一种基于GMM模型的语音情感识别方法

在人机语音交互系统中，机器不仅要具有理解人类语音的能力，还应当具有识别说话人情感的能力。本文提出了基于高斯混合模型(GMM)的序列分类和识别的改进方法，并将该方法引入到语音情感识别的研究中。本文提出了观测值次序均衡的方法。实验结果证明这种新的方法有效地提高了语音情感识别的准确率。
关键词：语音序列分类；语音情感识别；GMM评分函数

EMOTION CLASSIFICATION AND RECOGNITION OF SPEECH  USING GMM-BASED METHOD HUANG Feng, YIN Jun-xun (School of Elec.&Info. Engg., South China Univ. of Technology, Guangzhou 510640,China) Abstract: For human-machine speech interaction system, machine should not only have the ability to understand contents of human speech, but should also be able to recognize human emotion from input speech to perform properly. In this paper we improve the sequence classification and recognition method based on GMM scoring and score space mapping, and use the new method for speech emotion classification. We propose order compensation method for improvement. Experiment results show the new method significantly improves accuracy of emotion classification and recognition in speech.
Key words: Speech sequential classification, emotion recognition, GMM scoring