Robust Speech Recognition Using a Cepstral Minimum-Mean-Square-Error-Motivated Noise Suppressor

• Published in: IEEE transactions on audio, speech, and language processing Vol. 16; no. 5; pp. 1061 - 1070
• Main Authors: Dong Yu, Li Deng, Droppo, J., Jian Wu, Yifan Gong, Acero, A.
• Format: Journal Article
• Language: English
• Published: Piscataway, NJ IEEE 01.07.2008; Institute of Electrical and Electronics Engineers
• Subjects: Algorithms; Applied sciences; Cepstral analysis; Channels; Detection, estimation, filtering, equalization, prediction; Discrete Fourier transforms; Error analysis; Errors; Exact sciences and technology; Filter bank; Information, signal and communications theory; Mel frequency cepstral coefficient; Mel-frequency cepstral coefficient (MFCC); minimum-mean-square-error (MMSE) estimate; Miscellaneous; Noise; Noise level; Noise reduction; Noise robustness; phase asynchrony; robust automatic speech recognition (ASR); Signal and communications theory; Signal processing; Signal, noise; Spectra; Speech processing; Speech recognition; Statistics; Suppressors; Telecommunications and information theory; minimum-mean-square-error (MMSE) estimate; Noise reduction; Error rate; Discrete Fourier transformation; robust automatic speech recognition (ASR); phase asynchrony; Baseline; Algorithm; Mismatching; Optimization; Mean square error; Cepstral analysis; Mel-frequency cepstral coefficient (MFCC); Speech recognition; Signal processing; Noise immunity; Filter bank; Automatic recognition; Speech processing; Piecewise linearization
• ISSN: 1558-7916; 1558-7924
• DOI: 10.1109/TASL.2008.921761

We present an efficient and effective nonlinear feature-domain noise suppression algorithm, motivated by the minimum-mean-square-error (MMSE) optimization criterion, for noise-robust speech recognition. Distinguishing from the log-MMSE spectral amplitude noise suppressor proposed by Ephraim and Malah (E&M), our new algorithm is aimed to minimize the error expressed explicitly for the Mel-frequency cepstra instead of discrete Fourier transform (DFT) spectra, and it operates on the Mel-frequency filter bank's output. As a consequence, the statistics used to estimate the suppression factor become vastly different from those used in the E&M log-MMSE suppressor. Our algorithm is significantly more efficient than the E&M's log-MMSE suppressor since the number of the channels in the Mel-frequency filter bank is much smaller (23 in our case) than the number of bins (256) in DFT. We have conducted extensive speech recognition experiments on the standard Aurora-3 task. The experimental results demonstrate a reduction of the recognition word error rate by 48% over the standard ICSLP02 baseline, 26% over the cepstral mean normalization baseline, and 13% over the popular E&M's log-MMSE noise suppressor. The experiments also show that our new algorithm performs slightly better than the ETSI advanced front end (AFE) on the well-matched and mid-mismatched settings, and has 8% and 10% fewer errors than our earlier SPLICE (stereo-based piecewise linear compensation for environments) system on these settings, respectively.