Noise reduction for auditory prostheses based on harmonic detection

Abstract

Difficulty in understanding speech in background noise is one of the most common complaints of hearing-aid and cochlear-implant users. Various noise-reduction and spectral-enhancement algorithms have been designed and tested over the years, often with limited success in terms of improving speech intelligibility. A commonly used method involves spectral subtraction, which is based on the assumption that the noise spectrum can be estimated and the clean speech signal can be extracted by subtracting the noise spectrum from the noisy speech signal. However, recent studies have shown that such methods often result in poorer signal-to-noise ratios in the modulation-spectrum domain, which may explain why little benefit in speech intelligibility has been found. Also, identifying noise in terms of its stationarity runs the risk of misidentification in more stationary signals, such as music. Here a noise reduction algorithm based on harmonic detection and enhancement was explored. Simulation results showed that this algorithm could help suppress noise in both speech and music. Results from perceptual tests will be reported. [Supported in part by Advanced Bionics.]