Skip to Main content Skip to Navigation
Theses

Modélisation de l'espace articulatoire par un codebook hypercubique pour l'inversion acoustico-articulatoire

Abstract : In this thesis, we deal with the inversion of the articulatory-to-acoustic relation, i.e. given an acoustic signal we want to recover the trajectories of the corresponding articulatory parameters. For this purpose, we have to resolve three problems : modelling articulatory space by hypercubes, retrieving all the solutions, and recovering articulatory trajectories varying slowly. Our inversion method is based on the representation of the articulatory space by a hypercube codebook. This representation has the advantage of decomposing the articulatory space into regions where the mapping is quasi-linear. Each region is represented by a hypercube. The inversion procedure retrieves articulatory vectors corresponding to an acoustic entry from the hypercube codebook. As the dimension of the articulatory space is greater than the dimension of the acoustic space, the corresponding null space is sampled by linear programming to retrieve all the possible solutions. Retrieving articulatory trajectories is performed in two steps. We use non-linear smoothing method based on dynamic programming followed by smoothing with a variation al method. We have succeeded to retrieve smooth and realistic articulatory trajectories, which is confirmed by the experimental evaluation.
File URL :
http://docnum.univ-lorraine.fr/prive/SCD_T_2001_0210_OUNI.pdf
Complete list of metadatas

https://hal.univ-lorraine.fr/tel-01746502
Contributor : Thèses Ul <>
Submitted on : Thursday, March 29, 2018 - 10:39:09 AM
Last modification on : Tuesday, April 24, 2018 - 1:30:17 PM

Identifiers

  • HAL Id : tel-01746502, version 1

Collections

Citation

Slim Ouni. Modélisation de l'espace articulatoire par un codebook hypercubique pour l'inversion acoustico-articulatoire. Autre [cs.OH]. Université Henri Poincaré - Nancy 1, 2001. Français. ⟨NNT : 2001NAN10210⟩. ⟨tel-01746502⟩

Share

Metrics

Record views

31