Compensation of Speech Enhancement Distortion with Combination of CMN and PMC

The performance of speech recognition systems degrades severely when speech enhancement technique is applied under low the SNR ratio because of the distortion induced by speech enhancement.This paper presents a Parallel Model Combination (PMC) in model space and Cepstral Mean Normalization (CMN) in characteristic space to compensate the distortion.Resultsare presented that demonstrate the effective combination of robust methods in different space can significantly improve the recognition rate.