SLAM 3: An Updated Stylization Model for Speech Melody - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

SLAM 3: An Updated Stylization Model for Speech Melody

Résumé

This paper presents the newest version of the annotation software Stylization and Labelling of Speech Melody (SLAM), a language-independent prosodic model that automatically annotates pitch contours in linguistic units of arbitrary length. We review the core principles of SLAM before describing several shortcomings and the innovations introduced in SLAM 3 to address them. These notably include methods implemented to minimize the influence of F0 microvariations and alignment errors and to better model the perception of short-duration pitch changes. Secondly, we present additional functionality allowing speech segments to be annotated relative to the mean pitch of their nearest neighbors, reducing the influence of downdrift on annotations. Finally, we demonstrate the utility of these changes by comparing SLAM 3 against its predecessor in terms of measured distances between their stylized outputs and the natural pitch contours used as input.
Fichier principal
Vignette du fichier
SLAM3_accepted_corrected.pdf (366.02 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04171671 , version 1 (26-07-2023)

Identifiants

  • HAL Id : hal-04171671 , version 1

Citer

Emmett Strickland, Marc Evrard, Anne Lacheret-Dujour. SLAM 3: An Updated Stylization Model for Speech Melody. International Congress of Phonetic Sciences, Aug 2023, Prague, Czech Republic. ⟨hal-04171671⟩
116 Consultations
33 Téléchargements

Partager

Gmail Facebook X LinkedIn More