Analysis of intonation, prosody etc.

Speech and Multimedia Communication Based on the pitch-extraction algorithms used for speech, an experimental system for recognition of hummed tunes has been developed by 6th semester students in a project with focus on pattern matching. The results reported in the conference paper [Brøndsted, Augustensen, Fisker, Hansen, Klitgaard, Nielsen, Rasmussen, 2001] shows that many problems encountered in speech-like "creaky voice" hardly ever occur in hummed sound signals. Likewise, no typical halving/doubling errors were encountered in the system. Testing with seven persons humming 46 songs in total, shows a successful recognition rate of 75.6 % with a search space of 39925 songs. (Tom Brøndsted)