I was the main worker on the YorkTalk speech synthesis system 1992-4. I have been responsible for implementing and testing the timing model used. Some of the main results of this work show that (i) the basic unit of timing of polysyllabic utterances is the syllable, and the syllable in the foot, (ii) foot-initial, medial, and final syllables have to be timed differently, (iii) Latinate and Germanic words are timed differently from one another.
I have also developed a parser which can recognise words of English as Latinate or not, and, if Latinate, can break the word down into its component morphemes using a categorial and feature-based grammar. These steps are important for (i) getting the stress of words of English right (ii) having a dictionary of morphemes rather than of complete words (which is many times bigger). The parser does not accept ill-formed but Latinate-looking words; so that re-vol-ut-ion is acceptable, but re-vol-at-ion is not, since the root -vol- requires suffixes with the theme vowel -u-. This parser was the basis for a USA patent: 5,651,095 Speech Synthesis using word parser with knowledge base having dictionary of morphemes with binding properties and combining rules to identify input word classes .
We're starting work on YorkTalk again, funded by EPSRC. Click here for more information.
The IPOX system (related to YorkTalk)
Phonetics on the Net