Speech emotion recognition method based on hidden factor analysis

A robust speech emotion recognition system relies on a large number of training data, which are difficult to collect in practice. To tackle this problem, a novel speech emotion recognition method based on hidden factor analysis is presented. By utilising the mixture of factor analysers approach, the acoustic features are decomposed into an emotion-independent component and an emotion-specific component. The emotion-specific component, described by a low-dimensional emotion identity vector, is adopted for classification. The proposed approach is evaluated via cross-corpus emotion recognition, and the experimental results demonstrate the efficacy of the proposed method.