Speech Emotion Recognition Based On A Semi-supervised Learning Research

Today, the rapid development of IT makes the relationship between human and computer have become closer, and intelligent human-computer interaction has become a hot spotin the field of artificial intelligence. As one of the key technologies of intelligent human-computer interaction, emotion recognition plays an important role, which its study is also in full swing.This paper introduces the background and applications for speech emotion recognition first, and then summarizes the key technologies of speech emotion recognition status and the problems encountered, reviews the research progress and research results of speech emotion recognition.On the basis of the reference to the existing emotional voice recording, we established emotional speech database, and the voice signal has been carried out a series of pre-processing operations, and voice emotional characteristics analysis and extraction was determined for this study,11kinds of emotional features.Speech emotion recognition, in essence, belongs to the classification problem of pattern recognition. When we use the conventional supervised machine learning approach to sentiment classification, we need to build statistical model of the concept to be marked on the training set, so that the classifier generated by the experiment has good generalization performance. But marking label of the training samples artificially is a very time-consuming and laborious, especially in corpus of speech emotion recognition.Semi-supervised learning makes use of the information implicit in the unlabeled samples to improve the generalization capability of the classifier, which has been widely used in many fields. It can effectively reduce the manual workload necessary in the production of speech emotion recognition corpus. So that, the semi-supervised learning methods will be used for the voice emotion recognition. Conditional random is regarded as the initial classifier, marking classified against the few number of labelled samples, and then use self-training algorithm, to label the unlabeled samples in order to add them to the training set. So that, the final speech emotion recognition classifier can be obtained by iterations of the loop. In the paper, this method has been used in speech emotion recognition, and compared with the speech recognition method based on a single classifier. The experimental results show that this method has a better overall recognition performance with effectively improving the speech emotion recognition rate.