ABSTRACT
An automatic text-to-speech (TIS) system is a computer system which can take written text as input, and produce audible and intelligible speech as the output. The internal strategy of the main Linguistic Processor of the Edinburgh University Centre for Speech Technology Research (CSTR) TIS system consists of a number of operations:
( 1) examining each word in the input text and establishing its morphological make-up;
(2) stripping off any identified suffixes for separate treatment; (3) looking up the pronunciation and grammatical category of the
remaining morphological core of the word; ( 4) assigning word-stress; (5) making appropriate morphophonemic adjustments to the pronunci-
ation of the whole re-composed word; ( 6) passing the phonemic specification of the word to a speech synthesis
module able to generate an audible representation of the text concerned. Alternatively, if the word is not morphologically complex, and/ or is not represented in the main dictionary of the TIS system, then the pronunciation is generated by using a set of grapheme-to-phoneme spelling conversion rules. In addition, in both cases, the overall intonation and rhythm of the utterance will be generated on a phrasal basis, partly with the help of a syntactic parsing mechanism.
