Study On Speech Enhancement Based On Human Auditory Masing Properties

We detail the speech enhancement technique in the circumstance of adding noise. First, we provide basic theory of speech signal processing, which is foundation of the research and implement of speech enhancement. Second, we introduce traditional speech enhancement method such as spectral subtraction, wiener filtering and subspace method etc, and then we compare the advantage and shortcoming of these methods. Third we make detailed discussion on the method based on human auditory masking properties. The processing in the Bark domain allows us to take into account in an optimal manner the masking properties of the human auditory system. Considering that the clearness and intelligibility are more important, we provide an improved algorithm based on Virag's method. It is shown in the simulated waveform that it provide speech of better clearness and intelligibility. Subsequently we provide an algorithm based on wiener filter and human auditory masking properties, which has low residual music noise and better clearness than conventional methods do. In addition, we briefly introduce subspace and wavelet approaches for enhancing speech corrupted by colored noise, and give the simulated waveform of wavelet approach. At last we compare the performance of these methods in subjective and objective way.This algorithm is simulated in Matlab, and implemented in Visual c++ under Windows.