Natural language processing in a Japanese text-to-speech system for written-style texts

This paper introduces a Japanese text-to-speech (TTS) system, AUDIOTEX, that modifies written-style expressions for listening and generates prosodic information (Kanji reading, accent, and pause). Conventional TTS systems have several problems: most texts are written to be read with the human eye, thus they are inadequate for listening; and Kanji reading errors and accentuation errors reduce intelligibility. To solve these problems, we have developed a rewriting process for listening based on flexible tree conversion rules and a highly accurate morphological analysis which retains word candidates with an error correction environment