USING EM-TRAINED STRING-EDIT DISTANCES FOR APPROXIMATE MATCHING OF ACOUSTIC MORPHEMES

Our research concerns spoken language understanding within the domain of automated telecommunication services. In the recent papers we presented a new methodology for training of statistical language models for recognition and understanding of utterances from large corpora of phone sequences obtained as the output of a task-independent ASR-system. The advantage of this strategy compared to the traditional word-based strategy is that we don’t have to manually transcribe large amounts of data in order to extract acoustic morphemes to train the classifier. Since the b aseline strategy suffered high False Rejection Rates caused by findi ng no acoustic morphemes in the test data, we describe in this paper how approximate matching can be incorporated in the Bayes-classifier to reduce FRR. The experiments are evaluated for “How May I Help You?” -task.