Abstract:Enhancing explainability in speech self-supervised learning (SSL) is important for understanding and effectively utilising speech SSL representations. This study conducts a large-scale layer-wise probing analysis of 11 speech SSL models, examining speaker identity together with acoustic, prosodic, and paralinguistic attributes. The results confirm a general hierarchy wherein initial layers encode fundamental acoustics and middle layers synthesise abstract traits. The consensus that final layers purely abstract linguistic content is challenged, as larger models unexpectedly recover speaker-discriminative information in their deep layers. Furthermore, the intermediate representations of speech SSL models are found to provide stronger probing performance for several attributes than specialised speaker embeddings. These insights give a systematic empirical characterisation of speaker-related information in SSL models, providing guidelines for selecting representations for downstream tasks.
From: Aemon Yat Fei Chiu [view email]
[v1]
Thu, 9 Jan 2025 15:26:33 UTC (471 KB)
[v2]
Thu, 18 Sep 2025 11:22:07 UTC (107 KB)
[v3]
Thu, 5 Mar 2026 11:57:52 UTC (121 KB)
[v4]
Sat, 5 Sep 2026 11:50:51 UTC (119 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。