语音情感的维度特征提取与识别

Dimensional Feature Extraction and Recognition of Speech Emotion

研究了情绪的维度空间模型与语音声学特征之间的关系以及语音情感的自动识别方法。介绍了基本情绪的维度空间模型,提取了唤醒度和效价度对应的情感特征,采用全局统计特征减小文本差异对情感特征的影响。研究了生气、高兴、悲伤和平静等情感状态的识别,使用高斯混合模型进行4种基本情感的建模,通过实验设定了高斯混合模型的最佳混合度,从而较好地拟合了4种情感在特征空间中的概率分布。实验结果显示,选取的语音特征适合于基本情感类别的识别,高斯混合模型对情感的建模起到了较好的效果,并且验证了二维情绪空间中,效价维度上的情感特征对语音情感识别的重要作用。

The relation between the emotion dimension space and speech features is studied. The automatic speech emotion recognition problem is addressed. A dimensional space model of basic emotions is introduced. Speech emotion features are extracted according to the arousal dimension and the valence dimension. And statistic features are used to reduce the influence of the text variations on emotional features. Anger, happiness, sadness and neutral state are studied. Gaussian mixture model is adopted for modeling and recognizing the four categories of emotions. Gaussian mixture number is optimized through experiment for the probability distribution of the 4 categories in the feature space. The experimental results show that the chosen features are suitable for recognizing basic emotions. The Gaussian mixture model achieves satisfactory classification results. The valence features in the two-dimensional space plays a more important role in emotion recognition.