In this work, spectral features are extracted from speech to perform emotion classification. Linear prediction cepstral coefficients, Mel frequency cepstral coefficients and their derivatives (velocity and acceleration coefficients) are explored as features. Gaussian mixture models are proposed as classifiers. The emotions considered in this study are anger, fear, happiness, neutral, sadness and surprise. The emotional speech database used in this work is both simulated and semi-natural in nature. The semi-natural database has been collected from the dialogues of actors/actresses in popular Hindi movies. Average emotion recognition performance, in the case of male and female speaker is observed to be around 65.3% and 72% respectively. Recognition performance for semi-natural and simulated databases has been compared. © 2012 Springer-Verlag.
CITATION STYLE
Koolagudi, S. G., Barthwal, A., Devliyal, S., & Sreenivasa Rao, K. (2012). Real life emotion classification from speech using gaussian mixture models. In Communications in Computer and Information Science (Vol. 306 CCIS, pp. 250–261). https://doi.org/10.1007/978-3-642-32129-0_28
Mendeley helps you to discover research relevant for your work.