The Microphone Array Speech Enhancement Based On The Statistical Model After Filtering Method

Noise exists everywhere in the human living environment, and clean speech signal is easily disturbed by noise. The speech signal processing technologies(e.g., speech recognition,speech synthesis and speech control) require clean speech signal from the noisy speech for subsequent processing. However, the noise seriously affects the performance of the application of these technologies, even leads to failure of the system performance. So, it’s urgent to learn how to obtain pure speech from noise to make speech enhance. Has always been based on microphone array speech enhancement algorithm is an important research direction in the field of speech enhancement, space array has a good selectivity, can keep expected language to gain the same direction at the same time, weakening the other direction to do additional noise, so as to realize good denoising effect. Collected by these methods,however, the speech contained residual noise, to enhance the performance of the increase is limited, for this shortage, this paper puts forward after added to a model based on statistical filtering module, improve the enhanced speech data signal-to-noise ratio.In this paper, the microphone array beam forming combined with the method based on statistical model, put forward a new enhancement method based on statistical model filtering.The work of this paper is mainly has the following several aspects:1. This paper introduces several kinds of common speech enhancement algorithm, and in the third chapter, respectively, to evaluate their effect;2. Introduced basic method and microphone array speech enhancement method, and introduces their application in speech enhancement algorithm’s advantages and disadvantages;3. Improved filtering based on statistical model, beam forming method and using MATLAB and c++implementation of the method. At first,use of broadband microphone array beam forming to enhance the speech with noise, analysis the result of enhances; In the second,use HTK toolkit training of GMM model of clean speech and noise GMM model calculation is the voice of the GMM model with noise; At last, use of the speech with noise model of a prior information of microphone array to enhance residual noise. After the speech enhancement result evaluation, it is concluded that after filtering method based on statistical model of residual noise signal filtering effect is very significant.