Temporal Modulation Normalization for Robust Speech Feature Extraction and Recognition

Traditional noise reduction methods usually are based on the assumption that the short-term statistical distributions of speech and noise are different. Differently from that assumption, we have proposed a noise reduction method based on the assumption that the temporal modulations of noise and speech are different. Two steps are used in the proposed algorithm: one is the temporal modulation contrast normalization, another is the modulation events preserved smoothing. Since our proposed method can be used independently for noise reduction, it can be combined with the traditional noise reduction methods to further reduce the noise effect. We tested our proposed method as a front-end for robust speech recognition. Two advanced noise reduction methods, ETSI advanced front-end (AFE) method, and particle filtering (PF) with minimum mean square error (MMSE) estimation method, for comparison and combinations. Experimental results showed that our proposed method outperforms the advanced methods as an independent front-end processor, and further improved the performance consistently than using each method independently as combined front-ends.