Linguistic processing in the Kth multi-lingual text-to-speech system

Our ambition has been to deal comprehensively with the text-to-speech conversion process. We want to be able to do this for any kind of text and for several languages in a way that makes the developed algorithms easy to implement in a portable state-of-the-art technology. Linguistic knowledge has been applied in a variety of ways in the formulation of the system. It is mainly based on rules that are formulated in a special higher level programming language. This notation parallels closely the notation used in generative phonology. As a base for development of both grapheme-to-phoneme rules and dictionaries, test corpora of approximately 10,000 words for several languages have been phonetically transcribed. Several different studies based on these corpora will be presented in a separate paper at this conference. The system has been specially expanded to include symbol-to-speech conversion (Blissymbols). The pronunciation of the word(s) corresponding to each concept is stored in a lexicon with its part of speech and features defining inflection category. A grammar predicts clause and phrase boundaries and word forms, and a grammatically well-formed sentence results.