The methods for combining the information of various kinds of features in speech recognition

摘要

In speech recognition studies based on features, it is often discovered that the recognition performance of some features for some utterances is better than for the others, while the other features have opposite effects. They have, to some extents, the complementary re1ation in the recognition of some utterances. Based on HMM recognition which is now widelyused in speech studies, three efficient methods for combining multiple complementary features to improve the recognition performance of HMM recognition are presented in this paper. They are defined as the maximum parameter method, all parameter method and the most reliable parameter method. By using these three methods, the improvements of the performance in multi-speaker Chinese digit DHMM/VQ recognition from 89% achieve 92.3%, 95.7% and 94.3%respectively.