A method combining the global and time series structure features for emotion recognition in speech

Abstract

Combined the global and the vowel structure features, a novel method of utilizing modified quadratic discriminant function (MQDF) on emotional features extraction in speech signals is proposed in this paper. After 3000 emotional sentences involved such four emotions as happiness, anger, surprise and sadness were collected, the new method was used to analysis the features of the time, amplitude, pitch and formant construction. The experimental results show that the presented method in our paper can provide an effective performance in emotional recognition.