摘要

Though a great deal of research has been done to recognize emotions automatically from human speech, low recognition rate is still a serious problem. In order to improve recognition performance, we used an improved ranked voting fusion algorithm to combine the decisions from eight hidden Markov model (HMM) classifiers which are based on different feature vectors respectively. On the other hand, in view of the severe influence to emotion recognition precision from the individual differences of acoustic character and gender is a main factor leading to acoustic difference, gender distinction method was adopted. The recognition results show that compared with the isolated HMM classifier, the recognition results of the classifier fusion system is more satisfying. Besides, gender distinction method can also improved recognition rate evidently.

  • 出版日期2010