A post-processing of speech for hearing impaired integrate into standard digital audio decoders
Shinichi Hoshino, Itaru Kaneko, Hideaki Kikuchi, Katsuhiko Shirai · 1999
In this paper, we present LOADSTAR, a Mandarin Spoken Dialogue system developed for travel information retrieval, which could provide clients tourism information and plan proper itinerary according to clients ’ favor. In addition to description of the system architecture and individual modules, several significant strategies such as mixed class-based and word-based language model, robust semantic-based hierarchical parsing paradigm, dynamically refreshed tracing of dialog, unified scoring schema integrating multiple information sources, and mixed-initiatives control mechanism are proposed and applied to realize efficient interaction. Evaluation of system performance on 13 dialog include 120 utterances, which contains ample spontaneous speech phenomena is reported. The response accuracy of system achieved 90.9%. 1.