Classification of Emotional Attitudes in Pet-directed Speech

This paper describes the speech interface of a multimodal application developed as part of a study in intelligent multimedia. The application centres on a pet-like autonomous agent “Bouncy” and users communicate with the agent through speech, gestures, and eye contact. The interface concentrates on extra-linguistic properties of the speech signal, pitch and speech rate, involving no actual “speech-totext” recognition. Published studies on perception and synthesis of emotional speech furnish good grounds for believing that pitch and speech rate constitute useful parameters also for revealing basic emotional attitudes of humans when communicating with pets. The emotional attitudes in question are mainly “disapproval” and “approval”. The speech interface is tested in a set-up where Bouncy is placed between a bowl of candy and a bowl of vegetables. Through voice, gestures and eye contact, the test persons must keep Bouncy off the unhealthy candy and encourage him to eat the healthy (but less tasty!) vegetables.