摘要：Abstract Emotion recognition has become a “must have” for all systems that want to inspire user's confidence and to interact in a friendly and familiar way. This is why speech therapy, especially for young children, can be considered half-blind if it does not take into account the emotional state of the subject. Most research on affective state identification so far has focused on adult subjects, with a good pronunciation. However, little research has been conducted on adapting classical emotion recognition techniques in “narrow areas” such as children speech therapy, where emotions play a key role. In this paper we investigate which emotional states should be recognized by a CBST (Computer Based Speech Therapy System) in order to reduce the gap between classical and computer assisted therapy. A brief literature review is presented, exploring the recent work in the area. New hypothesis are tested using results from a focus group and from an experiment involving children. Appropriate affective states and measure model are identified. These results encourage us to develop an emotion recognition extension for our CBST named Logomon. While the vision of a CBST that can replace human SLT (Speech and Language Therapist) is beyond the horizon of the next decade, recent advances has proved that computer can be a valuable “primary assistant” in the therapy process.
关键词：Emotions models;affective computing;computer based speech therapy;emotion recognition ;
