Adaptation of Prosodic Phrasing Models

There is considerable variation in the prosodic phrasing of speech between different speakers and speech styles. Due to the time and cost of obtaining large quantities of data to train a model for every variation, it is desirable to develop models that can be adapted to new conditions with a limited amount of training data. We describe a technique for adapting HMMbased phrase boundary prediction models which alters a statistical distribution of prosodic phrase lengths. The adapted models show improved prediction performance across different speakers and types of spoken material.