line decor
line decor

Home>News & events


Building an articulatory model from ultrasound, EMA and MRI data

Postdoctoral position

until 28th February 2008

Research project

An articulatory model comprises both the visible and the internal mobile articulators which are involved in speech articulation: the lower jaw, tongue, lips and velum) as well as the fixed walls (the palate, the rear wall of the pharynx). An articulatory model is dynamic since the articulators deform during speech production. Such a model has a potential interest in the field of language learning by providing visual feedback on the articulation conducted by the learner, and many other applications.
Building an articulatory model is difficult because the different articulators have to be detected from specific image modalities: the lips are acquired through video, the tongue shape is acquired through ultrasound imaging with a high frame rate but these 2D images are very noisy. Finally, 3D images of all articulators can be obtained with MRI but only for sustained sounds (as vowels) due to the long acquisition time of MRI images.
The subject of this post-doc is to construct a dynamic 3D model of the entire vocal tract by merging the 3D information available in the MRI acquisitions and temporal 2D information provided by the contours of the tongue visible on the ultrasound images or X-ray images.
We are working on the construction of an articulatory model within the European project ASPI.
We already built an acquisition system which allows us to obtain synchronized data from ultrasound, MRI, video and EM modalities.
Only a few complete articulatory models are currently available in the world and a real challenge in the field is to design set-ups and easy-to-use methods for automatically building the model of any speaker from 3D and 2D images. Indeed, the existence of more articulatory models would open new directions of research about speaker variability and speech production.


The aim of the subject is to build a deformable model of the vocal tract from static 3D MRI images and 2D dynamic 2D sequences. Previous works have been conducted on the modelling of the vocal tract, and especially of the tongue (M. Stone[1] O. Engwall[2]). Unfortunately, important human interaction is required to extract tongue contours in the images. In addition, only one image modality is often considered in these works, thus reducing the reliability of the model obtained.
The aim of this work is to provide automatic methods for segmenting features in the images as well as methods for building a parametric model of the 3D vocal tract with these specific aims:

  1. * The segmentation process is to be guided by prior knowledge on the vocal tract. In particular shape, topologic as well as regularity constraints must be considered.
  2. * A parametric model of the vocal tract has to be defined (classical models are linear and built from a principal component analysis). Special emphasis must be put on the problem of matching the various features between the images.
  3. * Besides classical geometric constraints, both the building and the assessment of the model will be guided by acoustic distances in order to check for the adequation between the sound synthesized from the model and the sound realized by the human speaker.

Skill and profile

The recruited person must have a solid background in computer vision and in applied mathematics. Informations and demonstrations on the research topics addressed by the Magrit team are available at


[1] M. Stone: Modeling tongue surface contours from Cine-MRI images. Journal of Speech, language, hearing research, 2001.
[2] P. Badin, G. Bailly, L. Reveret: Three-dimensional linear articulatory modeling of tongue, lips and face based on MRI and video images, Journal of Phonetics, 2002, vol 30, p 533-553


Interested candidates are invited to contact Marie-Odile Berger:, +33 3 54 95 85 01

Important information

This position is advertised in the framework of the national INRIA campaign for recruiting post-docs. It is a one year position, renewable, beginning fall 2008. The salary is 2,320 euros gross per month.
Selection of candidates will be a two step process. A first selection for a candidate will be carried out internally by the Magrit group. The selected candidate application will then be further processed for approval and funding by an INRIA committee.
Doctoral thesis less than one year old (May 2007) or being defended before end of 2008. If defence has not taken place yet, candidates must specify the tentative date and jury for the defence.

Important - Useful links

Presentation of INRIA postdoctoral positions
To apply (be patient, loading this link takes times...)