Establishing degrees of closeness between audio recordings along different dimensions using large-scale cross-lingual models - Laboratoire Interdisciplinaire des Sciences du Numérique Access content directly
Conference Papers Year : 2024

Establishing degrees of closeness between audio recordings along different dimensions using large-scale cross-lingual models

Abstract

In the highly constrained context of low-resource language studies, we explore vector representations of speech from a pretrained model to determine their level of abstraction with regard to the audio signal. We propose a new unsupervised method using ABX tests on audio recordings with carefully curated metadata to shed light on the type of information present in the representations. ABX tests determine whether the representations computed by a multilingual speech model encode a given characteristic. Three experiments are devised: one on room acoustics aspects, one on linguistic genre, and one on phonetic aspects. The results confirm that the representations extracted from recordings with different linguistic/extra-linguistic characteristics differ along the same lines. Embedding more audio signal in one vector better discriminates extra-linguistic characteristics, whereas shorter snippets are better to distinguish segmental information. The method is fully unsupervised, potentially opening new research avenues for comparative work on under-documented languages.
Fichier principal
Vignette du fichier
2024.findings-eacl.154.pdf (526.64 Ko) Télécharger le fichier
Origin Publisher files allowed on an open archive
licence

Dates and versions

hal-04561819 , version 1 (29-04-2024)

Licence

Identifiers

Cite

Maxime Fily, Guillaume Wisniewski, Séverine Guillaume, Gilles Adda, Alexis Michaud. Establishing degrees of closeness between audio recordings along different dimensions using large-scale cross-lingual models. Findings of the Association for Computational Linguistics: EACL 2024, Association for Computational Linguistics, Mar 2024, St. Julian’s, Malta. ⟨hal-04561819⟩
43 View
10 Download

Altmetric

Share

Gmail Mastodon Facebook X LinkedIn More