Speech Enhancement And Separation Based On Microphone Array

Noise and speech signals from competing talkers are the common disturbance in speech communications. The human being's acoustical system can distinguish and follow the interested speech signal in several speakers' situation, and tell apart the speech that he wants. This ability of distinguishing is a special perceptual capacity of the speech comprehension mechanism in human body, that is, people have the speech separation ability. This is called "cocktail-party problem". However, the existing speech processing systems can not do this like human beings. Their performance is seriously influenced by noise and other talkers. For example, the speech recognition system can achieve high recognition rate with pure speech signals, but when the input is corrupted by noise, its performance is degraded dramatically.Speech enhancement is a useful forward means to restrain the interference from the received speech signals that are inevitably interfered by noise. In recent years, microphone array speech enhancement technique is widely applied in vehicle speech acquisition, multimedia conference and robot control, etc. Compared with signals received by single microphone, the signals received by microphone array can be processed not only in time/frequency domain, but also in spatial domain. Consequently, microphone array possesses capabilities of strong interference suppression, speech sources localization, tracking and separation, etc. For this reason, it has been proposed as a promising solution to high quality of speech communication in such applications as teleconference, hands-free mobile telephone and hearing aids. The research of microphone array is a new area in array signal processing with broad application prospects.Blind Source Separation (BSS) is to recover the unknown independent sources from several mixed signals according to the signal's statistical characteristic without any knowledge of the sources and channels. Independent component analysis (ICA) is a new blind separation technique which appears during the research of BSS. Since its appearance, ICA has become a hot topic in signal processing, data analysis, statistics and neural networks, etc. It has been widely used in speech processing, biomedical signal processing, pattern recognition, feature extraction, data compression, image processing, and telecommunications, etc.Presently, many methods for speech enhancement have been proposed. However, speech enhancement in highly noisy environments is still a challenging problem. The research on blind deconvolution plays an important role in real applications of speech separation technique. Blind deconvolution in the frequency domain is an effective solution, but the order ambiguity becomes a serious problem. My dissertation plays emphasis on the above two problems.In this dissertation, we summarize and analyze the previous work on microphone array speech enhancement and separation. We make research on microphone array speech enhancement in highly noisy environments and blind speech separation in the frequency domain, and present two valid methods to solve two problems, i.e., speech enhancement in highly noisy surroundings and order ambiguity in frequency-domain BSS:1.Speech enhancement in highly noisy environment based on microphone array speech enhancement technique and ICA. ICA can separate independent components from mixtures without any prior knowledge of sources and channels. In this method, we analyze the sensor signals using ICA to obtain signals of better quality. The signals after ICA processing are then taken as inputs to the following microphone array speech enhancement system. The experiments show that the performance by this method is remarkable in highly noisy surroundings.2.Solving order problem in frequency-domain blind speech separation using microphone sub-arrays. In this method, we estimate DOA of sources using sub-arrays. Permutations are aligned based on the DOA estimation in the frequencies where the confidence of DOA estimation is sufficiently high. For the remaining frequencies, permutations are aligned using correlation method. The experiments in real environment show that this method is more robust, especially in low frequencies.However, there are still many issues unresolved in the microphone array speech enhancement and separation. At the end of this dissertation, the future directions of our research are summarized and prospected.