Transfer Leaming in Automatic Speech Recognition for Serbian

In automatic speech recognition systems the training data used for system development and data expected to be obtained during the practical use of the system do not have to fit each other perfectly, but other similar data may be available. Transfer learning can help to exploit such similar data for training in order to boost the speech recognizer's performance in a certain domain. In this context, the paper presents the first application of transfer learning in speech recognition for the Serbian language. Several methods are proposed, with the goal of optimizing system performance on a specific part of the existing speech database for Serbian. The experimental results evaluated on a test set from the desired domain show significant improvement in both word error rate and character error rate.