Performance evaluation of a speech enhancement technique using wavelets

Abstract

In this work, a novel speech enhancement algorithm is proposed based on multiband spectral subtraction speech enhancement technique and wavelet thresholding. The algorithm is tested with noisy speech signal produced by a prosthetic device for laryngectomy patients. The performance of the proposed algorithm is compared with the multiband spectral subtraction algorithm in terms of perceptual evaluation of speech quality (PESQ). The objective measures such as signal-to-noise-ratio (SNR), log-likelihood ratio (LLR), segmental signal-to-noise-ratio (SegSNR), weighted spectral slope (WSS), itakura-saito distance (IS), Cepstral Distance, and frequency-weighted segmental signal-to-noise-ratio (fwSNR) are used to test the effectiveness of the algorithm.