next up previous contents
Next: Task 4.2: Status Up: Task 4.2: Unsupervised Adaptation Previous: Task 4.2: Unsupervised Adaptation

Task 4.2: Objective

In a continuous stream of broadcast audio, significant sections will not contain speech or will be very poorly matched to the acoustic models. In this task we have investigated techniques to segment such audio streams into ``recognizable'' and ``not recognizable'' segments. We have also looked at vocal track length normalization for unsupervised speaker adaptation.

Christophe Ris