Trajectories predicted by optimal speech motor control using LSTM networks - Université Grenoble Alpes Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Trajectories predicted by optimal speech motor control using LSTM networks

Résumé

The question of optimality and its role in trajectory formation is at the core of important debates in motor control research. We present the first speech control model that associates Optimal Feedback Control (OFC) for planning and execution of movements with a biomechanical model of the vocal tract. Simulated trajectories in the VCV sequences are compared with trajectories generated using the GEPPETO model that drives the same 2D biomechanical model; in GEPPETO, the scope of optimality is limited to movement planning and to phoneme-related target motor commands. In our OFC model commands are estimated via the minimisation of a cost that combines neuromuscular effort, and a penalty on accuracy of the auditory patterns reached for the phonemes. The biomechanics of the plant are implemented by an LSTM trained on simulations of a finite element model of the tongue. The comparison of the OFC model with GEPPETO relies on the time variation of the motor commands, the shape of the articulatory trajectories, and on auditory trajectories in the F1-F2 planes.
Fichier principal
Vignette du fichier
Paper_Template_for_INTERSPEECH_2022.pdf (1.33 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03788795 , version 1 (01-10-2022)

Identifiants

Citer

Tsiky Rakotomalala, Pascal Perrier, Pierre Baraduc. Trajectories predicted by optimal speech motor control using LSTM networks. Interspeech 2022 - 23rd Annual Conference of the International Speech Communication Association, Sep 2022, Incheon, South Korea. pp.630-634, ⟨10.21437/interspeech.2022-10604⟩. ⟨hal-03788795⟩
98 Consultations
113 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More