Abstract:
In the upcoming 21st century, the demand for ``global communication among speakers of different languages'' will become more and more pressing. According to recent survey results, spontaneous speech translation will come into practical use by about 2010--2020. Despite the great expectations of society for efficient speech translation systems, there remain a number of difficult problems related to acoustic and linguistic phenomena yet to be overcome. In particular, the variety of spontaneous speech makes it difficult to simply extend current work done on read speech. For example, speech rate can vary considerably and there are often disfluencies and diversities of pronunciation (with ensuing spectral variations) resulting from different speaker characteristics and situations. This presentation describes research targets for the year 2000 for spontaneous speech translation, emphasizing improvements in speech recognition, prosody processing, synthesis of natural-sounding speech, and system integration, for spoken-language translation between Japanese and other languages such as English, Korean, and German. This research activity is being carried out under international collaborations by the international Consortium for Speech Translation Advanced Research (C-STAR II), which was established in 1994 to begin research activities aimed at an international experiment on multilanguage speech translation planned for 1999.