Elamur Research⌕
☰
home / papers / 2026 / AURORA Model of Formant-to-tongue Inversion for Didactic and Clinical Applications

AURORA Model of Formant-to-tongue Inversion for Didactic and Clinical Applications

Summary

Phoneticians, clinicians and learners often explain vowel formants with a simple rule: F1 tracks tongue lowering and F2 tracks tongue fronting. This rule ignores nonlinear acoustics, the coupling between formants, and the fact that tongue height and shape depend on each other. The AURORA model offers a visual explanation that is still simple but more faithful to real articulation. It works by articulatory inversion, estimating a two-dimensional tongue contour from F1 and F2 alone. The model was trained on synchronised ultrasound and audio recordings of 40 Northern English speakers. A multivariate linear regression maps the formants and their interaction onto six tongue parameters, and a Procrustes-based reconstruction step turns these back into a smoothed contour. The authors chose not to score the model against a test set. They evaluated it qualitatively and found that its predictions separate the tongue configurations expected for different formant combinations. Two accessible tools build on the model: a Shiny app and a prototype for real-time tongue biofeedback. Intended uses include phonetics teaching and speech therapy, such as gender-affirming voice training.

Key points

  • Training data: 3,856 vowel tokens from a pre-existing ultrasound and acoustic corpus.
    • Speakers were 40 first-language English speakers aged 18–48.
    • Stimuli were ten b_d monosyllables, such as bead, bird and bud.
  • Measurement pipeline: Tongue contours were tracked with DeepLabCut using 11 points from the vallecula to the tongue tip. Formants were measured with FastTrack, and both were sampled at the vowel midpoint.
  • Model structure: F1, F2 and F1×F2 predict six outputs.
    • Four outputs are the X/Y positions of knot 1 and knot 11.
    • Two outputs are tongue-shape principal components from a Procrustes-plus-PCA analysis done with the R shapes package.
    • Together, these two PCs explain 67.54% of shape variance.
  • Evaluation: Tongue shapes were predicted over a grid of formant values between the 5th and 95th quantiles of the data. F1 ranged from 320 to 903 Hz and F2 from 828 to 2616 Hz.
  • Tools and limitations: The model is released through a Shiny app and a real-time biofeedback prototype, which uses a precomputed lookup table for speed. It has been trained only on English, so how well it generalises to other languages remains untested.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/2sw67miiiyc7

@inproceedings{strycharczuk2026aurora,
  title = {{AURORA Model of Formant-to-tongue Inversion for Didactic and Clinical Applications}},
  author = {Strycharczuk, Patrycja and Kirkham, Sam},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/2sw67miiiyc7},
  url = {https://doi.org/10.63317/2sw67miiiyc7}
}

References

27 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Articulate Instruments Ltd. 2014. Articulate Assistant Advanced ultrasound module user manual, revision 2.16. find
  2. Santiago Barreda. 2021. Fast track: fast (nearly) automatic formant-tracking using Praat. Linguistics Vanguard, 7. find
  3. Paul Boersma and David Weenink. 2009. Praat: doing phonetics by computer [Computer programme]. find
  4. Donald G Childers and Ke Wu. 1991. Gender recognition from speech. Part II: Fine analysis. The Journal of the Acoustical society of America, 90(4):1841–1856. find
  5. Ian L Dryden and Kanti V Mardia. 2016. Statistical shape analysis: with applications in R. John Wiley & Sons. find
  6. Gunnar Fant. 1971. Acoustic theory of speech production: with calculations based on X-ray studies of Russian articulations. Walter de Gruyter. find
  7. Erica Gold, Christin Kirchhübel, Kate Earnshaw, and Sula Ross. 2022. Regional variation in British English voice quality. English World-Wide, 43(1):96–123. find
  8. Kiyoshi Honda. 1996. Organization of tongue articulation for vowels. Journal of Phonetics, 24:39– 52. find
  9. Kiyoshi Honda, Sayoko Takano, and Hironori Takemoto. 2010. Effects of side cavities and tongue stabilization: Possible extensions of the quantal theory. Journal of Phonetics, 38(1):33–43. find
  10. Deanna Kawitzky and Tara McAllister. 2020. The effect of formant biofeedback on the feminization of voice in transgender women. Journal of Voice, 34:53–67. find
  11. John T. Kent. 1992. New directions in shape analysis. In Kanti V. Mardia, editor, The Art of Statistical Science, pages 115–127. Wiley, New York. find
  12. Ray D Kent and Y-J Kim. 2003. Toward an acoustic typology of motor speech disorders. Clinical linguistics & phonetics, 17(6):427–445. find
  13. Sam Kirkham and Patrycja Strycharczuk. 2025. Dynamical model parameters from ultrasound tongue kinematics. JASA Express Letters, 5(11). find
  14. Sam Kirkham, Patrycja Strycharczuk, Emily Gorman, Takayuki Nagamine, and Alan Wrench. 2023. Co-registration of simultaneous high speed ultrasound and electromagnetic articulography for speech production research. In Proceedings of the 20th International Congress of Phonetic Sciences, pages 918–922. find
  15. William Labov. 2006. A sociolinguistic perspective on sociophonetic research. Journal of Phonetics, 34(4):500–515. find
  16. John Laver. 1980. The Phonetic Description of Voice Quality. Cambridge: Cambridge University Press. find
  17. Yeptain Leung, Jennifer Oates, and Siew Pang Chan. 2018. Voice, articulation, and prosody contribute to listener perceptions of speaker gender: A systematic review and meta-analysis. Journal of Speech, Language, and Hearing Research, 61:266–297. find
  18. Clara Leyns, Paul Corthals, Marjan Cosyns, Tine Papeleu, John Van Borsel, Dominique Mor-somme, Guy T’Sjoen, and Evelien D’haeseleer. 2024. Acoustic and perceptual effects of articulation exercises in transgender women. Journal of Voice, 38:246–e15. find
  19. Justin J. H. Lo, Patrycja Strycharczuk, and Sam Kirkham. 2025. Articulatory strategy in vowel production as a basis for speaker discrimination. In Interspeech 2025, pages 3504–3508. find
  20. Shinji Maeda. 1990. Compensatory articulation during speech: Evidence from the analysis and synthesis of vocal-tract shapes using an articulatory model. In William J. Hardcastle and Alain Marchal, editors, Speech production and speech modelling, pages 131–149. Dordrecht: Kluwer. find
  21. Alexander Mathis, Pranav Mamidanna, Kevin M Cury, Taiga Abe, Venkatesh N Murthy, Mackenzie Weygandt Mathis, and Matthias Bethge. 2018. Deeplabcut: markerless pose estimation of user-defined body parts with deep learning. Nature neuroscience, 21(9):1281–1289. find
  22. Tara McAllister, Collin Eagen, Victoria S McKenna, Yi Shan, Peter Traver, Jayce Garner, Daphna Harel, Tae Hong Park, and Vesna D Novak. 2025. Real-time resonance biofeedback for gender-affirming voice training: Usability testing of the truvox web-based application. Journal of Voice. find
  23. Lorenzo Spreafico, Michael Pucher, and Anna Matosova. 2018. UltraFit: A speaker-friendly headset for ultrasound recordings in speech science. In Interspeech 2018, 19th Annual Conference of the International Speech Communication Association, Hyderabad, India, 2-6 September 2018, pages 1517–1520. Inte… find
  24. Kenneth N Stevens and Arthur S House. 1955. Development of a quantitative description of vowel articulation. The Journal of the Acoustical Society of America, 27:484–493. find
  25. Matthew Strimas-Mackey. 2023. smoothr: Smooth and Tidy Spatial Features. R package version 1.0.1. find
  26. Patrycja Strycharczuk, Sam Kirkham, Emily Gorman, and Takayuki Nagamine. 2025. Dimensionality reduction in lingual articulation of vowels: Evidence from lax vowels in northern anglo-english. Language and Speech, 68(3):689–721. find
  27. Alan Wrench and Jonathan Balch-Tomes. 2022. Beyond the edge: Markerless pose estimation of speech articulators from ultrasound and camera images using DeepLabCut. Sensors, 22:1133. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/strycharczuk-2026-aurora-model-formant-tongue-inversion/