Proceedings/Recueil Des Communications Année : 2023

From 'snippet-lects' to doculects and dialects: Leveraging neural representations of speech for placing audio signals in a language landscape

Résumé

XLSR-53, a multilingual model of speech, builds a vector representation from audio, which allows for a range of computational treatments. The experiments reported here use this neural representation to estimate the degree of closeness between audio files, ultimately aiming to extract relevant linguistic properties. We use max-pooling to aggregate the neural representations from a 'snippet-lect' (the speech in a 5-second audio snippet) to a 'doculect' (the speech in a given resource), then to dialects and languages. We use data from corpora of 11 dialects belonging to 5 less-studied languages. Similarity measurements between the 11 corpora bring out greatest closeness between those that are known to be dialects of the same language. The findings suggest that (i) dialect/language can emerge among the various parameters characterizing audio files and (ii) estimates of overall phonetic/phonological closeness can be obtained for a little-resourced or fully unknown language. The findings help shed light on the type of information captured by neural representations of speech and how it can be extracted from these representations.

Fichier principal
Vignette du fichier
Snippet_lects_NeuralRepresentations.pdf (155.25 Ko) Télécharger le fichier
DOI

Cite 10.24397/pangloss-0004337 Son Michaud, A. (2012). Enterrée vive [Audio recording]. Pangloss. https://doi.org/10.24397/PANGLOSS-0004337

DOI

Cite 10.24397/pangloss-0004339 Son Michaud, A. (2012). Femme céleste [Audio recording]. Pangloss. https://doi.org/10.24397/PANGLOSS-0004339

DOI

Cite 10.24397/pangloss-0004341 Son Michaud, A. (2012). Sister: The sister’s wedding (version 1) [Audio recording]. Pangloss. https://doi.org/10.24397/PANGLOSS-0004341

DOI

Cite 10.24397/pangloss-0004351 Son Michaud, A. (2012). Only daughter [Audio recording]. Pangloss. https://doi.org/10.24397/PANGLOSS-0004351

Dates et versions

hal-04108652 , version 1 (27-05-2023)
hal-04108652 , version 2 (08-08-2023)

Licence

Identifiants

Citer

Séverine Guillaume, Guillaume Wisniewski, Alexis Michaud. From 'snippet-lects' to doculects and dialects: Leveraging neural representations of speech for placing audio signals in a language landscape. Proceedings of SIGUL 2023: 2nd Annual Meeting of the ELRA/ISCA Special Interest Group on Under-resourced Languages (SIG-UL 2023), ISCA, pp.29-33, 2023, ⟨10.21437/SIGUL.2023-7⟩. ⟨hal-04108652v1⟩
418 Consultations
344 Téléchargements

Altmetric

Partager

  • More