AI Paper Atlas

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

NeurIPS 2023 · Vision-language and multimodal › Human and embodied modeling · approach: Layout-aware document understanding · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
scale MAP-MERT from 95M to 330M parametersMAP-MERT-v1 · GuitarSet chord estimationroot accuracy: 50.5 -> 56.3Improved
scale MAP-MERT from 95M to 330M parametersMAP-MERT-v1 · MulJam lyrics transcriptionWER: 77.9 -> 77.0Improved
scale MAP-MERT from 95M to 330M parametersMAP-MERT-v1 · Jamendo lyrics transcriptionWER: 82.2 -> 83.1Hurt
scale MAP-MERT from 95M to 330M parametersMAP-MERT-v1 · MTT taggingAP: 39.3 -> 39.5Improved
scale MAP-MERT from 95M to 330M parametersMAP-MERT-v1 · GiantSteps keyrefined accuracy: 63.5 -> 61.7Hurt
use public-data MERT-v0 instead of in-house-data MERT-v0MAP-MERT-v0-95M · GiantSteps keyrefined accuracy: 64.1 -> 67.3Improved
use public-data MERT-v0 instead of in-house-data MERT-v0MAP-MERT-v0-95M · EMO emotionR2 valence: 52.9 -> 59.1Improved
use public-data MERT-v0 instead of in-house-data MERT-v0MAP-MERT-v0-95M · MelodyDB melody extractionaccuracy: 60.0 -> 36.1Hurt
self-supervised Music2Vec on 1k hours instead of supervised MusiCNNMAP-Music2Vec · NSynth pitchaccuracy: 64.1 -> 93.1Improved
self-supervised Music2Vec instead of contrastive CLMRMAP-Music2Vec · GiantSteps keyrefined accuracy: 14.8 -> 50.6Improved
larger-scale Jukebox-5B/MULE vs MAP models on taggingJukebox-5B · MTT taggingAP: 38.2 -> 40.6Improved
fine-tune-free frozen features vs previous SOTA on source separationMAP-MERT-v1-330M · MUSDB18SDR vocals: 9.3 -> 5.3Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy