LOLOL: Laugh Out Loud On Laptop

Significant progress in the domains of speechand singingsynthesis has enhanced communicative potential of machines. To make computers more vocally expressive, however, we need a deeper understanding of how nonlinguistic social signals are patterned and perceived. In this paper, we focus on laughter expressions: how a phrase of vocalized notes that we call “laughter” may be modeled and performed to implicate nuanced meaning imbued in the acoustic signal. In designing our model, we emphasize (1) using high-level descriptors as control parameters, (2) enabling real-time performable laughter, and (3) prioritizing expressiveness over realism. We present an interactive system implemented in ChucK that allows users to systematically play with the musical ingredients of laughter. A crowdsourced study on the perception of synthesized laughter showed that our model is capable of generating a range of laughter types, suggesting an exciting potential for expressive laughter synthesis.