Application of HMM Adaptation and Robust Features to Sub-Band Speech Recognition in Noise

In recent years, sub-band speech recognition has been found useful in robust speech recognition, especially for speech signals contaminated by band-limited noise. In sub-band speech recognition, full band speech is divided into several frequency sub-bands. Sub-band feature vectors or their generated likelihoods by corresponding sub-band recognizers are combined to give the result of recognition task. In this paper, we concatenate sub-band feature vectors, where we extract phase autocorrelation (PAC) MFCC and one type of group delay based MFCC, called MFPSCC, as noise robust features from each subband. Furthermore, we used a model adaptation method, named weighted projection measure (WPM), to adapt HMM Gaussian mean vectors to concatenated sub-band feature vectors in noisy conditions. The experimental results indicate that the proposed methods significantly improve the sub-band speech recognition system performance in presence of additive noise