Speech synthesis

Techniques used to generate synthetic speech

• waveform concatenation
• formant synthesis
• articulatory synthesis

For a recent introduction to the field see Keller 1994.

YorkTalk and IPOX use formant synthesis because

• unlike waveform concatenation, it offers fine-grained control over speech output
• it is relatively easy to obtain data for a formant synthesizer

Question

• Why is so little progress made in Speech Synthesis research?

Answers

• too much engineering
• emphasis on robustness rather than completeness
• linguistic model based on SPE
• phonetic model based on segmental units

Solutions

• computational linguistics
• speech generation rather than text-to-speech conversion
• linguistic model based on prosodic theory
• phonetic model based on overlapping constituents

Arthur Dirksen / adirksen@prl.philps.nl / January 1995