The Signal Analysis and Interpretation Laboratory (SAIL)
has attempted construction of a ‘Synthetic Laughter Generator’ – which
could be of use as a responder to their ‘Automatic Sarcasm Recognizer’
previously described.
The lab published a paper on the subject of Automatic acoustic synthesis of human-like laughter in the Journal of the Acoustical Society of America,
121:1(527-535) 2007. It points out that “… natural expressive speech
quality is essential for synthesizing long exchanges of human-machine
dialogs and for information relaying monologs.” Something which a
synthetic laughter generator may be able to enhance – say, for example
in an interaction with a robotic agent at a call-centre.
Using a computerised spring-damper paradigm which mimics the
oscillatory behaviour of real laughter, the lab managed to generate
bouts of synthetic laughs – but questions remain as to their
acceptability for real-world use : “It is [also] difficult to define a
quantitative measure or a quantitative set of parameters to define an acceptable
form of laughter.” explain the researchers.
Sadly, the lab’s web-page
on their synthetic laughter project has, for some reason, disappeared –
but it has been partially archived by the Wayback Machine (though, disappointingly, the audio samples themselves have gone )
Note: The work reported in this paper was supported in part by grants from the NSF and the U.S. Army.
