Wavelet Filter-bank And Psychoacoustic Modeling For Speech Enhancement

In general,speech is often corrupted inevitably by ambient noise, so a system for speech enhancement is needed strongly to solve the problem. The objective of speech enhancement may be to improve the quality, to increase the intelligibility, to ensure the reliability of digital voice system. Depending on the specific application, the enhancement system may be directed at different objectives.This thesis addresses the problem of single channel speech enhancement at low signal-to-noise ratios. At first,the conventional enhancement methods are reviewed, focusing on spectral subtraction and minimum mean-square error method, as well as kalman filter method which based on speech model. The advantage and disadvantage of these methods are introduced. Wavelet transform has the character of mutil- resolution, it is applied to non-stationary signals, such as speech signals. In this paper, the method based on wavelet transform for speech enhancement is researched in- depth.The main points of this thesis are given as following:1. The applications of wavelet in speech signal processing are studied. A new thresholding value, Neyman-Pearson criterion is proposed compared with the conventional criterion. Experiment show that the proposed approach has the best performance.2. A wavelet-based auditory model is realized with a perceptual wavelet filter-bank to map the frequency response of human auditory system into the wavelet domain. Psychoacoustic experiments have revealed that the peripheral auditory system behaves as a filter-bank. The concept of the auditory filter is closely linked to the behavior of the basilar membrane in the inner ear. Since wavelet transforms offer the capability of producing a good frequency resolution at low-frequency and good time resolution at high-frequency, its characteristic is similar to human audition. So we propose a new model adopting the basic structure of traditional auditory model but replace the time-invariant bandpass filters with wavelet transforms. By virtue of matching time-frequency representations, the masking effect can be integrated into the proposed model for effective noise suppression.3. A new sub-band adaptive filter based on wavelet filter-bank for speech enhance- ment is proposed. The adaptation of Kalman filter in wavelet domain has effectively reduced the non-stationary noise. A perceptual weighting filter exploiting the masking properties of psychoacoustic model is concatenated with the Kalman filter to further improve the intelligibility of speech. The proposed method owns its merits from the successful porting of Kalman filter into the wavelet domain so that speech analysis and enhancement can be carried out in time-frequency spectrum based on the human auditory model. Experimental results show that the new speech enhancement system is capable of reducing noise with little speech degradation in adverse noise environments and the overall performance is superior to conventional methods.