Robust Speech Recognition Based On Feature Classification Histogram Equalization

After more than half century research, speech recognition systems have become nealy applicable. In the laboratory, recognition rate of continuous speech with large amount of words is beyond 90%. But in the working environment, the recognition performance degrades significantly due to different human, enrivonment and noise. The problem of robust noise is the environmental mismatch between training and recognition. This mismatch generally lies in the difference of feature parameter probability distribution. Normalization of feature parameters can be realized through parameter transform of feature space and relief the mismatch between training and application to improve recognition performance of the system. In this paper, an improved Histogram Equalization (HEQ) method is proposed for robust speech recognition system, which improved the anti-noise property of the recognition system quite successful. Experiment results show that the proposed method can significantly improve the recognition rate of the system under environment with low signal to noise ratio.