Update: The results are in, and you can read them here
It would be fun to have a robot throw its head back and guffaw convulsively. Acousticians like me, however, simply want to synthesise a convincing sound. That鈥檚 a tough task, given how much our laughs and giggles can vary.
For one thing, humans can change the vowel sound of their laughs 鈥� from 鈥渢ee-hee鈥� to 鈥渉o ho ho鈥�, though 鈥渉a ha ha鈥� is the most common (). We also vary the prosody 鈥� the stresses, rhythm and intonation 鈥� within and between laughs, helping us to convey different characteristics such as joy and ridicule (). Shrieks of laughter, in which the vocal folds vibrate to give a distinctive pitch to the sound (see diagram), differ from snorts, grunts and pants, where the vocal folds do not vibrate. These 鈥渦nvoiced鈥� sounds seem less popular with listeners ().
Advertisement
We want you to help us test the best in laughter synthesis in an experiment that pits examples from leading researchers against one another. Each bionic chuckle was produced using a different technique, but the goal is the same: to add laughter to synthesised speech to make it sound more natural.
and colleagues from the University of Saarland in Saarbr眉cken, Germany, for example, model the movements of the vocal tract and air flow. at the Pierre and Marie Curie University in Paris, France, on the other hand, takes an algorithm that turns text into speech and then alters the prosody to try and .
Meanwhile at Deutsche Telekom in Berlin, Germany, uses a technique called linear predictive speech coding to generate individual laughs (鈥渉a鈥�), and a simple algorithm to work out their timing. from the University of Mons, Belgium, mixes and manipulates single laughs drawn from real laughter.
Which technique is most effective? You decide. Our online experiment, found at , is very simple to use. Just audition a set of sounds, and after each one say whether it was a computer or a human laughing. Can you spot the fakes?
Read more: The secrets of laughter
