MFCC based robust features for English word Recognition

In this paper we propose robust features of MFCC for speech recognition in noisy environments. The state-of art Mel-frequency Cepstral Coefficients technique were extensively used in speech Recognition. The MFCC are derived from speech frames by using Fast Fourier Transform and Discrete Cosine Transform. Performance of MFCC features is degrading in noisy environments. These features further normalized to get new set of features known as Cepstral Mean Normalized (CMN) features. Normalized features improves recognition rate in presence of additive noise. Performance of these noise robust features evaluated on own created English digit database using Vector Quantization technique. The effectiveness of the new feature set is demonstrated by the results of both Speaker-Independent and speaker-dependent recognition tasks in presence of white Gaussian noise.