文件:Multi-Microphones Reverberation Cancellation for Distant Speech Recognition.pdf

In distant speech recognition(DSR), a target signal is disrupted by reverberation and noise effects so that the target signal is reduced and distorted. Recognising distant speech robustly, however, still remains a challenge. The contributions are that a global overview of multi-microphones reverberation cancellation for distant speech recognition is provided, the problem of errors in microphones array processing is analysed, and above all fundamental and recent reverberation cancellation techniques are illustrated, such as multi-input multi-output inverse filtering of room acoustics, spectral subtraction multi-channel linear prediction based on short time Fourier transform representation, the proposed linear-predictive multi-input equalization algorithm, the method using neural network front-ends and beamforming, along with theoretical analysis and experimental results verifying the effectiveness of the various algorithms.