ROBUST ANTI-NOISE MFCC FOR SPEECH RECOGNITION

This paper concerns the problem of automatic speech recognition in noisy environments. The main goal of the proposed work is the definition, implementation and evaluation of a novel noise robust speech signal parameterisation algorithm. The proposed procedure is based on time frequency speech signal representation. To improve the performance of ASR systems, a new method is proposed to extract features capable of operating at a very low signal to noise ration. The basic idea in this paper is to enhance to speech quality from the noise signal during the front end processing of the signal. The new feature extraction method using SMN, CMN and CVN is applied for the regular MFCC algorithm and its recognition accuracy over speech sentences have been evaluated. The hidden Markov model toolkit was used throughout our experiments, which were conducted for various noise types provided by noisex-92 database at different SNRs. Comparison results of the proposed approach with the MFCC and antinoise method are presented in this paper. The proposed method improves the recognition accuracy by 3-5% computed on five SNR levels for different types of noise conditions.