Noise Robust Novel Approach to Speech Recognition

Abstract

Most practical methods of the speech recognition (SR) are dependent on the feature extraction schemes used in the implementation. The performances of these SR systems are highly affected by the presence of noise. By passing speech signal through cochlear filter bank (CFB) prior to the feature extraction reduces the impact of the noise on the system. In this paper, noise robust approach of feature extraction, cochlear filter bank with zero crossing as a feature is discussed. The comparative analysis of CFB with Mel frequency Cepstral Coefficients (MFCC) approach of feature extraction in terms of recognition accuracy (RA) is discussed. It may be implied that former approach gave a good fit to the experimentation in presence of noise.