Research On Microphone Array Speech Enhancement

With the development of digital signal processing theory and application of electronic technology, in people's life the system of noise reduction get more and more applications. Under the conditions of the vehicle phone, video conferencing and other occasions, there are not only the existence of environmental noise, but also to an echo and reverb, so the traditional single-channel speech enhancement method look powerless. Therefore, it has posed a multi-channel speech enhancement methods. Microphone array has become an important component of these methords.Time-delay estimation and beamforming are two core technologies to microphone array speech enhancement, this paper is to study the beamforming methods for the post-beamforming filtering method. If we choose the suitable methords for the microphone array or improve a methord, we can get better test result.Post-filtering beamforming methord has two parts, the one is delay-sum Beamformer and the other is post-filter. The method of Post-filter make the direct impact on results, thus improving the traditional post-filtering algorithm or design a new enhancement algorithm have a direct effect. In this paper, we combine the sub-space method and the wavelet threshold method, the microphone array signals collected post-processing of related algorithms have been refined and simulated, and the results obtained were analyzed. In the sub-space part, we use the eigen decomposition method based on frequency domain to enhanced the voice, of which noise is estimated by the minimum statistics noise estimation method instead of the traditional VAD detection of noise estimation method, which can estimate noise much better, so as to provide a more accurate basis for the decomposition of noisy speech signal. For the wavelet threshold denoising part , with the analysis of the nature of the wavelet functions in the wavelet analysis , this paper selects the db6 wavelet in the db wavelet speech enhancement system as the wavelet decomposition wavelet function. The set-valued function using a compromise threshold function which combine the soft and hard thresholding function, which can not only gets the advantages of both but also to overcome the shortcomings of both to a certain extent. At last we find that in this paper the speech enhancement methods which we design has the advantages of the sub-space method and wavelet thresholding method have. Compare with the traditional post-filtering method, the one we designed in this paper can achieve a better test result and achieve a better denoising effect.