Project 7: Integration of multiple units in computational models
One of the challenges of making use of FPD in ASR is how to incorporate long-term structures which represent speech dynamics and suprasegmental processes into existing recognizers which employ short-term representations. The frame-based nature of most current work in ASR is at odds with the richer, multiple tree-based representations implied by approaches such as Polysp. Indeed, in ASR, suprasegmental information is typically seen as the source of distracting variation rather than as valuable information. The purpose of this project is to attempt to develop an effective, statistical framework for ASR which is capable of exploiting the information available at multiple time scales. The study has two components. In the first, researchers will build on existing work at Naples and Nijmegen into novel speech feature representations and ASR architectures. The second study is equally adventurous and will examine multimodal FPD.
Working on this project: » Dr Francesco Cutugno » Dr Jon Barker » Dr Louis ten Bosch » Dr Anna Corazza » Bogdan Ludusan » Dr Gianpaolo Coro » Dr Jonas Beskow » Prof Rolf Carlson » Prof David House » Prof Björn Granström
