A study on new strategy of emotion recognition in speech

本研究提出了一种语音中的情感识别方法，该方法不仅可以估计语音中的情感本身，而且可以估计包含多个情感的语音中每种情感的程度。这种方法将每种情绪表示为基本因素向量的合成向量，即唤醒-价位-优势。将该方法与我们提出的情感感知模型应用于语音中的情感识别，结果表明，语音到情感空间的映射最符合人的反应。此外，与GMM相比，该方法的识别率也有很大提高。

This study proposed a method of emotion recognition in speech, which can estimate not only the emotion itself but the degree of each emotion from speech that plural emotions are included in. This method represents each emotion as a resultant vector of the basic factor vectors, Arousal-Valence-Dominance. As the results of applying this method with our already proposed emotion perception model to emotion recognition in speech, the mapping of speech to the emotional space is the most correspondent to human responses. In addition, the recognition accuracy is also greatly excellent at the recognition rate compared with that by GMM.