Research map / Vision-language and multimodal / Human and embodied modeling
MARBLE: Music Audio Representation Benchmark for Universal Evaluation
Ruibin Yuan, Yinghao Ma, Yizhi Li, Ge Zhang, Xingran Chen, Hanzhi Yin, Le Zhuo, Yiqi Liu, Jiawen Huang, Zeyue Tian, Binyue Deng, Ningzhi Wang and 13 more
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 |
|---|---|---|---|
| scale MAP-MERT from 95M to 330M parameters | MAP-MERT-v1 · GuitarSet chord estimation | root accuracy: 50.5 -> 56.3 | Improved |
| scale MAP-MERT from 95M to 330M parameters | MAP-MERT-v1 · MulJam lyrics transcription | WER: 77.9 -> 77.0 | Improved |
| scale MAP-MERT from 95M to 330M parameters | MAP-MERT-v1 · Jamendo lyrics transcription | WER: 82.2 -> 83.1 | Hurt |
| scale MAP-MERT from 95M to 330M parameters | MAP-MERT-v1 · MTT tagging | AP: 39.3 -> 39.5 | Improved |
| scale MAP-MERT from 95M to 330M parameters | MAP-MERT-v1 · GiantSteps key | refined accuracy: 63.5 -> 61.7 | Hurt |
| use public-data MERT-v0 instead of in-house-data MERT-v0 | MAP-MERT-v0-95M · GiantSteps key | refined accuracy: 64.1 -> 67.3 | Improved |
| use public-data MERT-v0 instead of in-house-data MERT-v0 | MAP-MERT-v0-95M · EMO emotion | R2 valence: 52.9 -> 59.1 | Improved |
| use public-data MERT-v0 instead of in-house-data MERT-v0 | MAP-MERT-v0-95M · MelodyDB melody extraction | accuracy: 60.0 -> 36.1 | Hurt |
| self-supervised Music2Vec on 1k hours instead of supervised MusiCNN | MAP-Music2Vec · NSynth pitch | accuracy: 64.1 -> 93.1 | Improved |
| self-supervised Music2Vec instead of contrastive CLMR | MAP-Music2Vec · GiantSteps key | refined accuracy: 14.8 -> 50.6 | Improved |
| larger-scale Jukebox-5B/MULE vs MAP models on tagging | Jukebox-5B · MTT tagging | AP: 38.2 -> 40.6 | Improved |
| fine-tune-free frozen features vs previous SOTA on source separation | MAP-MERT-v1-330M · MUSDB18 | SDR vocals: 9.3 -> 5.3 | Hurt |
Closest papers by meaning
- LLark: A Multimodal Instruction-Following Language Model for Music (ICML 2024 · similarity 0.88)
- CMI-RewardBench: Evaluating Music Reward Models with Compositional Multimodal Instruction (ICML 2026 · similarity 0.86)
- Aria-MIDI: A Dataset of Piano MIDI Files for Symbolic Music Modeling (ICLR 2025 · similarity 0.86)
- Deep Learning for Style Transfer and Experimentation with Audio Effects and Music Creation (AAAI 2024 · similarity 0.86)
- Massive Sound Embedding Benchmark (MSEB) (NeurIPS 2025 · similarity 0.85)
- MeLFusion: Synthesizing Music from Image and Language Cues using Diffusion Models (CVPR 2024 · similarity 0.85)
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.
