X. Luo, I.Y. Soon, C.K. Yeo, and S.N. Koh (Singapore)
Automatic Speech Recognition, Auditory Modeling, Lateral Inhibition, AURORA2
Performance of automatic speech recognition system drops dramatically in the presence of background noise unlike the human auditory system which is more adept at noisy speech recognition. Auditory modeling, which simulates some properties of the human auditory system, can thus be applied to speech recognition system to enhance its robustness. This paper proposes a highly effective and extremely simple noise robust front-end based on the traditional mel-frequency cepstral coefficients (MFCC) feature extraction algorithm. The proposed method is a novel integration of lateral inhibition, a simultaneous masking technique to MFCC, which sharpens signals’ power spectrum in the frequency domain and inherently removes noise. Experiments carried out on AURORA2 database show that the word recognition rate using our proposed feature extraction method has been effectively increased. Another major merit of our approach is the extremely low computational load involved compared to other known methods.
Important Links:
Go Back
