基本信息
- 来源: arxiv
- 原始来源: https://arxiv.org/abs/2603.03096v1
- 作者: Kyle Janse van Rensburg, Benjamin van Niekerk, Herman Kamper
- 分类: eess.AS
- 论文时间: 2026-03-03T15:33:39Z
- 论文 PDF: https://arxiv.org/pdf/2603.03096v1.pdf
来源摘要/节选
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. Using WavLM, 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. Finally, in synthesis experiments we show that most characteristics can be controlled by changing the corresponding dimensions. This provides a simple method to control characteristics of the output voice in synthesis applications.
来源说明
当前只保存了官方论文摘要，不代表论文全文。请以原始来源为准。
本页只呈现已做哈希绑定的来源证据，不包含基于旧正文或缺失原文的扩展推断。