[en] In this paper, the AVLaughterCycle database is presented. The database consists in the audio and facial motion capture recordings of 24 subjects watching a funny video. It is the first database of laughter combining these 2 modalities. There are around 1000 laughter episodes in the database, covering a large variety of shapes. The database annotation protocol is meant to not only build a large laughter class grouping all the utterances, but distinguish different kinds of laughter. The database is publicly available for research purposes. It can be used for many laughter studies. For example, in the AVLaughterCycle project it was used to endow a virtual agent with the capability of appropriately joining its conversational partner's laughter.
