Learning vocal tract variables with multi-task kernels

Hachem Kadri 1 Emmanuel Duflos 2 Philippe Preux 1, 3, 4
1 SEQUEL - Sequential Learning
LIFL - Laboratoire d'Informatique Fondamentale de Lille, LAGIS - Laboratoire d'Automatique, Génie Informatique et Signal, Inria Lille - Nord Europe
2 LAGIS-SI
LAGIS - Laboratoire d'Automatique, Génie Informatique et Signal
Abstract : The problem of acoustic-to-articulatory speech inversion continues to be a challenging research problem which sig- nificantly impacts automatic speech recognition robustness and accuracy. This paper presents a multi-task kernel based method aimed at learning Vocal Tract (VT) variables from the Mel-Frequency Cepstral Coefficients (MFCCs). Unlike usual speech inversion techniques based on individual esti- mation of each tract variable, the key idea here is to consider all the target variables simultaneously to take advantage of the relationships among them and then improve learning per- formance. The proposed method is evaluated using synthetic speech dataset and corresponding tract variables created by the TAsk Dynamics Application (TADA) model and com- pared to the hierarchical ε-SVR speech inversion technique.
Type de document :
Communication dans un congrès
36th International Conference on Acoustics, Speech and Signal Processing (ICASSP), May 2011, Prague, Czech Republic. 2011
Liste complète des métadonnées

Littérature citée [20 références]  Voir  Masquer  Télécharger

https://hal.inria.fr/hal-00826050
Contributeur : Preux Philippe <>
Soumis le : mardi 4 juin 2013 - 09:18:41
Dernière modification le : jeudi 11 janvier 2018 - 06:26:40
Document(s) archivé(s) le : jeudi 5 septembre 2013 - 04:19:23

Fichier

ICASSP2011.pdf
Fichiers produits par l'(les) auteur(s)

Identifiants

  • HAL Id : hal-00826050, version 1

Collections

Citation

Hachem Kadri, Emmanuel Duflos, Philippe Preux. Learning vocal tract variables with multi-task kernels. 36th International Conference on Acoustics, Speech and Signal Processing (ICASSP), May 2011, Prague, Czech Republic. 2011. 〈hal-00826050〉

Partager

Métriques

Consultations de la notice

446

Téléchargements de fichiers

155