No abstract

No abstract

Abstract. The main objective of this paper is to design a noise-resilient and speaker independent speech recognition system for isolated word recognition. Mel-frequency Cepstral Coefficients (MFCCs) has been used for feature extraction. Noise robust performance of MFCC under mismatched training and testing conditions is enhanced by the application of wavelet based denoising algorithm and also to make MFCCs as robust to variation in vocal track length (VTL) an invariant-integration method is applied. The resultant features are called as enhanced MFCC Invariant-Integration Features (EMFCCIIFs). To accomplish the objective of this paper, classifier called feature-finding neural network (FFNN) is used for the recognition of isolated words. Results are compared with the results obtained by the traditional MFCC features. Through experiments it is observed that under mismatched conditions, the EMFCCIIFs features remains high recognition rate under low Signal-to-noise ratios (SNRs) and their performance are more effective under high SNRs too.

No abstract

No abstract