Aslı Özyürek

Decomposing co-speech gesture generation into separate semantic and beat streams, controlled by a learned gating mechanism, yields more realistic and expressive human-computer interaction.