S. Petrik and F. Pernkopf (Austria)
Language Modelling, Data Mining, Classification, Text Alignment, Similarity Measurement
Automatic phonetic reconstruction of medical dictations from non-literal and automatically recognized speech tran scripts leads to closer-to-literal transcripts for training lan guage models of speech recognizers. In this paper, we in troduce an extended alignment method assessing multiple levels of text segmentation and show how open issues like wrong segmentation in the recognized transcript can be re solved. Furthermore, we compare a rule-based text recon struction approach with an automatic classifier, using the multi-level alignment and a stochastic phonetic similarity measure as features. Experiments show better performance for the rule-based system in terms of Recall and Precision, but superiority of the automatic classifier in terms of lan guage model perplexity. The overall increase in precision compared to the simple system in [1] is between 0.7% and 4.7% absolute without loss in recall.
Important Links:
Go Back
