Emotion Recognition in Speech Using MFCC and Energy Based Ratio Features

Emotion recognition from speech signal has emerged as a prominent research focus within the realm of human computer interaction in the last decade. Nonetheless, the existing recognition accuracy requires enhancement. To improve the recognition accuracy the proposed framework adopts a global feature extraction approach, decomposing the input speech signal into Intrinsic Mode Functions (IMFs). After that, the Mel Frequency Cepstral Coefficients (MFCC), and energy-based ratio features are extracted from each IMF. The ReliefF algorithm is employed to select pertinent features, which are subsequently tested on an optimizable ensemble classifier for evaluation. The proposed method provides 95.3% emotion recognition accuracy.