R. Gretter, G. A. Mian, R. Rinaldo and M. Salmasi
This work describes the linguistic analysis done in a real-time text-to-speech system for unrestricted Italian. The system, which is built up around a linguistic processor, includes a lexical analyzer and a syntactic preanalyzer, and synthesizes speech by joining LPC coded diphones. At a word level, the linguistic analysis carries out the transcription from grapheme to phoneme and assigns stress markers corresponding to primary stress, secondary stress and unstressed syllables; at a sentence level it allows to individuate suprasegmental features, namely Phonological Words and Intonational Groups, which are linguistically relevant in spoken language. As a result, the text is converted into a sequence of phonemes, stress marks and syntactic indicators which are eventually used to control the prosodic features (fundamental frequency, duration and intensity) of the synthesized speech at word, intonational group and sentence level.