Being able to process and express emotional signals when interacting with humans is an important feature for machines acting in roles like companions or tutors. Laughter is a very important signal that regulates human conversations. It is however hard for machines, which usually have no real comprehension of the phenomenon that triggered laughter, to understand the meaning of human laughs and, in consequence, to react accordingly. In this paper, we explore one dimension to characterize laughs: their intensity. Without better understanding the conversation, a machine that can infer the intensity of users' laughs will be better equipped to select an appropriate answer (which can be laughing at an intensity related to the detected laugh).