Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Repeatable Speech Embeddings Using An Intra-class Correlation Regularizer
Jianwei Zhang, Suren Jayasuriya, Visar Berisha
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 |
|---|---|---|---|
| add ICC regularizer to GE2E loss (ablation) | VGG-M-40 · VoxCeleb1 TI-SV | EER: 4.39% -> 3.96% | Improved |
| add ICC regularizer to GE2E loss (ablation) | FastResNet-34 · VoxCeleb1 TI-SV | EER: 2.49% -> 2.39% | Improved |
| add ICC regularizer to AngleProto loss (ablation) | VGG-M-40 · VoxCeleb1 TI-SV | EER: 4.36% -> 4.02% | Improved |
| add ICC regularizer to AngleProto loss (ablation) | FastResNet-34 · VoxCeleb1 TI-SV | EER: 2.28% -> 2.17% | Improved |
| add ICC regularizer to SupCon loss (ablation) | VGG-M-40 · VoxCeleb1 TI-SV | EER: 3.91% -> 3.78% | Improved |
| add ICC regularizer to GE2E loss (ablation) | VGG-M-40 · VoxCeleb1 TI-SV | ICC: 0.4494 -> 0.5487 | Improved |
| add ICC regularizer to GE2E loss (ablation) | FastResNet-34 · VoxCeleb1 TI-SV | ICC: 0.7215 -> 0.7366 | Improved |
| add ICC regularizer to SupCon loss (ablation) | VGG-M-40 · VoxCeleb1 TI-SV | minDCF: 0.2791 -> 0.2597 | Improved |
| add ICC regularizer to AngleProto loss (ablation) | VGG-M-40 · VoxCeleb1 TI-SV | minDCF: 0.2809 -> 0.2790 | No clear effect |
| train AutoVC with GE2E+ICC speaker embeddings | AutoVC · VCTK zero-shot conversion AB test | selection ratio: 38.10% -> 61.90% | Improved |
| train AutoVC with GE2E+ICC speaker embeddings | AutoVC · VCTK zero-shot conversion | WER: 0.5810 -> 0.5109 | Improved |
| train AutoVC with GE2E+ICC speaker embeddings | AutoVC · VCTK zero-shot conversion | speaker similarity: 0.2231 -> 0.2309 | Improved |
Closest papers by meaning
- Self-Supervised Disentangled Representation Learning for Robust Target Speech Extraction (AAAI 2024 · similarity 0.84)
- SimReg: Achieving Higher Performance in the Pretraining via Embedding Similarity Regularization (NeurIPS 2026 · similarity 0.84)
- Learning from Teaching Regularization: Generalizable Correlations Should be Easy to Imitate (NeurIPS 2024 · similarity 0.83)
- Manifold-Based Verbalizer Space Re-embedding for Tuning-Free Prompt-Based Classification (AAAI 2024 · similarity 0.83)
- Unified Speech Recognition: A Single Model for Auditory, Visual, and Audiovisual Inputs (NeurIPS 2024 · similarity 0.83)
- Large Language Models are Efficient Learners of Noise-Robust Speech Recognition (ICLR 2024 · similarity 0.83)
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.
