MERT-v2-30s

MERT-v2-30s is a bidirectional music encoder pretrained on 30-second excerpts. Extract general-purpose music representations at the frame or recording level. Load with standard Hugging Face Transformers, using the familiar MERT workflow.

For complete recordings, use MERT-v2-FullSong, which continues pretraining on full songs.

MERT-v2-30s uses the bidirectional backbone in (B); MERT-v2-FullSong continues through the full-song branch in (C). The causal branch is used for YuE2 tokenization.

🚀 Quick start

Install the matching PyTorch packages:

python -m pip install torch==2.6.0 torchaudio==2.6.0 transformers==4.53.2 huggingface-hub safetensors soundfile

import soundfile as sf
import torch
import torchaudio.functional as AF
from transformers import AutoFeatureExtractor, AutoModel

repo = "m-a-p/MERT-v2-30s"
device = "cuda" if torch.cuda.is_available() else "cpu"
processor = AutoFeatureExtractor.from_pretrained(repo, trust_remote_code=True)
model = AutoModel.from_pretrained(repo, trust_remote_code=True).eval().to(device)

audio, sr = sf.read("music.wav", dtype="float32", always_2d=True)
audio = audio[:30 * sr].mean(axis=1)  # First 30 seconds, mixed to mono.
waveform = AF.resample(torch.from_numpy(audio), sr, processor.sampling_rate)
inputs = processor(
    waveform.numpy(), sampling_rate=processor.sampling_rate, return_tensors="pt",
).to(device)

with torch.inference_mode():
    output = model(**inputs, output_hidden_states=True)

frames = output.last_hidden_state         # [batch, frames, 1024], 25 Hz
layers = output.hidden_states             # 24 tensors, one per block
mask = output.feature_attention_mask[..., None]
embedding = (frames * mask).sum(1) / mask.sum(1).clamp_min(1)  # [batch, 1024]

hidden_states[0] is block 1; hidden_states[23] is block 24. For full songs, switch to MERT-v2-FullSong and remove the audio slice.

📊 MARBLE

Reported frozen-encoder results on MARBLE. All scores are multiplied by 100; higher is better. Bold marks the best displayed value in each column, including ties.

MERT2-30s leads on 14 of 15 metrics; MERT2-FS (full-song) leads on 13 of 15. Each count compares one encoder against the strongest external baseline on each metric; the two MERT2 encoders are not counted as baselines for each other. The model IDs below are MERT-v2-30s and MERT-v2-FullSong, respectively.

Reported values are the best observed results across representations and probe settings. Each ROC-AUC / AP pair uses the same representation.

Baseline scores and parameter counts are reproduced from Tables III and V of PupuJEPA; the baselines were not rerun for this release. Evaluation protocols may differ across sources.

Naming note: the authors renamed PupuJEPA to PupuM2D after the paper’s initial release. We retain PupuJEPA in this comparison.

General music understanding. MTT = MagnaTagATune; key = GiantSteps refined key accuracy; genre and beat = GTZAN; valence and arousal = EmoMusic. ROC = ROC-AUC; AP = average precision.

MTG-Jamendo tagging. Mood denotes mood/theme tags.

Supplemental: Chords1217 frame accuracy is 78.48 for MERT-v2-30s and 77.79 for MERT-v2-FullSong.

HF reproduction verified: Both models reproduce all ten MARBLE tasks with the fixed evaluation settings. Scores and best settings.

🎛️ Layer guide

Recommended probe settings with the encoder frozen. L1 = hidden_states[0]; All-layer MLP uses all 24 layers.

📚 Citation

Please cite the YuE2 paper when using MERT-v2-30s in your research:

@article{yuan2026yue2,
  title = {{YuE2}: Unifying Symbolic and Audio Music Generation at Frontier Quality},
  author = {Yuan, Ruibin and Pan, Jiahao and Jiang, Junyan and Wu, Zhiyue and Zhou, Ziya and Sun, Jiankai and Li, Yizhi and Zhang, Ge and Gu, Yicheng and Tian, Zeyue and Dai, Junyu and Lin, Hanfeng and Li, Kai and Wu, Shangda and Liu, Xuanjie and Wang, Jiaming and Liu, Zihan and Wang, Yue and Ma, Yinghao and Yin, Hanzhi and Chen, Kangrui and Zhang, Xinyue and Ma, Ziyang and Liao, Mengqi and Zhao, Hejia and Huang, Guowei and Yan, Chao and Ke, Lei and Yu, Jianwei and Liu, Bei and Guo, Joe and Xue, Liumeng and Xia, Gus and Xue, Wei and Guo, Yike},
  journal = {arXiv preprint arXiv:2609.33757},
  year = {2026},
  eprint = {2609.33757},
  archivePrefix = {arXiv},
  primaryClass = {eess.AS},
  url = {https://arxiv.org/abs/2609.33757}
}

For the original MERT model and the MARBLE benchmark, please also cite the relevant work:

@inproceedings{li2024mert,
  title = {MERT: Acoustic Music Understanding Model with Large-Scale Self-supervised Training},
  author = {Li, Yizhi and Yuan, Ruibin and Zhang, Ge and Ma, Yinghao and Chen, Xingran and Yin, Hanzhi and Xiao, Chenghao and Lin, Chenghua and Ragni, Anton and Benetos, Emmanouil and Gyenge, Norbert and Dannenberg, Roger and Liu, Ruibo and Chen, Wenhu and Xia, Gus and Shi, Yemin and Huang, Wenhao and Wang, Zili and Guo, Yike and Fu, Jie},
  booktitle = {International Conference on Learning Representations},
  year = {2024},
  url = {https://proceedings.iclr.cc/paper_files/paper/2024/hash/33dffa2e3d2ab74a783d1a8c292f66d9-Abstract-Conference.html}
}

@inproceedings{yuan2023marble,
  title = {MARBLE: Music Audio Representation Benchmark for Universal Evaluation},
  author = {Yuan, Ruibin and Ma, Yinghao and Li, Yizhi and Zhang, Ge and Chen, Xingran and Yin, Hanzhi and Zhuo, Le and Liu, Yiqi and Huang, Jiawen and Tian, Zeyue and Deng, Binyue and Wang, Ningzhi and Lin, Chenghua and Benetos, Emmanouil and Ragni, Anton and Gyenge, Norbert and Dannenberg, Roger and Chen, Wenhu and Xia, Gus and Xue, Wei and Liu, Si and Wang, Shi and Liu, Ruibo and Guo, Yike and Fu, Jie},
  booktitle = {Advances in Neural Information Processing Systems},
  volume = {36},
  pages = {39626--39647},
  year = {2023},
  doi = {10.52202/075280-1722},
  url = {https://proceedings.neurips.cc/paper_files/paper/2023/hash/7cbeec46f979618beafb4f46d8f39f36-Abstract-Datasets_and_Benchmarks.html}
}

Weights: CC BY-NC 4.0. Dependency notices.

YuE2 family: Song generation · Audio decoder · Benchmark decoder.