Research On Space-Time Adaptive Filtering And Implementation Of Distant Speech Enhancement

Acoustic waves,light waved and electromagnetic waves are three kinds of the media carrying information.Among acoustic signal processing,the retrieval and understanding of weak speech is one of the main research points.In the outdoor environment,speech signals from the far field are affected by all kinds of interference and noise,along with geometrical attenuation.In this paper we propose a space-time processing method,focusing on how to reduce noise,suppress interference,enhance speech signal,and guarantee the intelligibility of speech.Facing with the weak signal from the far field,we design and construct a microphone array with large aperture to collect speech signal,which is then enhanced by the beamforming technology based on the plane wave propagation model.Further the MVDR beamformer and the adaptive generalized sidelobe canceller are utilized to suppress interference for speech enhancement.At the same time,a wideband frequency-invariant beamformer is designed to enhance the speech signal while reduce the distortion of speech.Also,outdoor speech faces the affect of wind noise.Research implicates that wind noise has the similar non-stationary characteristics as speech signals have.To eliminate the influence of wind noise,we propose a voice activity detection method based on sequential Gaussian mixture model so as to distinguish the speech and noise in the time-frequency domain.The voice activity detection is then combined with an improved coherent comb filter based on the feature of voiced/unvoiced speech for removal of the wind noise and speech enhancement.Finally,we design and build a system for distant acoustical signal acquisition and processing.Results of outdoor experiments demonstrate that the joint space-time processing method is able to improve the signal-to-noise ratio and the intelligibility of speech signal,so as to lay the foundation for the subsequent speech recognition.