Video: Sneeze-sensing software gives avatars a good laugh
Having a laugh with friends? Your computer may soon be able to join in. Software that can automatically recognise 鈥渘on-linguistic鈥� sounds, such as laughter, and generate an appropriate facial animation sequence, could improve the quality of web-based avatars or computer-animated movies.
Animated characters are already 鈥渓earning鈥� to lip sync when played human speech. But this is only part of the picture 鈥� we laugh, cry, yawn and sneeze our way through life, and realistic computer animations must be able to mimic the facial expressions that accompany these sounds too.
at the University of Bath, UK, and at the University of Cardiff, UK, have developed software to automatically recognise some of these vocalisations and generate appropriate animation sequences.
Advertisement
Sobs, sneezes and yawns
Cosker and Holt used optical motion capture to record the facial expressions of four participants as they performed a number of laughs, sobs, sneezes and yawns. The researchers also recorded the participant鈥檚 voices during their performances.
The researchers then developed software that correlates the key audio features of each non-linguistic vocalisation with the relevant facial motion-capture data.
This motion capture data was used to animate a standardised facial model, with help from at the University of Surrey, UK.
The result is a software model which, when played a new laugh or cry, can automatically animate an avatar in an appropriate manner.
鈥楽tandard鈥� laugh
鈥淧roviding a person laughs with this standard structure, the computer can take their voice and create an animation sequence,鈥� says Cosker.
But there are still limitations with the technique, he says. A loud guffaw sounds different to a snigger, and the software cannot yet get to grips with this level of variation.
鈥淭here is also some ambiguity in the audio,鈥� says Cosker. 鈥淥ne person鈥檚 laugh can sound similar to another person鈥檚 crying. In terms of classifying actions on the basis of audio alone, we still need to do more work.鈥�
Cosker and Holt presented their work at the held in Porto, Portugal.
