Emotion in user interface, voice interaction system
V. Kostov, Shuichi Fukuda · 2002
An approach towards a personalized voice-emotion user interface regardless of the speaker's age, sex or language is presented. An extensive set of carefully chosen utterances provided a speech database for investing acoustic similarities among eight emotional states: (unemotional) neutral, anger, sadness, happiness, disgust, surprised, stressed/troubled and scared. Based on those results, a voice interaction system (VIS) capable of sensing the user's emotional message was developed. In efforts to detect emotions, several primary parameters from human speech were analyzed: pitch, formants, tempo (rhythm) and power of human voice. First the individual basic speaker's voice characteristics were extracted (pitch or/and formants in neutral speech, normal speech rate, neutral speech power) and based on those parameters the emotional message of the subject's utterance was successfully extracted. The VIS interacts with the user while changing its response according to the user's utterances.