On Endpoint Detection in Isolated Word Recognition

Abstract

The paper deals with the use of dynamic programming for word endpoint detection in isolated word recognition. Endpoint detection is based on likelihood maximization. Expectation maximization approach is used to deal with the problem of unknown parameters. Speech signal and background noise energy is used as features for making decision. Performance of the proposed approach was evaluated using isolated Lithuanian words speech corpus. Performance evaluation based on recognition error rate showed that for low noise level endpoint detection based on dynamic programming slightly outperforms threshold based endpoint detection. For high background noise level preference of dynamic programming based algorithm is more noticeable. The main advantage of dynamic programming based approach is that this method does not need any threshold. This endpoint detection method is applied in development of isolated word recognition system. Ill. 1, bibl. 9 (in Lithuanian; summaries in Lithuanian, English and Russian).