NORISC : Next generation template based Recognition for Interactive man-machine Speech Communication

It is studied in this research program how a state-of-the-art approach to ASR, can be adapted, or even be replaced by an approach in which some of the implementation choices are each made better suited to describe the long-span speech processes of pronunciation variation. This research program focuses on three different domains: (1) the choice of acoustic features, (2) the choice of the recognition units and the attending challenge for the dynamic programming search, and (3) the choice of how to define new distance measures. In this manner, the research program aims at developing a new approach to acoustic decoding in ASR, a dvancing the state-of-the-art in HMMs, while keeping its advantages.

Project leader: Prof. dr. L.W.J. Boves (RU)