Maria João Medeiros de Vasconcelos

Towards the Automatic Study of the Vocal Tract From Magnetic Resonance Images

Over the last few decades, researchers have been investigating the mechanisms involved in speech production. Image analysis can be a valuable aid in the understanding of the morphology of the vocal tract. The application of magnetic resonance imaging to study these mechanisms has been proven to be reliable and safe. We have applied deformable models in magnetic resonance images to conduct an automatic study of the vocal tract; mainly, to evaluate the shape of the vocal tract in the articulation of some European Portuguese sounds, and then to successfully automatically segment the vocal tracts shape in new images. Thus, a point distribution model has been built from a set of magnetic resonance images acquired during artificially sustained articulations of 21 sounds, which successfully extracts the main characteristics of the movements of the vocal tract. The combination of that statistical shape model with the gray levels of its points is subsequently used to build active shape models and active appearance models. Those models have then been used to segment the modeled vocal tract into new images in a successful and automatic manner. The computational models have thus been revealed to be useful for the specific area of speech simulation and rehabilitation, namely to simulate and recognize the compensatory movements of the articulators during speech production.

Using Statistical Deformable Models to Reconstruct Vocal Tract Shape from Magnetic Resonance Images

The mechanisms involved in speech production are complex and have thus been subject to growing attention by the scientific community. It has been demonstrated that magnetic resonance imaging (MRI) is a powerful means in the understanding of the morphology of the vocal tract. Over the last few years, statistical deformable models have been successfully used to identify and characterize bones and organs in medical images and point distribution models (PDMs) have gained particular relevance. In this work, the suitability of these models has been studied to characterize and further reconstruct the shape of the vocal tract in the articulation of Portuguese European (EP) speech sounds, one of the most spoken languages worldwide, with the aid of MR images. Therefore, a PDM has been built from a set of MR images acquired during the artificially sustained articulation of 25 EP speech sounds. Following this, the capacity of this statistical model to characterize the shape deformation of the vocal tract during the production of sounds was analysed. Next, the model was used to reconstruct five EP oral vowels and the EP fricative consonants. As far as a study on speech production is concerned, this study is considered to be the first approach to characterize and reconstruct the vocal tract shape from MR images by using PDMs. In addition, the findings achieved permit one to conclude that this modelling technique compels an enhanced understanding of the dynamic speech events involved in sustained articulations based on MRI, which are of particular interest for speech rehabilitation and simulation.

Human motion segmentation using active shape models

Human motion analysis from images is meticulously related to the development of computational techniques capable of automatically identifying, tracking and analyzing relevant structures of the body. This work explores the identification of such structures in images, which is the first step of any computational system designed to analyze human motion. A widely used database (CASIA Gait Database) was used to build a Point Distribution Model (PDM) of the structure of the human body. The training dataset was composed of 14 subjects walking in four directions, and each shape was represented by a set of 113 labelled landmark points. These points were composed of 100 contour points automatically extracted from the silhouette combined with an additional 13 anatomical points from elbows, knees and feet manually annotated. The PDM was later used in the construction of an Active Shape Model, which combines the shape model with gray level profiles, in order to segment the modelled human body in new images. The experiments with this segmentation technique revealed very encouraging results as it was able to gather the necessary data of subjects walking in different directions using just one segmentation model.