A gated recurrent unit approach for improving audio quality in hearing induction loop systems

Hearing induction loop systems often encounter uneven field distribution, metal-induced signal loss, and overspill, which may reduce speech intelligibility. This study presents a digital signal processing framework centered on gated recurrent unit (GRU) networks to address these challenges. Real-world noise measurements, including interference from power and data cables, are integrated into the training process to represent practical field conditions. By monitoring the signal-to-noise ratio, the GRU-based approach aims to reduce interference while preserving key speech cues. A simplified attention mechanism and spectral feature extraction are also incorporated, allowing the model to highlight crucial time-frequency segments. Preliminary tests, conducted in both simulated and real-world scenarios, suggest improvements in noise mitigation and intelligibility compared with baseline methods. These findings may help align theoretical low-spill designs with actual performance, ultimately supporting broader accessibility in venues such as lecture halls and event spaces.