This study aims to improve the listening experience of people with hearing impairments by enhancing the audio quality of Hearing Induction Loop Systems (HILS). These systems deliver sound through magnetic fields to telecoil-equipped hearing aids in public venues such as theaters and lecture halls. However, their performance can be degraded by environmental noise, electromagnetic interference (EMI), and metal loss. To address these issues, a deep-learning denoising model based on the Gated Recurrent Unit (GRU) architecture was developed and experimentally validated.A 3.5 × 3.5 m low-spill multi-loop system was designed to minimize magnetic-field overspill and maintain uniform signal strength in accordance with IEC 60118-4 standards. The GRU model was evaluated against conventional approaches including Wiener filtering, STFT-based spectral subtraction, and VAD-based masking. Experimental results showed that the GRU network consistently improved perceptual speech quality and signal-to-distortion ratio while retaining clearer time-frequency features of speech signals. These findings demonstrate that AI-based methods can significantly reduce noise without compromising natural speech clarity.The research was presented at the International Congress on Acoustics (ICA 2025, New Orleans, USA), where it received strong interest from both academia and industry. Discussions highlighted the potential of real-time deployment in hearing aids and cochlear implants, addressing issues such as computational efficiency and low-latency signal processing. Participants also emphasized the importance of developing accessible sound environments for public spaces.