|
One of their major research topics is speaker-independent
continuous speech recognition (SI-CSR). A SI-CSR system called T-RecS
has been developed in the ELIS Speech Lab. The important features of
this system are: auditory model front-end, an initial segmentation
stage, a two-level phonetic classifier based on Artificial Neural
Networks, and a heuristic search based decision module. T-RecS was
designed in such a way that it is easily adapted to a new task or
language. Their research in this domain is currently focussed on:
improved phonetic classification at the speech unit level, new models
for speech recognition, integration of higher level knowledge in the
decision module.
|