Efficiently using speaker adaptation data

Transformation based speaker adaptation techniques, such as Maximum Likelihood Linear Regression (MLLR) [1] require a large amount of adaptation data to robustly estimate the transform matrices. In this paper, we present a new adaptation scheme that adjusts the adaptation data according to the feedback from recognizer. By giving different weights to different parts of the adaptation data, the proposed scheme can make use of the adaptation data more efficiently. Experiments on the WSJ 20K task show that this method achieved an additional 10% relative word error rate reduction in supervised adaptation and 2% reduction in unsupervised adaptation compared with conventional MLLR approach.