STiLL - Speech Technology in Language Learning
May 25-27, 1998
In developing computer-based systems for language learning, it is important to model some of the characteristics of the disfluent speech typical in nonnative speakers. We observed that beginning language learners often pause within words while reading. We also observed that our automatic algorithms for scoring segmental quality of pronunciation were affected by these intra-word pauses (IWPs). In this work we propose a method for modeling IWPs. As a result we are able to produce more robust segmental scores. Our experimental study also suggests that the insertion rate of IWPs could be a good predictor of fluency.
Bibliographic reference. Franco, Horacio / Neumeyer, Leonardo / Bratt, Harry (1998): "Modeling intra-word pauses in pronunciation scoring", In STiLL-1998, 87-90.