Automatic Recognition of Emotion in Speech

Northeastern University, Boston MA

Investigators

Abstract

The goal of this project is to automatically classify emotion in speech. In addition to using standard acoustic parameters, this work use will also use syllabic patterns and speech formant contours within syllables. These features, along with pattern recognition techniques, will first be used to produce an optimal separation of presence or absence of selected emotions in single words and short phrases. The software will then be adapted to recognizing and classifying these emotions where they occur in longer speech samples. Experiments will also be conducted in applying the software to "impoverished speech" where the linguistic content has been obscured. If successful, this will be a major contribution to automatic recognition and classification of emotion in speech.