Vous êtes ici

Séminaire du jeudi 16/10/2014

Jeudi, Octobre 16, 2014 - 14:00 to 16:00
IMS, Bat A31, Salle: Amphi J.-F. Dom

Orateur : Zuheng Ming

Affiliation : Labri

Titre : Spectral Parameters to Cued Speech Parameters Mapping: Multi-linear and GMM approaches (applied to French vowels)

Résumé : Cued Speech (CS) is a visual communication system that uses hand shapes placed in different positions near the face, in combination with the natural speech lip-reading, to enhance speech perception from visual input for deaf people. This is a system where the speaker, facing the perceiver, moves his hand in close relation with speech. It proves that CS is largely improving speech perception by eliminating the ambiguity in the lip reading. Moreover, CS offers to deaf people a thorough representation of the phonological system, inasmuch as they have been exposed to this method since their youth, and therefore it has a positive impact on the language development. While the CS method offers a real advantage for complete speech perception for deaf people, one of the important challenges is the question of speech communication between normal hearing people who do not practice CS but produce acoustic speech and deaf people with no audition rests . To solve this question, one can use a human cuer as a translator or developing the automatic translation systems. In this work, a framework of automatic translation system for mapping the speech to CS in the case of French vowels is presented. The classic automatic system for converting audio speech to CS components consists of coupling a recognition system to a text-to-visual speech synthesizer. This work proposed a new approach using a low signal level based on the multi-linear regression or GMM-based regression methods for mapping speech spectral parameters to the visual components of CS coupling with the lip shape.

Beside, a direct estimation of lip geometry features from natural image of mouth region-of-interest prediction based on the 2D Discrete Cosine Transform combined with a Principal Component Analysis is proposed. The results show the possibility to estimate the geometric lip features with a good accuracy using a reduce set of predictors derived from the DCT coefficients.

Orateur : Tülay Adali

Affiliation : Department of Computer Science and Electrical Engineering, University of Maryland Baltimore County

Titre : ICA and IVA: Theory, Connections, and Applications to Medical Imaging

Résumé :
Data-driven methods are based on a simple generative model and hence can minimize the underlying assumptions on the data. They have emerged as promising alternatives to the traditional model-based approaches in many applications where the unknown dynamics are hard to characterize. Independent component analysis (ICA), in particular, has been a popular data-driven approach and an active area of research. Starting from a simple linear mixing model and the assumption of statistical independence, one can recover a set of linearly-mixed components to within a scaling and permutation ambiguity. It has been successfully applied to numerous data analysis problems in areas as diverse as biomedicine, communications, finance, geophysics, and remote sensing.

This talk reviews the fundamentals and properties of ICA, and provides a unified view of two main approaches for achieving ICA, those that make use of non-Gaussianity and sample dependence. Then, the generalization of ICA for analysis of multiple datasets, independent vector analysis (IVA), is introduced and the connections between ICA and IVA are highlighted, especially in the way both approaches make use of signal diversity. Examples are presented to demonstrate the application of ICA and IVA to analysis of functional magnetic resonance imaging data as well as fusion of data from multiple imaging modalities.

Tülay Adali is a Professor in the Department of Computer Science and Electrical Engineering at the University of Maryland Baltimore County. She has actively assisted the IEEE in organizing numerous international conferences and workshops, chaired and served on various technical committees and editorial boards including the Proceedings of the IEEE. She is a Fellow of the IEEE and the AIMBE, an IEEE Distinguished Lecturer, and the recipient of an NSF CAREER Award, 2010 IEEE Signal Processing Society Best Paper Award, and 2013 University System of Maryland Regents' Award for Research. Her research interests include statistical signal processing, machine learning for signal processing, and medical image analysis.