Zooming in on Spatiotemporal V-to-C Coarticulation with Functional PCA

Michele Gubian, Manfred Pastätter, Marianne Pouplier


It has long been proposed in speech production research that in CV sequences, the movement for consonant and vowel are initiated synchronously. However, mostly due to limitations on the statistical analysis of articulator motion over time, this could only be shown in a limited fashion, based on positional differences at a single time point during consonantal constriction formation. It is unknown to which extent this observation generalizes to earlier timepoints. In this paper, we illustrate the use of functional principal component analysis (FPCA) for the statistical analysis of articulator motion over time. Using articulography data, we quantify CV coarticulation during constriction formation of [k] in two vowel contexts. We show how FPCA enables us to analyse both horizontal and vertical movement components over time in a single model while preserving information on temporal variability. We combine FPCA with linear mixed modelling to obtain estimated mean trajectories and confidence bands for [k] in the two vowel contexts. Results show that well before the timepoint of peak velocity the vowel causes a substantial spatial separation of the consonantal trajectories, estimated to be at least 3 mm at peak velocity. This lends support to the hypothesis that vowel and consonant are initiated synchronously.


 DOI: 10.21437/Interspeech.2019-2143

Cite as: Gubian, M., Pastätter, M., Pouplier, M. (2019) Zooming in on Spatiotemporal V-to-C Coarticulation with Functional PCA. Proc. Interspeech 2019, 889-893, DOI: 10.21437/Interspeech.2019-2143.


@inproceedings{Gubian2019,
  author={Michele Gubian and Manfred Pastätter and Marianne Pouplier},
  title={{Zooming in on Spatiotemporal V-to-C Coarticulation with Functional PCA}},
  year=2019,
  booktitle={Proc. Interspeech 2019},
  pages={889--893},
  doi={10.21437/Interspeech.2019-2143},
  url={http://dx.doi.org/10.21437/Interspeech.2019-2143}
}