Interpreting Speaker Characteristics in the Dimensions of Self-Supervised Speech Features
IEEE Signal Processing Letters · 2026 · 33 · 3132-3136
DOI: 10.1109/lsp.2026.3716838 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Kyle Janse van Rensburg und 2 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben.
Abstract (Originalsprache)
How do speech models trained through self-supervised learning structure their representations? Previous studies have looked at how information is encoded in feature vectors across different layers. But few studies have considered whether speech characteristics are captured within individual dimensions of SSL features. In this paper, we specifically look at speaker information using PCA on utterance-averaged representations. For a range of SSL models, we find that the principal dimension that explains most variance encodes pitch and associated characteristics like gender. Other individual principal dimensions correlate with intensity, noise levels, the second formant, and higher frequency characteristics. We then use synthesis analyses to show that the dimensions for most characteristics are isolated from each other’s influence. We further show that characteristics can be changed by manipulating the corresponding dimensions.
Zitieren
APA
Rensburg, K. J. V., Niekerk, B. V., & Kamper, H. (2026). Interpreting Speaker Characteristics in the Dimensions of Self-Supervised Speech Features. IEEE Signal Processing Letters, 33, 3132-3136. https://doi.org/10.1109/lsp.2026.3716838
Vancouver
Rensburg KJV, Niekerk BV, Kamper H. Interpreting Speaker Characteristics in the Dimensions of Self-Supervised Speech Features. IEEE Signal Processing Letters. 2026;33:3132-3136. doi:10.1109/lsp.2026.3716838
BibTeX
@article{Rensburg2026,
author = {Rensburg, Kyle Janse van and Niekerk, Benjamin van and Kamper, Herman},
title = {Interpreting Speaker Characteristics in the Dimensions of Self-Supervised Speech Features},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {3132--3136},
doi = {10.1109/lsp.2026.3716838}
}
EUVIMED