Using Smap and Smllr for Arabic Speaker Adaptation

In this paper we address the problem of the adaptation of a speech recognition system to a new speaker. The aim of adaptation is to compensate the mismatch between training and testing condition, but how? The work of Illina and Mostefa (2001) propose to compensate the means of the Gaussian representing the acoustic model. In this paper, we proposes two methods to speaker adaptation of acoustic models in speech recognition systems, we have investigated two techniques which integrate the concepts of both structural maximum a posteriori (SMAP), and structural maximum likelihood linear regression (SMLLR) to adapt the Gaussian means of the speaker independent models. The experiments were evaluated using the Arabic words engine Laboratories "LASA", they show that all of the proposed techniques can improve the performances of an automatic speech recognition system (ASRS) in unsupervised batch adaptation as efficiently. But the approach SMAP, gives the best results