Speech enhancement using perceptual multi_band Wiener filter

This paper introduces a new perceptual speech enhancement method based on an improved Wiener filter. The classic noise reduction techniques, including the Wiener filter, introduce musical noise and distortions that remains in the enhanced speech signal, which consequently affects the perceptual appearance. Since the spectrum of speech signal is not affected uniformly by the noise it is interesting to analyze and process the signal in each frequency subband. For this speech signal will undergo a first filtering by a gammachirp filterbank. Signal will be divided into frequency subbands. Each frequency subband will be analyzed and processed by a second filter that is the classic Wiener filter. The latter will allow us to eliminate noise in each frequency subband as well as to calculate the threshold of hearing according to the Johnston model. The calculated threshold is then used in the third filter in order to improve the perceptual appearance of the resulting signal. The objectives and subjective results show that the proposed method is more effective not only in reducing noise but also in optimizing distortions in the enhanced speech signal.