The ESCA Workshop on Speech Synthesis
September 25-28, 1990
In this paper I discuss some computational-linguistic reasons why conventional text-to-speech systems which employ segmental representations and phonological rewrite-rules are less than satisfactory. I describe the YorkTalk system, which implements the "no-segment, no-rewrites" hypothesis to generate phonetic parameters for the Klatt formant synthesizer. The resulting synthetic speech is fluent, articulate and very human-like.
Bibliographic reference. Coleman, John (1990): "Yorktalk: "synthesis-by-rule" without segments or rules", In SSW1-1990, 13-16.