Research map / Label-efficient and robust learning / Recognition and long-tail
Disentangling Voice and Content with Self-Supervision for Speaker Recognition
Tianchi Liu, Kong Aik Lee, Qiongqiong Wang, Haizhou Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace TSP with Xi aggregation | tResNet34 · VoxCeleb1/SITW | EER and minDCF: 5.02%/4.19% average reductions | Improved |
| replace Xi with RecXi | tResNet34 · VoxCeleb1/SITW | EER and minDCF: 6.96%/5.82% average reductions | Improved |
| replace Xi with RecXi | ECAPA-TDNN · VoxCeleb1/SITW | EER and minDCF: 12.15%/10.66% | Improved |
| replace Chan.&Con. with Xi | ECAPA-TDNN · VoxCeleb1-O | EER: 1.377 -> 1.340 | Mixed |
| replace ResNet34 with tResNet34 | TSP aggregation · VoxCeleb1-H | EER: 2.500 -> 2.257 | Improved |
| add self-supervised speaker-preserving loss L_ssp (ablation) | RecXi · VoxCeleb1/SITW | EER and minDCF: 5.07%/5.44% average reductions | Improved |
| add L_ssp (ablation) | RecXi(phi, phi_lin) ECAPA · SITW eval | EER: 2.351 -> 2.105 | Improved |
| use phi only instead of phi and phi_lin, without L_ssp (ablation) | RecXi tResNet · VoxCeleb1-H | EER: 2.130 -> 2.185 | Hurt |
| use phi only instead of phi and phi_lin, with L_ssp (ablation) | RecXi tResNet · VoxCeleb1-H | EER: 2.097 -> 2.117 | No clear effect |
| use MSE loss instead of similarity-preserving loss for L_ssp (ablation) | RecXi ECAPA · SITW eval | EER: 2.105 -> 2.292 | Hurt |
| MSE L_ssp vs no L_ssp (ablation) | RecXi ECAPA · SITW eval | EER: 2.351 -> 2.292 | Improved |
| reduce transition matrices from 16 to 1 (ablation) | RecXi · SITW eval | EER: 1.832 -> 3.882 | Hurt |
Closest papers by meaning
- Self-Supervised Disentangled Representation Learning for Robust Target Speech Extraction (AAAI 2024 · similarity 0.90)
- SelfVC: Voice Conversion With Iterative Refinement using Self Transformations (ICML 2024 · similarity 0.86)
- Enhancing Zero-Shot Multi-Speaker TTS with Negated Speaker Representations (AAAI 2024 · similarity 0.85)
- Can Diffusion Models Disentangle? A Theoretical Perspective (NeurIPS 2025 · similarity 0.85)
- DDDM-VC: Decoupled Denoising Diffusion Models with Disentangled Representation and Prior Mixup for Verified Robust Voice Conversion (AAAI 2024 · similarity 0.84)
- Let There Be Sound: Reconstructing High Quality Speech from Silent Videos (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
