A novel approach for auditory spectrum enhancement to improve speech recognition's robustness

摘要

The success of acoustic features derived from MFCC and PLP turned them into a standard choice in specific conditions but their performances degrade with additive noise. Recently, we have proposed an auditory feature extractor based on gammachirp wavelet transform, they are obtained by replacement of the filter bank used in above methods by a gammachirp wavelet transform. We found that proposed feature give a significant improvement in robust speech recognition than conventional acoustic feature.