 |
ITRW on
|
 |
|
ITRW on
|
|
In this paper we address the problem of speaker adaptation in noisy environments. We aim at estimating speaker adapted models from noisy data by combining unsupervised speaker adaptation with model-based noise compensation. Speaker adapted models obtained with this method should contain as little information about the environment as possible, so that they can be reused in different environments. We propose a method based on linear approximations of the cepstral operator that allows for the computation of environment independent speaker adapted models from environment dependent models. The method is used in conjunction with MLLR speaker adaptation and jacobian model compensation. Results for a 2000 word task on real car noise show that unsupervised speaker adaptation combined with noise compensation provides an error rate reduction of 20% to 30% compared to noise compensation alone.
Bibliographic reference. Rigazio, Luca / Kryze, David / Nguyen, Patrick / Junqua, Jean-Claude (2001): "Joint environment and speaker adaptation", In Adaptation-2001, 93-96.