Skip to Main content Skip to Navigation
Theses

Towards a 3 dimensional dynamic generic speaker model to study geometry simplifications of the vocal tract using magnetic resonance imaging data

Ioannis Douros 1
1 MULTISPEECH - Speech Modeling for Facilitating Oral-Based Communication
Inria Nancy - Grand Est, LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery
Abstract : In this thesis we used MRI (Magnetic Resonance Imaging) data of the vocal tract to study speech production. The first part consist of the study of the impact that the velum, the epiglottis and the head position has on the phonation of five french vowels. Acoustic simulations were used to compare the formants of the studied cases with the reference in order to measure their impact. For this part of the work, we used 3D static MR (Magnetic Resonance) images. As speech is usually a dynamic phenomenon, a question arose, whether it would be possible to process the 3D data in order to incorporate dynamic information of continuous speech. Therefore the second part presents some algorithms that one can use in order to enhance speech production data. Several image transformations were combined in order to generate estimations of vocal tract shapes which are more informative than the original ones. At this point, we envisaged apart from enhancing speech production data, to create a generic speaker model that could provide enhanced information not for a specific subject, but globally for speech. As a result, we devoted the third part in the investigation of an algorithm that one can use to create a spatiotemporal atlas of the vocal tract which can be used as a reference or standard speaker for speech studies as it is speaker independent. Finally, the last part of the thesis, refers to a selection of open questions of the field that are still left unanswered, some interesting directions that one can expand this thesis and some potential approaches that could help someone move forward towards these directions.
Complete list of metadatas

Cited literature [164 references]  Display  Hide  Download

https://hal.univ-lorraine.fr/tel-03008224
Contributor : Thèses Ul <>
Submitted on : Monday, November 16, 2020 - 4:47:22 PM
Last modification on : Tuesday, November 17, 2020 - 11:54:04 AM

Identifiers

  • HAL Id : tel-03008224, version 1

Citation

Ioannis Douros. Towards a 3 dimensional dynamic generic speaker model to study geometry simplifications of the vocal tract using magnetic resonance imaging data. Computation and Language [cs.CL]. Université de Lorraine, 2020. English. ⟨NNT : 2020LORR0115⟩. ⟨tel-03008224⟩

Share

Metrics

Record views

44

Files downloads

20