ABSTRACT
Our motivation is to develop an emotional speech detection system for survivors and rescue robots in the context of disasters (earthquake, fire, building collapse, etc.). With emotional speech recognition ability, a teleoperated rescue robot could detect, analyze, and feed a survivor’s vital signs and emotional states back to the control base outside the disaster area, and it could also be a friend to accompany trapped people while rescuers are trying to reach them. The main novelty is that we adopt Higher Order Spectra (HOS) to analyze nonlinear aspects of affective speech signals, and try to search for embedding dimension and associated features that best show the emotional class structure of our data.
