Authors: Clara Yaiche¹, Boltzmann Li²
¹AAL; ²Aizip
Background: Hearing aids primarily restore audibility through amplification calibrated to the patient’s audiogram. However, amplification alone is insufficient to improve speech intelligibility. While AI denoisers have shown breakthroughs in performance, most do not meet the real-time and computational constraints of edge audio devices, specifically regarding the echo perception threshold and the preservation of speech quality and environmental context.
Method: This presentation details the integration of a neural network-based denoiser into Absolute Audio Labs’ hearing aid software chain, developed by Aizip. The work explains the technical requirements for real-time edge AI processing, the challenges of maintaining low latency, and the methodology for evaluating performance through specific metrics and perceptual testing.
Results: We demonstrate how an edge-AI neural network can achieve a breakthrough in denoising without compromising speech signal quality. The submission outlines how the integration manages to balance noise removal with the preservation of essential auditory information required by the hearing impaired.
Conclusion: This work demonstrates that a state-of-the-art AI denoiser can successfully be implemented within the strict hardware constraints of hearing aids. Such advancements improve speech intelligibility and provide significant real-world benefits for hearing aid users in complex acoustic environments.
