Emotional speaker recognition based on i-vector space model

Abstract

I-vector space feature has been recently proved to be very efficient in speaker recognition field. In this paper, we assess using the i-vector approach for emotional speaker recognition in order to boost the performance which is deteriorated by emotions. The key idea of the i-vector algorithm is to represent each speaker by a fixed length and low dimensional feature vector. The concatenation of these speaker dependent i-vector features is used as an input characteristic vectors in the Support Vector Machines (SVM) classifier. In the feature extraction step, the Mel Frequency Cepstral Coefficients (MFCC) were used. All experiments were on spontaneous emotional context using the IEMOCAP data base. Results reveal that the i-vector solve the problem of large scale of SVM model and give a promising results for speaker recognition in spontaneous emotional context.