The use of discriminative training techniques in speaker adaptationInvestigator: Matt Gibson Supervisor: Thomas Hain
Discriminative training techniques (e.g. maximum mutual information and minimum phone error) have been shown to outperform maximum likelihood training for the task of large vocabulary speech recognition.
The aim of this project is to investigate the effectiveness of discriminative training techniques when applied to the speaker adaptation task. Of particular interest is the unsupervised adaptation of large vocabulary speech recognition systems.
This research project is funded by the EPSRC.