Emotional feature selection of speaker-independent speech based on correlation analysis and Fisher

Abstract

Feature selection is a crucial step in the development of a system for identifying emotions in speech. Recently the interaction between features generated from the same audio source was rarely considered, which may produce redundant features and increase the computational costs. To solve this problem, emotional feature selection of speaker-independent speech based on correlation analysis and Fisher is proposed, which can remove the redundant features that have high correlations with each other. Experiment on the speech emotion recognition based on Support Vector Machines (SVM) is performed, where the speaker-independent features selected by the proposal and the features selected without correlation analysis are used for emotion recognition respectively, and the experimental results show that the proposal achieved 70.2% recognition rate on average. Using speaker-independent features, it would be fast and efficient to discriminate emotional states of different speakers from speech, and it would make it possible to realize the interaction between speaker-independent and computer/robot in the future.