Statistical analysis of prosodie parameters and emotion recognition of multilingual speech

The features of prosodie parameters are considered as the direct reflection of emotional information in speech signals. In order to research the feasibility of emotion recognition based on basic prosodie parameters and improve the robust of language-independent emotion recognition system, statistical analysis of pitch, energy and time parameters of multilingual emotional speech is discussed. A corpus of emotional speech spoken by one speaker in Chinese, English, and Japanese is collected. Principle Component Analysis (PCA) method is used to recognize the states of emotion in multilingual speech. The mean error rate of recognition is 27.74% and the lowest error rate is 11%. The statistical analysis shows that language factor doesn't effect pitch variation features of some given emotion obviously. And according to the recognition results we can conclude that basic emotion states in multilingual speech can be recognized by a few simple prosodie parameters.