arXiv：cs.CL· Beimnet Bekele Guta, Xiaoyu Yang, Guangzhi Sun, Philip C. Woodland·· 3 小时前
用路由稀疏自编码器解耦语音中的语言学与副语言学信息
Disentangling Linguistic and Paralinguistic Information with Routed Sparse Autoencoders
AI 导读
研究者将 TopK 稀疏自编码器与路由专属监督、跨因子对抗训练结合，在冻结的 SPEAR 和 WavLM 编码器上实现因子特定保留与抑制：语言学信息在语言学路由中更强，说话人身份、情感、韵律等副语言学因子则保留在副语言学路由、并在语言学路由中大幅减少。
正文
Abstract:Self-supervised speech encoders contain linguistic and paralinguistic information in a shared, entangled representation space. We combine a TopK sparse autoencoder with route-specific supervision and cross-factor adversaries. Across frozen SPEAR and WavLM encoders, independent probes show factor-specific retention and suppression: linguistic information remains stronger in the linguistic route, while paralinguistic factors, including speaker identity, emotion, and prosody, are retained in the paralinguistic route and substantially reduced in the linguistic route. The route organisation learned on LibriSpeech persists on MSP-Podcast without representation-side retraining. Feature-space route interventions further transfer the swapped factor while largely preserving the information carried by the unchanged route. These results show consistent route-selective separation across encoders, corpora, independent probes, and representation-level interventions.
| Comments: | In submission |
| Subjects: | Computation and Language (cs.CL); Audio and Speech Processing (eess.AS) |
| Cite as: | arXiv:2610.10865 [cs.CL] |
| (or arXiv:2610.10865v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.10865 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Xiaoyu Yang [view email]
[v1]
Wed, 7 Oct 2026 20:10:07 UTC (920 KB)
来源：arXiv：cs.CL · arxiv.org