Less Given Aliasing Voice Signal Blind Separation Method

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 apperception capacity of the speech comprehension mechanism in the human body, that is, people have the speech separation ability, and it is called "cocktail-party problem".In the speech and acoustical signal processing area, how to imitate human's speech separation ability and separate or extract the original or interested speech from the observed speech mixtures, has become an important problem. This is also a significant research direction in speech signal processing, and it has very active signification to speech recognition and speech enhancement.Blind Source Separation (BSS), also called blind signal separation, is to recover the unknown independent sources from several observed mixed signals according to the signal's statistical characteristic without any knowledge of the sources and channels. Generally, to simplify the research work, BSS methods mostly concentrate on the overcomplete case or complete case, that is, the number of observed signals is not less than that of source signals. However, in the real speech communications, we always meet the underdetermined case, that is, the number of observed signals is less than that of source signals. So it is of great significance to resolve problem of speech separation in the underdetermined case.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. And it has been widely used in speech processing, biomedical signal processing, pattern recognition, feature extraction, data compression, image processing, and telecommunications, etc. Presently, there have been several ICA algorithms given by some scholars, however, most of these ICA algorithms ignore the underdetermined case, because they can't separate speeches well under the underdetermined case. Sparse representation of signals has received a great deal of attention in recent years. For the signal in time domain is usually not sparse, so an effective sparse representation of the signal is necessary. Sparse component analysis (SCA) is a method of signal processing based on sparse representation. SCA estimate the sparsest signals from observations which is different from ICA. The speech sources can be sparse in the Time-Frequency (TF) domain if a suitable linear transformation is performed. Sparse representation has been widely applied in underdetermined BSS.Currently, most of speech separation algorithms consider the noise-free case. But in the real speech communications, speech signals can inevitably be interfered by the background noise. So it is of great value and significance to resolve problem of speech separation in the noisy environment. Noisy speech mixtures involve not only the speeches of different people but also the background noise, so it is very difficult to separate the original speeches. Presently some scholars are devoting themselves to the research of noisy BSS algorithms, but the achievement is little.In this dissertation, we analyze and summarize the previous work of BSS. We make research on blind speech separation in underdetermined case and in noisy environment, and then present several valid resolving methods mainly based on the SCA and ICA technique:1. A new two-stage approach to underdetermined BSS based on sparse representation is proposed. According to SCA and basic criterion of ICA, a new sparse representation based on high-order statistics in transformed domain, which is called statistically sparse component analysis (SSCA), is proposed to estimate the sources. Compared with the existing two-stage methods, better separation performance is obtained in the proposed approach.2. To resolve the noise, wavelet transform is added to de-noise for the above method. Computer simulation results exhibit good separating performance.3. Underdetermined speech separation in transform domain is discussed.We combine ICA and Binary Time Frequency Masking technique to separate the mixed speech signals.Computer simulation results exhibit a good level of separating effect.In addition, there are very few research results of underdetermined speech separation. Two different methods are researched combining SCA and ICA, and get good results.At the last of this dissertation, the future research directions of this thesis are summarized and prospected.