We present an improved speech enhancement method based on Wiener filtering in the wavelet domain for automatic speech recognition (ASR). The wavelet coefficients that are contaminated by the effects of late reflection and background noise are filtered using a Wiener gain. We optimize the wavelet parameters for speech, background noise and late reflection to achieve a better estimate of the Wiener gain for effective filtering. Wiener gains to compensate for the effects of late reflection and background noise are independently estimated and then combined. Moreover, we introduce the noise profile and reverberation time identification to cope with different noise and reverberant conditions. Experimental results in large vocabulary continuous speech recognition (LVCSR) show that the proposed method outperforms the conventional methods.
展开▼