Robust feature extraction for speech recognition based on perceptually motivated MUSIC

A novel feature extraction algorithm was proposed aiming at improving speech recognition rate in noise environmental conditions. Core technology was the Multiple Signal Classification (MUSIC), which estimated MUSIC spectrum from the speech signal and incorporated perceptual information directly into the spectrum estimation, then the cepstrum coefficients were extracted as the feature parameter. We evaluated the technique using improved Hidden Markov Model (HMM) in different noisy environment, six Chinese vowels were taken as the experimental data. The experimental results show that the novel feature has very good robustness and effectiveness relative to the previously proposed Mel Frequency Cepstral Coefficient (MFCC) technique and the improved HMM can make speech recognition system robust in noise environmental conditions.