Speech Enhancement Based On Auditory Masking And Auditory Wavelet Packet Decomposition

With the progress of society, speech processing systems and communication technology are required and developing very quickly. Many of the speech systems are using the speech enhancement technology. The purpose of speech enhancement is to remove the background noise and improve speech quality, raise the definition of the speech and improve the robust of the follow-up speech processing systems.As the research going on in wavelet analysis, the researchers have been proposed an auditory wavelet packet speech decomposing method in the field of speech processing systems. And it plays an important role and opens up an exciting new field in the speech enhancement. Given the human auditory characteristics and the minimum value of statistical methods in the frequency domain speech enhancement method has been made good results, this article will study the basic principles of these methods and modify these methods to apply in the time-frequency wavelet domain speech enhancement processing.This paper studied the speech enhancement algorithms based on the characteristics of auditory perception in time-frequency domain, the main work are as follows:1. Analyze the auditory wavelet packet decomposition in the speech enhancement, introduce a auditory wavelet packet decomposition method in the single-channel speech signals. This decomposition method is basing on the reality that there are critical bands in the human ears. Decomposing the speech signal to every critical band and the corresponding wavelet sub-bands makes the speech signal effectively concentrated and easy to deal with in the further de-noising processing.2. Propose a threshold estimation method using the minimum statistic method. It can track the noise energy dynamically in the auditory wavelet sub-bands and set up a reasonable threshold. Experiment show that this method can dynamically track the noise energy and set the wavelet threshold in the time-varying colored noise environment, providing an effective guarantee to the follow-up wavelet de-noising process.3. Propose a masking threshold function based on human auditory. According to the phenomenon that the signals in different frequencies will influence each other and effect the human ear's hearing threshold. A threshold function is proposed that can adjust the threshold parameter to control the suppression of the noise according to the masking threshold resulted by the current distribution of the speech. The experiments show that this method have good de-noising ability and have a small speech processing distortion, achieve a good enhancement result.