Speech Emotion Analysis: the Production-Perspective

Exploring Acoustic Cues and Spectral Features for Speech Emotion Recognition

Abstract

Automatic recognition of human emotion (anger) inspeech aims at recognizing the underlying emotional state of a speaker from thespeech signal. The area has received rapidly increasing research interest overthe past few years. However, designing powerful spectral features forhigh-performance speech emotion recognition (SER) remains an open challenge.Our purpose was to investigate whether emotion (anger), as perceived by a panelof listeners, were observable in various acoustic cues of the speech signal.The cues were chosen, by examining earlier studies on the same subject and theywere: the syllable rate, the minimum, maximum, median and mean of the pitch,the amplitude and the first six formants, % jitter and %shimmer. Application isfirst trained on person voice will provide signal if detected as angry, and aperson, who is signaled that his/her manner of speaking is classified as angry,becomes aware of his mental state and could regulate his way of expressing histhoughts. It is of great advantage in this situation if the corrective comesfrom a machine that does not play a part in the situation and shows no emotionsin itself.