Exploring emotion specific features for emotion recognition system using PCA approach

In this paper, we address emotion specific features for emotion classification using speech signals using PCA approach. Emotional speech is produced by human speech production mechanism and the emotion information lies in both excitation source (prosody features) and the vocal tract system (spectral features). However, studies have shown that the emotion recognition performance of the system decreases by using either spectral or prosody features individually. We propose emotion specific features by combining both spectral and prosody features for Emotion Recognition. Principal Component Analysis (PCA) is applied on the combination of Mel Frequency Cepstral Coefficients (MFCC), Pitch Chroma (PC), Fundamental Frequency (F0) and Formants Features. The resultant features are used for modeling using Gaussian Mixture Models (GMMs) and testing. It is identified that, the recognition performance has increased significantly for the four emotions viz. angry, fear, happy and neutral. The database used for the study is IITKGP.