Speaker adaptations in sparse training data for improved speaker verification

摘要

The over-training problem in speaker verification occurs when modelling a speaker with sparse training data. The authors propose to solve this problem by employing effective speaker adaptations using a hybrid version of the maximum a posteriori (MAP) and maximum likelihood linear regression (MLLR) methods. Experimental results show that the speaker verification system using the proposed hybrid adaptation scheme outperforms systems based on speaker models without adaptation by a factor of up to 5.