Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
MusicDET: Zero-Shot AI-Generated Music Detection
Chaolei Han, Hongsong Wang, Jie Gui
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 |
|---|---|---|---|
| train a one-class flow on real music only (zero-shot) instead of discriminative detectors trained on one generator | MusicDET · FakeMusicCaps cross-generator | EER (%): 11.46 -> 4.51 | Improved |
| add class-conditional prior with generated samples | MusicDET · FakeMusicCaps cross-generator | EER (%): 4.51 -> 0.89 | Improved |
| add class-conditional prior with generated samples | MusicDET · SONICS cross-generator | EER (%): 2.89 -> 0.00 | Improved |
| zero-shot MusicDET vs spectrogram baseline Spec-ViT | MusicDET · SONICS cross-generator | EER (%): 1.23 -> 2.89 | Hurt |
| use multiple frequency bands instead of a single band (ablation) | MusicDET · FakeMusicCaps | EER | Improved |
| increase prior mean mu_real (ablation) | one-class MusicDET · FakeMusicCaps | EER | Improved |
| fully fine-tune self-supervised front end | W2V2-AASIST · FakeMusicCaps | EER (%): 25.80 -> 11.46 | Improved |
| fully fine-tune self-supervised front end | MERT-AASIST · SONICS | EER (%): 18.98 -> 28.89 | Hurt |
| hold out jazz or piano subdomain from training (ablation) | MusicDET · FakeMusicCaps | EER (%): jazz 2.5, piano 4.1 | No clear effect |
| apply pitch shifting to test audio (ablation) | MusicDET · FakeMusicCaps robustness | EER (%): +40.22 (to 44.73) | Hurt |
| apply white noise to test audio (ablation) | MusicDET · FakeMusicCaps robustness | EER (%): +39.6 (to 44.11) | Hurt |
| apply time stretching to test audio (ablation) | MusicDET · FakeMusicCaps robustness | EER (%): 2.44 (↓2.07) | No clear effect |
Closest papers by meaning
- SONICS: Synthetic Or Not - Identifying Counterfeit Songs (ICLR 2025 · similarity 0.87)
- Transcoda: End-to-End Zero-Shot Optical Music Recognition via Data-Centric Synthetic Training (NeurIPS 2026 · similarity 0.86)
- Detecting Music Performance Errors with Transformers (AAAI 2025 · similarity 0.85)
- Zero-Shot Unsupervised and Text-Based Audio Editing Using DDPM Inversion (ICML 2024 · similarity 0.85)
- MGE-LDM: Joint Latent Diffusion for Simultaneous Music Generation and Source Extraction (NeurIPS 2025 · similarity 0.84)
- MIDI-GPT: A Controllable Generative Model for Computer-Assisted Multitrack Music Composition (AAAI 2025 · 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.
