In this paper, we present a speech database with children's speech. We first describe the database, part of the emotional labelling conducted, and the distribution of 'non-linguistic'/para-linguistic phenomena, esp. of 'real' laughter vs. speech laughter, i.e. laughter modulated onto speech. Especially, we will characterize interactional attitude and speaker-specific behaviour. In a second, more technologically oriented part, we will address the automatic recognition of these phenomena.