Model Releases
Revisiting Label-Free Speaker Embedding Enhancement with vMF Profile Likelihood
arXiv:2610.06691v1 Announce Type: cross Abstract: Embedding enhancement improves speaker verification under acoustic mismatch without modifying a frozen backbone. Recent work has established a practic
arXiv:2610.06691v1 Announce Type: cross Abstract: Embedding enhancement improves speaker verification under acoustic mismatch without modifying a frozen backbone. Recent work has established a practical label-free setting for this task, but often adopts increasingly structured formulations. Here, the clean target is directly observed during training, making enhancement a matching problem on the unit hypersphere. We model the clean target with a von Mises--Fisher (vMF) likelihood and profile out a sample-wise concentration parameter, yielding a simple closed-form objective with adaptive weighting. Across VoxCeleb1, VoxSRC23, CN-Celeb, VOiCES, and VC-Mix, the proposed method largely preserves the baseline and gives clearer gains on challenging mismatch sets. It also remains stable under a broad single-view recipe, where a recent diffusion baseline becomes less reliable in controlled comparisons. These results suggest that effective label-free embedding enhancement in this setting does not require a highly structured formulation.
Related
- Audio-JEPA: Joint-Embedding Predictive Architecture for Audio Representation Learning
- Stashbird: Efficient Speaker-Indexed Memory for Conversational Agents
- SepRQ : Self-Supervised Speech Mixture Representation Learning via Mask-Free, Multi-Scale Source Separation
Source: arXiv cs.LG | 2026-10-06