In this paper, feature extraction for speech recognition is discussed. For that purpose, different normalization methods are evaluated using a small-size DTW-based recognition tool. To minimize the word error rate, several parameters of the feature extraction as well as the parameters of the DTW procedure are optimized. The evaluation shows that optimizing the DTW and other processing steps such as the limiter is successful regarding improvements of the recognition rate. Furthermore, the evaluation shows that the introduced WCMN achieves better results than the CMN, but is outperformed by the CVN.
展开▼