Speaker-Specific Articulatory Assessment and Measurements during Portuguese Speech Production Based on Magnetic Resonance Images

摘要

The development of two and three-dimensional magnetic resonance imaging (MRI) opened new options for a better understanding of speech production; in particular, for the articulation process, comparing with other imaging techniques, such as x-rays. Several MRI studies have been carried out considering different languages, but concerning European Portuguese the available information is far from being completely achieved. Recently, the knowledge gained with the application of deformable models in magnetic resonance images towards the automatic study of the vocal tract, has allowed an enhanced identification and description of the articulatory mechanism and its organs. Our aim is to extract and evaluate the main characteristics of the movements of vocal tract during European Portuguese speech production to achieve speaker-specific articulatory assessment from MRI. For this, we used active shape and active appearance models to identify, i.e. to segment, and simulate the vocal tract's shape in MR images and concluded that both are suitable for such tasks being the later more proficient. The findings obtained are believed to be useful for speech rehabilitation and simulation purposes, namely to recognize and simulate the compensatory movements of the articulators during speech production. Therefore, this chapter gains particular interest within the communities of speech study and rehabilitation, medical imaging and bioengineering. It is organized as follows: the introduction section starts with a literature review concerning the use, application and challenges of MRI in speech production study, in particular for speech articulation. In addition, image analysis techniques based on deformable templates, more specifically by using geometrical shapes driven by parameterized functions, are introduced. In the second section, the adopted methodology of MRI acquisition and data assessment are described. Based on this image analysis approach, in the results section the key aspects of articulatory movements during the production of relevant European Portuguese speech sounds are addressed. In the final section of this chapter, conclusions are presented and further suggestions for future work are indicated.