A Study of Zero-Crossings with Peak-Amplitudes in Speech Emotion Classification

A number of recent studies have focused on the speech emotion recognition. Emotional features extraction is the key of the whole system. Most of previous work on feature extraction is database dependent. This paper describes the zero-crossings with peak-amplitudes (ZCPA) feature extraction approach, and applies ZCPA approach to emotion recognition. Two kinds of experiments were carried out. The first kind is one language experiments with TYUT database and Berlin database. The results of this kind experiments show that ZCPA features can represent speech emotion effectively. The second kind experiment is cross-database of three languages. Experiments of the second kind indicate that the ZCPA features are database independent.