Study On Microphone Array Speech Enhancement Algorithm Based On Delay-sum Beamforming

In the speech communication application such as information freeway, multimedia technology, office automation, modern communication and intelligent system and so on, speech signals are usually inevitably affected by noise, reverberation and even other voice interference. These interferences will not only cause the degradation of speech quality, but also lead to a sharp deterioration performance of the speech processing system. The speech enhancement technique, as a pre-processing program, is an effective way to suppress the interferences. Therefore, the study in which speech enhancement algorithms to get as clean as possible speech from noisy speech, has an important significance.This thesis describes and compares several classic single channel and multi-channel speech enhancement algorithms in detail. Focuses on the microphone array speech enhancement algorithms, this thesis introduces an improved algorithm. The delay-sum beamforming algorithm is good in eliminating coherent noise, but it is of no effect in incoherent noise and musical noise, thus the improved algorithm mainly includes three modules: delay-sum beamforming, based on absence or no absence of speech short-time logarithmic amplitude minimum mean-square error estimate (LSA-MMSE), postfiltering. Whereas the shortcoming of the delay-sum beamforming, it obtains the better effect to make the speech pass the module of the improved LSA-MMSE to eliminate the noise of background, however, there is a small part of the non-coherent noise exists. The thesis adds a post-wiener filter aiming at this shortage, the simulations of the improved algorithm with matlab are given, experiments indicate that the improved algorithm surely obtain better stabilization and capability of eliminating noise. It can be shown from the simulation result that: compared with traditional delay-sum beamforming, the improved microphone array speech enhancement algorithm surely achieves better robustness, and has better output signal-to-noise ratio.