Spectral Features for Emotion Classification

This paper aims at exploring short term spectral features for Emotion Recognition (ER). Linear predictive cepstral coefficients (LPCC), mel frequency cepstral coefficients (MFCC) and log frequency power co-efficients (LFPC) are explored for classification of emotions. For capturing the emotion specific knowledge from the above short-term speech features vector quantizer (VQ) models are used in this paper. Indian Institute of Technology, Kharagpur-Simulated Emotion Speech Corpus (IITKGP-SESC) is used for developing the emotion specific models and validating the models by emotion recognition task. The emotions considered for the study are anger, compassion, disgust, fear, happy, neutral, sarcastic and surprise. The recognition performance of the developed models is observed to be about 40%, where as the subjective listening tests show the performance about 60%.