A wavelet based speech enhancement method using noise classification and shaping

Speech enhancement systems performing in Fourier or wavelet domain usually generate musical noise and distortion. It is possible to reduce musical noise and speech distortion by shaping residual noise. In this paper, we propose to implement a noise shaping method for a wavelet based noise reduction system. For noise shaping, we propose a noise classification method based on spectral shape of input noise. Using this classification method, we transform input noise to another noise which is more acceptable from listening point of view. Objective and subjective test results show that using noise shaping method; we obtain less distortion in speech signal in comparison to basic wavelet noise reduction system. Furthermore, listening test results illustrate that background shaped noise is less annoying.