Learning emotional speech by using Dirichlet Process Mixtures

Our aim in this paper is to illustrate the effectiveness of the Dirichlet Process Mixture (DPM) model for emotional speech class density estimation when the number of Gauss mixture components are unknown. The problem is modeled as a two-class classification problem where the classes are anger and-no-anger. Performance of the algorithm is evaluated on the features extracted from the emotion dataset EMO-DB, it is observed that the prior information inclusion led to increased non-anger recall rate. The introduced feature set performs perceptual analysis in time, spectral and Bark domains based on the Perceptual Evaluation of Audio Quality (PEAQ) model as described by the standard, ITU-R BS.1387-1 which provides a mathematical model resembling the human auditory system. Unlike the existing systems, the proposed feature set learns statistical characteristic of emotional differences hence enables us to represent the statistics of emotional audio with a small number of features.