Evaluation of modulation spectrum equalization techniques for large vocabulary robust speech recognition

Previous approaches for modulation spectrum equalization were evaluated only for the Aurora 2 small vocabulary task. We further apply these approaches on the Aurora 4 large vocabulary task. In the spectral histogram equalization (SHE) approach, we equalize the histogram of the modulation spectrum for each utterance to a reference histogram obtained from clean training data. In the magnitude ratio equalization (MRE) approach, we equalize the magnitude ratio of lower to higher frequency components on the modulation spectrum to a reference value also obtained from clean training data. Experimental test results indicate significant performance improvements using these approaches when cascaded with cepstral mean and variance normalization (CMVN). Cascading MRE with more advanced feature normalization approaches such as histogram equalization (HEQ) and higher-order cepstral moment normalization (HOCMN) yielded additional performance improvements.