No laughing matter.

Laughter matters! From an analysis of a very large corpus of naturally-occurring conversational speech we have con-firmed that approximately one in ten utterances contains laughter. From among these laughing utterances, we were able to distinguish four types of laughter according to what each re-vealed about the speaker’s affective state, and we were able to recognise these different types automatically, by use of Hidden Markov Models trained on laugh segments, with a success rate of greater than 75%. The paper also presents a speech synthesis interface that enables the control of such emotional expression for use in an interactive conversation.