Parameter Compensation for Mel-LP based Noisy Speech Recognition

This study deals with a noise robust distributed speech recognizer for real-world applications by deploying feature parameter compensation technique. To realize this objective, Mel-LP based speech analysis has been used in speech coding on the linear frequency scale by applying a first-order all-pass filter instead of a unit delay. To minimize the mismatch between training and test phases, Cepstral Mean Normalization (CMN) and Blind Equalization (BEQ) have been applied to enhance Mel-LP cepstral coefficients as an effort to reduce the effect of additive noise and channel distortion. The performance of the proposed system has been evaluated on Aurora-2 database which is a subset of TIDigits database contaminated by additive noises and channel effects. The baseline performance, that is, for Mel-LPC the average word accuracy for test set A has found to be 59.05%. By applying the CMN and BEQ with the Mel-LP cepstral coefficients, the performance has been improved to 68.02 and 65.65%, respectively.