Emotion Recognition with Speech for Call Centres using LPC and Spectral Analysis

Emotion recognition from human speech is a challenge for the researchers. It is mostly considered under ideal acoustic conditions. The performance of such system is degraded while there is existence of environmental mismatches between training and testing phases. For robust speech recognition it requires for reduction of redundancy, variability, and capturing ability of speech signals in noisy environments. Cepstral coefficients are popularly used features and derived from linear predictive coding (LPC). In that case speech signal is assumed to be the output of the all-pole linear filter simulating the vocal tract of a human being. Such recognition systems with LPC-derived cepstrum work well in clean environments, where the performance is carried with these features. The result obtained using this method has a great achievement. In this paper, the overall emotion recognition process has two goals. The first goal is to provide an update record of the available emotional speech data. The number of emotional states, the number of speakers, and the kind of speech are briefly addressed. The second goal is to present the most frequent features used for emotional speech recognition and to assess how the emotion affects them. Two features have been considered for emotion recognition as linear predictive coding (LPC) and spectral analysis.