IPOX Lecture Notes

Declarative Metrical Phonology and Speech Synthesis

An introduction to

LOT Lecture Notes, January 16-20, 1995, Tilburg

Arthur Dirksen, Institute for Perception Research/IPO

Abstract

These pages describe IPOX, an experimental speech synthesis system, based in part on YorkTalk. YorkTalk and IPOX differ from main-stream speech synthesis systems such as DecTalk in that no use is made of rewriting rules. Instead, input strings are analyzed using declarative, constraint-based phrase structure grammars. The representation thus obtained, a metrical-prosodic tree, is assigned a compositional phonetic interpretation in terms of paramaters for a formant synthesizer.

Samples of synthetic speech, generated by IPOX, are included as illustrations, as well as screen clips of the graphic representations displayed during analysis and generation.

Note: In order to play the speech samples, your Web browser must be able to handle 16-bit .wav files sampled at 11025 Hz.

Contents

General

• Introduction
• IPOX versus YorkTalk
• Declarative phonology
• Speech synthesis

Architecture

• Data structure
• Modules
• Constraint-based grammars
• Temporal interpretation
• Parametric interpretion
• Parameter generation and synthesis

Application (British English)

• Vowels and consonants
• Syllables and metrical feet
• Phonological and phonetic coarticulation
• Ambisyllabicity and syllable overlap
• Vowel reduction and elision
• Fundamental frequency

• References

Acknowledgements

The material presented here is co-authored by John Coleman (Oxford University). The research of the present author has been made possible by a fellowship of the Royal Netherlands Academy of Arts and Sciences.

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