Second International Conference on Spoken Language Processing (ICSLP'92)

Banff, Alberta, Canada
October 13-16, 1992

J-SUMMIT: A Japanese Segment-Based Speech Recognition System

Shinsuke Sakai, Michael Phillips

Spoken Language Systems Group, Laboratory for Computer Science, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA

This paper describes the application of the SUMMIT speech recognition system to continuously spoken Japanese speech within a limited domain. Summit is a speech recognition system being developed at MIT which makes explicit use of acoustic-phonetic knowledge, embedded in a segmental framework that can be trained automatically [1], and has been ported to many applications in English [2, 3]. In order to cope with Japanese speech, we prepared a word lexicon with baseform pronunciations and morphological categories, wrote a set of phonological rules, and developed a morphological category-pair grammar for Japanese. Preliminary recognition experiments within the VOYAGER domain [2] were performed using data read by a single male speaker. Word and sentence error rates of 7.1% and 41.6%, respectively were obtained.

Full Paper

Bibliographic reference.  Sakai, Shinsuke / Phillips, Michael (1992): "J-SUMMIT: a Japanese segment-based speech recognition system", In ICSLP-1992, 1515-1518.