Effect of Time Constant on Speech Enhancement in Hearing Aids Based on Auditory Neural Feedback

摘要

The medial olivocochlear neuron feedback loop in the human auditory system plays a critical role in speech-in-noise perception. Recent studies have successfully simulated the mechanism of the neuron feedback and developed corresponding speech enhancement algorithms for hearing applications. These algorithms apply time-varying attenuation to hearing applications by reducing the gain in each frequency band over time. The speed of gain reduction is characterized by a time constant. In hearing applications, it is known that speech intelligibility is related to the gain regulating time constant. However, the effect of the time constant on enhancement performance is unclear. This paper studies its effect on speech intelligibility in hearing aids by incorporating a time constant variable speech enhancement algorithm modified from an existing neuron feedback model into a hearing aid model. The study demonstrates intelligibility improvement of over 30% at SNR between 5 and 10 dB. The longer time constants (≥ 1000 ms) show greater intelligibility improvement (about 20%) at SNR ≤ 10 dB, whilst the shorter time constants show more benefits at SNR > 10 dB.