Study On Methods For Speech Enhancement Based On Microphone Array

In an intelligent speech communication application such as the car, videoconference andmultimedia conference, the speech signal received by a microphone is often corrupted by thebackground noise, reverberation and interference. These may cause the degradation of speechquality, which will have a bad influence on the the performance of speech processing system.For such an adverse acoustic environment, therefore it is necessary to develop effective noisesuppress methods to recover the original speech from the microphone signal.The performance of some conventional single-microphone speech enhancement methodsseriously degrades in the environment where reverberation and interference coexist. Withspatial filtering and higher signal gain, microphone array (or multi-microphones) becomes anideal way of improving speech quality in the intelligent communication system. Microphonearray speech enhancement method has recently been a research focus in speech processing.This paper deals with some speech enhancement methods, the purpose of which is todevelop framework of microphone array speech enhancement technology in the differentacoutstic environments. All the approaches are studied along three lines, far-field vs.near-field beamforming, full-band vs. subband beamforming, and conventional vs. blindbeamforming, respectively. In addition, a detailed analysis of the relationship betweenmicrophone array configuration and array system performance is also given in the paper.The main contributions of the paper are as follows:(1) Study on combing full-band beamforming with single channel speech enhancementmethod. With the advantages of spatial filtering, a method of speech enhancement based onmicrophone array in frequency domain is proposed to suppress the noise in an enclosure. Amasking threshold based spectral subtraction method is additionally used to suppress theresidual noise of beamformer output. Some simulations results demonstrate that besides lowcomputation complexity, noise reduction performance of the proposed method is better thanthat of other methods.(2) Study on subband adaptive beamforming based microphone array speechenhancement method. Considering slow convergence speed and computation complex offull-band adaptive beamforming, a subband idea with the oversampled Cosine-modulatedfilter banks is applied into the full-band beamforming. A subband adaptive beamformingbased microphone array speech enhancement method is thus presented in the paper. Thenormalized leaky LMS algorithm is used in blocking filter and a subband mean coherence function is introduced to control the subband adaptive processing, all of which can reduce thespeech signal cancellation brought by reverberation. A modified postfilter is adopted tosuppress residual noise. Simulations results show that the proposed method achieves fastconvergence speed and a good noise suppression performance.(3) Study on near-field beamforming based microphone array speech enhancementmethod. With the advantage of spherical nature of sound wavefront, a microphone arrayspeech enhancement method based on near-field beamforming is presented in the paper.Distance discrimination used here reduces the bad effect of both reverberation and noise onthe speech signal. Simulations results demonstrate that the proposed method exhibits a betternoise reduction performance than other methods.(4) Study on blind beamforming based microphone array speech enhancement method.Considering the effect of steering vector errors on performance of speech enhancement basedon conventional beamforming, a blind beamforming based microphone array speechenhancement method is proposed in the paper. It does not need any prior knowledge ofdirection of arrival (DOA) and array manifold. Simulations results show that the proposedblind beamforming method exhibits good noise suppression performance even with theunknown prior knowledge such as DOA.(5) Study on the relationship between microphone array spatial placement and theperformance of array speech enhancement. Criterions including array beampattern response,array gain and Cramer-Rao bound of time delay estimation are derived to evaluate theperformance of microphone array system for a given array configuration. Experimental resultsshow that spatial placement of microphones affects the performance of speech enhancementsystem, and optimal microphones placement is able to significantly improve the speechquality.