Optimal Control Theory of speech production using probabilistic articulatory-acoustic models

Benjamin Elie, Juraj Šimko, Alice Turk

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract / Description of output

This paper introduces the concept of using probabilistic articulatory-acoustic models in optimization-based models of speech articulatory planning. In these models, speech articulatory movements are assumed to be performed such that they satisfy conflicting task requirements, such as least effort and intelligibility. Our model is used to account for intelligibility by computing the posterior probability of a vowel given a vector of formant values. These models can be trained on formant values and associated vowel labels extracted from available audio corpora. We present a minimal example using a model of the American-English vocalic system trained on formant values extracted from the TIMIT database. A preliminary experiment illustrates the interest of the approach by reproducing vowel centralization when least effort is weighted more highly than intelligibility.
Original languageEnglish
Title of host publicationProceedings of the 20th International Congress of Phonetic Sciences
EditorsRadek Skarnitzl, Jan Volín
Place of PublicationPrague
PublisherGuarant International
Pages942-946
ISBN (Electronic)9788090811423
Publication statusPublished - 10 Aug 2023
Event20th International Conference of Phonetic Sciences (ICPhS) - Prague Congress Centre, Prague, Czech Republic
Duration: 7 Aug 202311 Aug 2023
https://www.icphs2023.org/

Conference

Conference20th International Conference of Phonetic Sciences (ICPhS)
Abbreviated titleICPhS 2023
Country/TerritoryCzech Republic
CityPrague
Period7/08/2311/08/23
Internet address

Keywords / Materials (for Non-textual outputs)

  • articulatory planning
  • speech production
  • Optimal Control Theory

Fingerprint

Dive into the research topics of 'Optimal Control Theory of speech production using probabilistic articulatory-acoustic models'. Together they form a unique fingerprint.

Cite this