Introduction

IPOX is an all-prosodic speech synthesizer, based in part on YorkTalk (Coleman 1992, Local 1994), currently being developed by Arthur Dirksen at IPO and John Coleman at Oxford University.

All-prosodic means:

• no principled distinction between segmental and supra-segmental prosodic structure
• speech generated directly from prosodic structure

Properties:

• declarative, no rewriting rules
• analysis of input using constraint-based grammars
• compositional phonetic interpretation
• metrical structure (prosodic headedness) used in both analysis and phonetic interpretation
• fully integrated system, including formant synthesis and speech output

Goals:

• further formalize and extend the YorkTalk architecture
• develop a multi-lingual version
• make the YorkTalk model accessible to phonologists and speech researchers

Current capabilities (British English):

• syllable overlap (used to glue syllables together)
• syllable compression (used to generate the correct rhythm)
• a first attempt at connected speech

These demo's were played at the Second ESCA/IEEE Workshop on Speech Synthesis (September 1994), Dirksen and Coleman 1994.