Gender Dependent and Independent Emotion Recognition System for Telugu Speeches Using Gaussian Mixture Models

Abstract: Speech based emotion recognition has its utility in the real world applications. Gaussian Mixture Model (GMM) is used to achieve it. This model is used to train dataset required for the task and suitable for recognizing speech based emotions. GMM makes use of Exception Maximization (EM) algorithm for estimation maximum likelihood of parameters. Based on the GMM model, conditional probabilities are computed for data points which are not known priori. To extract emotional features from the speech Mel Frequency Cepstral Coefficients (MFCCs) method is used. The parameters used for this include vocal tract functions, spectra and pitch formants. In this paper we build a prototype application to demonstrate the concept of speech based emotion recognition. The empirical results revealed that the proposed application is effective.

Keywords: Emotion recognition, GMM, MFCCs, EM