Finished projects

This FWO-funded project aims at combining bottom-up phonetic recognition and long span example based recognition into a single speech recognition architecture that beats mainstream state-of-the-art HMM systems in terms of accuracy, be it at a higher computational cost. The project runs in collaboration with KULeuven.

This IWT funded project runs in close collaboration with Aristo Music, a company that is specialized in creating music channels that meet the desires of the customer. In order to accomplish its objective, the company must have access to a very large music database and to annotations that were made for each song in that database. In the past, these annotations were created manually by musical experts, but this method is too time consuming and expensive to scale up to the ever growing database sizes.

Since more and more music consumers have larger and larger music collections, and since buying and downloading (legally) of music from the internet will become more and more the standard way of acquiring music, it is important to develop new search methods for making relevant selections (e.g. Playlists) of a large collection. Obviously, one can store title of the song, name of the composer, etc. but based on advanced signal processing techniques one is presently investigating the development of more natural and flexible search methods. One such technique is query by imitation.

This STEVIN project is coordinated by TNO Utrecht and is about the setting up of a large-scale benchmark for speech recognition in Dutch, and with participation of different research groups in the Netherlands (Nijmegen, Twente, Delft), Flanders (Gent, Leuven) and abroad (TU-Brno and LIMSI)