Research On Robust Speech Recognition

Noise robustness is one of the major obstacles to the commercial use of speech recognition techniques. Most of the current speech recognizers are designed to work in controlled environments using clean speech. Though they can achieve rather high recognition rates in these environments, their performance degrades rapidly when noises exist. This phenomenon is mainly caused by the mismatches between training and testing conditions and makes the current speech recognizers unsuitable for many real world applications. In this dissertation, techniques for robust speech recognition are discussed. The main achievements are listed as follows:1. Model compensation techniques for the MFCCs and the LPCCs fail to achieve high performance when SNR is low because these features can't provide good discrimination in such situations. We propose a new model compensation technique for the 0SA-MFCCs. Experimental results show that the proposed technique can efficiently improve the performance of the OSA-MFCC recognizer and outperform the MFCC recognizer with the same compensation scheme.2. Based on the analysis of the influence of noises to the RAS-MFCCs , we propose a new model compensation technique that uses the RAS-MFCCs as front-end. Experimental results show that the proposed technique improves the performance of the RAS-MFCC recognizer efficiently and outperforms the compensated MFCC recognizer.3. A new MFCC based marginalisation approach is proposed. Two independent criteria are proposed to form a joint decision for detecting the unreliable cepstral components. Experimental results show that the proposed approach improves the performance of MFCC recognizer efficiently and gives higher average...