Explainable by-design Audio Segmentation through Non-Negative Matrix Factorization and Probing - Equipe Signal, Statistique et Apprentissage Access content directly
Conference Papers Year : 2024

Explainable by-design Audio Segmentation through Non-Negative Matrix Factorization and Probing

Abstract

Audio segmentation is a key task for many speech technologies, most of which are based on neural networks, usually considered as black boxes, with high-level performances. However, in many domains, among which health or forensics, there is not only a need for good performance but also for explanations about the output decision. Explanations derived directly from latent representations need to satisfy "good" properties, such as informativeness, compactness, or modularity, to be interpretable. In this article, we propose an explainable-by-design audio segmentation model based on non-negative matrix factorization (NMF) which is a good candidate for the design of interpretable representations. This paper shows that our model reaches good segmentation performances, and presents deep analyses of the latent representation extracted from the non-negative matrix. The proposed approach opens new perspectives toward the evaluation of interpretable representations according to "good" properties.
Fichier principal
Vignette du fichier
nmf_probing2024-4.pdf (592.3 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-04617131 , version 1 (19-06-2024)

Identifiers

  • HAL Id : hal-04617131 , version 1

Cite

Martin Lebourdais, Théo Mariotte, Antonio Almudévar, Marie Tahon, Alfonso Ortega. Explainable by-design Audio Segmentation through Non-Negative Matrix Factorization and Probing. Interspeech 2024, International Speech Communication Association (ISCA), Sep 2024, Kos / Greece, France. ⟨hal-04617131⟩
104 View
12 Download

Share

Gmail Mastodon Facebook X LinkedIn More