in comparison to direct microphone output mixing, this method reduces the stationary residual echo level by the reduction of acoustic coupling. © 1989, Acoustical Society of Japan. All rights reserved.