Hearing-impaired listeners often have great difficulty understanding speech in a noisy background. The problem has motivated the development of a new speech enhancement scheme with the goal of improving speech in noise perception for the hearing impaired listeners. In this thesis, a novel wavelet packet based noise reduction algorithm and hearing loss compensation are presented for a single microphone hearing aids application.
The noise reduction scheme utilizes noise masking threshold based suppression rule to remove additive noise. The perceptual noise suppression rule is optimized to achieve a balance between noise removal and speech distortion. Both objective and subjective evaluations have shown superior performance of the proposed technique in a good combination of low residual noise and low signal distortion.
The hearing loss compensation is realized by the wavelet-based loudness compression in each critical band. The compensated speech is guaranteed above hearing-impaired listener's threshold of hearing and with growth of loudness corrected in the dynamic range. Preference test among normal hearing person with simulated hearing loss has shown compensated speeches are favored in various conditions.