模糊矢量量化在语音情感识别中的应用

Improved Fuzzy VQ Algorithm in Speech Emotion Recognition

提出了一种将改进的模糊C均值聚类算法与矢量量化相结合的语音情感识别方法,实现了对4种情感的识别:高兴、生气、悲伤和惊奇。首先提取情感语句全局结构和时序结构特征参数并进行性别规整,再利用改进后的模糊矢量量化方法来设计码本,最后对待识别语音进行辩识。该算法不但解决了模糊C均值算法对初始值敏感、易陷入局部最优的问题,而且性别规整改善了特征参数的有效性,使识别率得以进一步提高。实验结果表明该算法能够有效改善识别率。

A method which combines improved fuzzy c-mean clustering and VQ（Veetor Quantization） is proposed. Four emotions, namely happiness, angry, sadness and surprise, are recognized. Firstly, globe and time sequence features are extracted from speech signals, and modified according to the gender difference. Then code book is designed by improved fuzzy VQ. Finally the emotion of the speech is recognized. The problem of sensitive to initial condition is settled, and the local optimization is also avoided. In addition, the features validity is improved by gender modification. The result shows the better recognition rate.