Study On Speech Separation And Speech Enhancement Methods

In our lives, speech signals are often disturbed by various interferences such as background noise and room reverberation. The existing of noise and reverberation not only affects human hearing, but also has influence on other steps of speech signal processing. So it is important to enhance speech using signal processing technology. In fact, besides speech enhancement, speech separation can also get rid of the influence of noise and reverberation. Blind Source Separation (BSS) is a challenging subject and becomes to be a popular research area in signal processing field in recent years. In the last two decades, a number of algorithms addressing the instantaneous BSS problems have been proposed and gained some results. However, current results are far from its mature solution. Moreover research on real-world convolutive speech signals has just begun.This thesis focuses on the methods of speech separation and speech enhancement which can be divided into three parts:Signal sparsity based undetermined blind source separation algorithm. In the case of anechoic and undetermined mixing mode, the independent component analysis based blind source separation algorithm can not achieve desired performance. To solve this problem, signal sparsity based time-frequency masking algorithm and line orientation separation technique is studied. Theory analysis and computer simulations indicate that both of the algorithms have achieved desired performance. However, there is a large number of zeros in the mask constructed by time-frequency masking algorithm. To solve this problem, some improvement has been made on it. As a result, the quality of the enhanced speech signal has been improved further.Independent component analysis with reference. In the blind source separation algorithm, people usually know something about the sources. Using this a priori information so as to extract the desired source is the subject of the algorithm, independent component analysis with reference. Experiment results indicate that the algorithm can perfectly extract the interested original signal witch has the strongest relationship with the reference signal in some sense. At the same time, comparing to the conventional independent component analysis algorithm, it has less computation consuming.Speech enhancement based on blind source separation with post-processing in subband. In the case of noise and reverberation coexisting, the whole algorithm is studied. The method has achieved high performance in depressing noise and reverberation. Regrettably, the noise was depressed at the expense of speech distortion. To solve this problem, some improvements have been made on it. This is that independent component analysis operations in low frequency subbands are replaced by the efficient time-frequency masking method. Experimental results indicate that a higher performance has been achieved through these improvements.