Study On Blind Source Separation Based Speech Enhancement Methods

In our lives, Speech is often corrupted acoustically by ambient noise which produces aesthetically undesirable effects on the performance of digital voice processor and even diminishes communication system's ability to convey information across the interface. Therefore a speech enhancement system is strongly needed whose responsibility is improving the speech quality and ensuring reliability of digital voice communication systems. For many years progresses have been made in this field to find good methods for speech enhancement.This thesis focuses on the methods of speech enhancement which can be divided into three parts:Voice activity detection has been discussed. The voice activity detection is a very important accessory measure in digital speech processing. The previous methods either require the statistical information of the speech and the noise, which make it difficult to perform in time, or can not perform well in low signal to noise ratio circumstance. Furthermore, they are mostly based on the detection technology with one-channel signal. That is to say each array signal is detected by using existing voice activity detection method with one-channel signal respectively. This will result in more computation consuming. Aiming at these problems, voice activity detector based on the noise type distinguishing with microphone array is proposed. Experimental results show the validity and less computation consuming of the method.Speech enhancement method based on blind source separation with post-processing in subband in the case of noise and reverberation environments is concerned. Performance analysis and computer simulations indicate the poor performance of this method in suppressing reverberation, uncorrelated and mild correlated noises, and it introduced audible distortion to the enhanced signal. To apply the method in practical environment, an improvement has been made on it. That is the adaptive noise cancellers are only used in the subbands with poor separation results. Through this method, the quality of the enhanced speech signal has been improved. Simulation results show the effectiveness of the proposed method.Speech enhancement for multi-speaker is studied. To obtain clearer desired signal under multi-speaker environment, the time-frequency masking effect is used in post-processing of speech enhancement using blind source separation. Compared with expectation maximum...