Adapting a duration synthesis model to rate children's oral reading prosody.