Japanese speech synthesis system in a book reader for the blind

This paper describes a Japanese speech synthesis system in a book reader for the blind and its performance evaluation results. A prototype book reader, which reads Japanese language books, has been developed. The speech synthesis system for this book reader is based on speech synthesis by rule using formant synthesis and diphone like segment compilation method. This system is small and inexpensive. This system has six speech synthesis rules, an accent rule, a pause rule, a devoicing rule, a duration rule, a pitch rule and a voice source rule. Intelligibility test results showed 81% monosyllable intelligibility and 96 % word intelligibility. The speech synthesized from texts such as novels and essays was sufficiently intelligible even at the fastest reading speed (more than ten syllables per second). Moreover, results of book reading tests showed that the 96% of the accents in accentual phrases were inferred correctly by the accent rule.