Jul 2026
InsideSSL: Understanding Self-Supervised Speech Representations using a Model-Centric Perspective
This work establishes a task-agnostic analysis from three intrinsic per-layer perspectives: compression, geometry, and robustness to perturbations, and introduces the cross-layer Generative Compatibility Matrix (GCM) to evaluate functional transferability.
Samir Sadok, Xavier Alameda-Pineda
· arXiv.org · 1 citation