Speech Enhancement Based On MMSE-STSA Estimation And Residual Noise Reduction
Zheng Wentao, Zhigang Cao · 2005
Three types of noisy speech enhancement algorithms based on Minimum Mean-Square Error Short-Time Spectral Amplitude estimation (MMSE-STSA) are experimentally compared, especially under low SNR. Among them, MMSE-STSA and MMSE-LOG-STSA have been reported [1],[2] and MMSE Short-Time Relative Spectral Amplitude (MMSE-REL-STSA) estimation is derived in this paper. The modified gain function and center clipping are proposed to reduce 'residual noise'. It is shown that these algorithms can improve the speech SNR of 5- -10dB by 3.4- 12dB, and provide enhanced speech with colorless noise.