Acoustic-phonetic labels in a Japanese speech database

A large sized Japanese speech database at ATR(JSDB-ATR) is introduced. Thesespeech data are transcribed in multiple ways using acoustic-phonetic symbols for various data access requests and for the convenience of fine acoustic-phonetic analysis. For multiple transcription, three types of categories are considered: linguistic and phonemic categories, acoustic event categories and some alophonic variation categories. To date, about 8500 words respectively uttered by eight professional announcers have been collected with half of them being acoustically-phonetically transcribed. INTRODUCTION Recently, the construction of speech databases has been undertaken in many languages to obtain much knowledge for speech recognition, perception and syntheses[l]-[3]. However, there are few Japanese speech database (JSDB) large enough for various research purposes.ln this paper, a large Japanese speech database that is being built at ATR (JSDB-ATR) is introduced focusing on its multiple acoustic-phonetic transcriptions.