|
Title:
|
Bimodal Approach of Automatic Speech Processing
|
|
Author:
|
Pierre Jourlin
|
|
Email:
|
click here to access email
|
|
Homepage:
|
http://lia.univ-avignon.fr/Perso/JOURLIN/jourlin_gb.html
|
|
Degree Awarded:
|
University of Avignon
, Computer Science
|
|
Degree Date:
|
1997
|
|
Linguistic Subfield(s):
|
Computational Linguistics
|
|
Director(s):
|
Marc El-Beze
|
|
|
Abstract:
|
|
|
This work addresses the problem of integrating visual information about the lip movements into acoustic-based speech processing systems.
In a first part, the different problems and methods related to this particular approach are discussed.
The first chapter is dedicated to the asynchrony phenomena (or temporal independence) of the two sources of information and to the different ways of handling it in HMM-based speech processing systems.
The second chapter concerns the problem of combining two modalities which may have very different reliabilities : Different methods for finding an optimal weighting are discussed.
The second part report the different experiments I have done in the acoustic-labial speech processing field.
In the chapter 3 are reported my own experiments in the acoustic-labial speech recognition field and in the framework of AMIBE (French CNRS project).
I also present in this chapter the results of the novel methods described in chapter 1 and 2.
Chapter 4 is dedicated to acoustic-labial person authentication systems and experiments, in the framework of the European project M2VTS (program ACTS).
These experiments are the very first conduced in the field of acoustic-labial speaker recognition and verification.
|
|
|
E-mail this message to a friend
