Research map / Label-efficient and robust learning / Recognition and long-tail
Learning from Between-class Examples for Deep Sound Recognition
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply BC learning to EnvNet-v2 with strong augmentation | EnvNet-v2 · ESC-50 | error rate (%): 21.2 ±0.3 -> 15.1 ± 0.2 | Helped |
| Apply BC learning to EnvNet-v2 with strong augmentation | EnvNet-v2 · ESC-10 | error rate (%): 10.9 ±0.6 -> 8.6 ±0.1 | Helped |
| Apply BC learning to EnvNet-v2 with strong augmentation | EnvNet-v2 · UrbanSound8K | error rate (%): 24.9 -> 21.7 | Helped |
| Apply BC learning | EnvNet · ESC-50 | error rate (%): 29.2 ±0.1 -> 24.1 ±0.2 | Helped |
| Apply BC learning | EnvNet · ESC-10 | error rate (%): 12.8 ±0.4 -> 11.3 ±0.6 | Helped |
| Apply BC learning | EnvNet · UrbanSound8K | error rate (%): 33.7 -> 28.9 | Helped |
| Use A-weighted level-adjusted mixing rather than unadjusted mixing (ablation) | BC learning with EnvNet · ESC-50 | error rate (%): 26.8 ± 0.1 -> 24.1 ±0.2 | Helped |
| Use A-weighted levels rather than RMS levels for sound mixing (ablation) | BC learning with EnvNet · ESC-50 | error rate (%): 26.5 ± 0.2 -> 24.1 ±0.2 | Helped |
| Use the ratio target instead of a single dominant-class label (ablation) | BC learning with EnvNet · ESC-50 | error rate (%): 26.5 ±0.2 -> 24.1 ±0.2 | Helped |
| Use the ratio target instead of a multi-label target (ablation) | BC learning with EnvNet · ESC-50 | error rate (%): 25.0 ±0.3 -> 24.1 ±0.2 | Helped |
| Mix two examples from different classes rather than two from the same class (ablation) | BC learning with EnvNet · ESC-50 | error rate (%): 27.3 ±0.2 -> 24.1 ±0.2 | Helped |
| Mix two different-class sounds rather than sometimes sampling same-class pairs (ablation) | BC learning with EnvNet · ESC-50 | error rate (%): 24.8 ±0.3 -> 24.1 ±0.2 | Helped |
Closest papers by meaning
- Between-Class Learning for Image Classification (CVPR 2018 · similarity 0.92)
- Independency Adversarial Learning for Cross-Modal Sound Separation (AAAI 2024 · similarity 0.82)
- Contrastive Learning from Synthetic Audio Doppelgängers (ICLR 2025 · similarity 0.81)
- SoundCount: Sound Counting from Raw Audio with Dyadic Decomposition Neural Network (AAAI 2024 · similarity 0.81)
- Weakly-supervised Audio Separation via Bi-modal Semantic Similarity (ICLR 2024 · similarity 0.81)
- AND: Audio Network Dissection for Interpreting Deep Acoustic Models (ICML 2024 · similarity 0.81)
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.