ITRW on Speech and Emotion September 5-7, 2000 Newcastle, Northern Ireland, UK

Analysis of an emotional speech corpus in Hebrew based on objective criteria

Noam Amir (1), Samuel Ron (2), Nathaniel Laor (2,3)

(1) Holon Academic Institute of Technology, Holon, Israel
(2) Tel Aviv-Brull Community Mental Health Center, Tel-Aviv, Israel
(3)Sackler Faculty of Medicine, Tel-Aviv University, Tel-Aviv, Israel and Yale Child Study Center, New Haven, CT, USA

To build a corpus of emotion in spontaneous speech, subjects were asked to recall an emotional event. Physiologic evaluations were then applied to validate the emotional content. This evaluation consisted of measuring several physiological parameters. Forty subjects were examined, each expressing five basic emotions (anger, fear, joy, sadness and disgust). Subjects who responded within the set criterion of the physiological evaluation were considered to represent the said emotion. This was followed by computer analysis in order to determine a set of criteria that could represent each emotion. The method employed was based on the analysis of the signal over sliding windows and extracting a representative feature set. At each instantce, the distance of the measured feature set from a set of reference points is calculated, and used to compute a fuzzy membership index for each emotion, which distance we called "emotional index". We present initial results of this analysis: the characteristics of various emotions are presented, and we discuss the extent to which they are common across different individuals. Our initial results show that when tested on a Hebrew emotional speech corpus the proposed method yields reliable results.

Full Paper

Bibliographic reference.  Amir, Noam / Ron, Samuel / Laor, Nathaniel (2000): "Analysis of an emotional speech corpus in Hebrew based on objective criteria", In SpeechEmotion-2000, 29-33.