A bio-inspired emotion recognition system under real-life conditions

This paper pulls together the advances of recognizing emotion theory with advances in speech feature in order to improve understand of emotion under real life condition. It presents the application of a recently proposed feature extraction method based on spectral features, used for speech emotion recognition purposes. Specifically, the performance of the proposed approach is evaluated on real condition speech signal (IEMOCAP database) with real world noise using various SNR levels. We examined an assessment of emotion error rate using classical descriptors (MFCC, PLP) and also new type of speech features considered as more robust to noise and reverberation distortions (PNCC with different variants). The results reveal that the used methods give better performance using MFCC under clean environment, and that PNCC shows an advantage compared to other features methods in noisy environment.