Title :
Spectral emotion profile
Author :
Fewzee, Pouria ; Karray, Fakhri
Author_Institution :
Fac. of Electr. & Comput. Eng., Univ. of Waterloo, Waterloo, ON, Canada
Abstract :
Proposed in this work is the notion of spectral emotion profile. The purpose of spectral emotion profile is to highlight the spectral differences of individuals in expressing emotions, and to make use of those differences towards personalizing recognition of emotions in speech. To define spectral emotion profile, we have taken advantage of the spectral energy distribution as a set of speech features. To demonstrate the idea in a more sensible way, we have made use of two emotional speech datasets. Results of our experimental study based on the proposed notion of spectral emotion profile show how different spectral intervals of individual speakers, as well as those of different genders, vary in the amount to which they contribute to the expression of emotions. It is also shown that using spectral emotion profile results in higher prediction accuracy.
Keywords :
emotion recognition; speaker recognition; spectral analysis; emotion recognition; emotional speech datasets; spectral emotion profile; spectral energy distribution; speech features; speech recognition; Databases; Emotion recognition; Feature extraction; Predictive models; Speech; Speech recognition; Visualization; Emotional Speech; Personalization; Spectral Analysis; Speech Features; Variable Selection;
Conference_Titel :
Electrical and Computer Engineering (CCECE), 2015 IEEE 28th Canadian Conference on
Conference_Location :
Halifax, NS
Print_ISBN :
978-1-4799-5827-6
DOI :
10.1109/CCECE.2015.7129450