search menu icon-carat-right cmu-wordmark

A Novel Approach to Emotion Recognition from Voice

October 2018 Poster
Oren Wright

This poster describes work toward a new emotional speech database and a set of techniques that can capture finer nuances than the current state of the art.

Publisher:

Software Engineering Institute

Abstract

This research explores a novel approach that capitalizes on recent advances in micro-articulometry, the measurement of speech properties at the phoneme level. Expected results are twofold: first, the creation of a novel, open-source, statistically labeled emotional speech database that maps to a continuum of emotions rather than discrete labels, and second, a working end-to-end emotion recognition prototype.