Planning the Articulation of Spoken Utterances

The EU-funded PlanArt: Planning the Articulation of Spoken Utterances project, led by Professor Alice Turk, explores mental representations and processes involved in planning speech articulation.

Due to systematic phonetic variability, for example, different pronunciations for each word[i] in different contexts in an utterance, planning speech articulation is complicated. This involves accessing mental representations of speech sounds, sequencing them into an appropriate structure that includes phrasing, as well as planning the articulatory movements that produce the sounds.

The nature of those mental representations is a major theoretical debate in Linguistics.

• If the mental representations are symbolic sound categories, they require phonetic planning, an extra processing component to plan the details of how the sounds are produced.
• If they are spatiotemporal they have some details of how the sounds are to be produced, and mechanisms to adjust these in different contexts in an utterance are required.

The project tests these theories in two ways, with experiments and software development. An international, interdisciplinary team designs and conducts experiments on speech articulation and analyses results; the project’s PlanArt software is developed on a single platform to enable the comparison of different types of speech articulation planning models. The experiments incorporate techniques such as Electromagnetic Articulometry (EMA), Laryngography and acoustic recordings. Data are automatically post-processed, anonymised and then stored in DataStore.

The project’s internal bespoke software is used for trajectory modelling using a parametrized model; the project also makes use of open-source and commercial software packages for articulatory visualisation and analysis. Statistical analysis is completed using R and SPSS (Statistical Package for the Social Sciences). The project’s PlanArt software incorporates Python code to implement different theories, stored at a GitLab repository with analysis completed in Eddie and making resources publicly available to the wider public using DataShare.

[i] Examples are available in the Edinburgh Speech Production Facility (ESPF) DoubleTalk dataset.

This case study was developed after the CAHSS Research in Action Event in November 2023.
University of Edinburgh: Alice Turk, Benjamin Elie, Dave Lee, Cedric Macmartin, Satsuki Nakai, Zihang Peng
External Collaborators: James Kirby, Jim Scobbie, Juraj Simko, Stefanie Shattuck-Hufnagel

• Elie, B., Šimko, J. and Turk, A., 2024. Optimization-based modeling of Lombard speech articulation: Supraglottal characteristics. JASA Express Letters, 4(1).
• Elie, B. and Turk, A., 2023, May. Estimating virtual targets for lingual stop consonants using general Tau theory. In Interspeech 2023. ISCA. 
• Elie, B., Šimko, J. and Turk, A., 2023. Optimal control of speech with context-dependent articulatory targets. Interspeech 2023. 
• Elie, B., Šimko, J. and Turk, A., 2023, March. Optimal control theory of speech production using probabilistic articulatory-acoustic models. In 20th International Conference of Phonetic Sciences (ICPhS).
• Turk, A., Elie, B. and Simko, J., 2023, March. PlanArt: A modular platform for computational modeling of articulatory planning. In 20th International Conference of Phonetic Sciences (ICPhS).
• Elie, B., Lee, D.N. and Turk, A., 2023. Modeling trajectories of human speech articulators using general Tau theory. Speech Communication, 151, pp.24-38.