A segment-based algorithm of speech enhancement for robust speech recognition

Accurate recognition of speech in noisy environment is still an obstacle for wider application of speech recognition technology. Noise reduction, which is aimed at cleaning the corrupted testing signal to match the ideal training conditions, remain to be an effective approach to improving the accuracy of speech recognition in noisy environment. This paper introduces a new algorithm of noise reduction that combines a tree-based segmentation method with the maximum likelihood estimation to accommodate the nonstationarity of speech while efficiently suppressing the possibly nonstationary noise. Numerical results are obtained from the experiments on an speech recognition system, showing the effectiveness of the proposed algorithm in improving the accuracy of Chinese speech recognition.