Ask for Correspondence Between Folders and TTM Models

[ciwei6107563]

We are truly delighted to see such an outstanding dataset from you. However, we have some questions regarding the details of the dataset. After downloading it, we noticed that there are multiple folders under the FakeMusicCaps/FakeMusicCaps directory. We observed that your dataset contains music data generated using five different TTM models. Could you kindly clarify which model each folder corresponds to? Are all the data synthetic, or does any of it include real data? We look forward to your reply and greatly appreciate your assistance. Thank you very much!

[lucacoma]

Thank you for the kind words! You are right, each folder was generated using a different text-to-music model.
Full details are available on our preprint paper here https://arxiv.org/pdf/2409.10684.
To summarize, we used the following models:

• MusicGen https://proceedings.neurips.cc/paper_files/paper/2023/file/94b472a1842cd7c56dcb125fb2765fbd-Paper-Conference.pdf
• AudioLDM2 https://ieeexplore.ieee.org/iel7/6570655/10304349/10530074.pdf
• MusicLDM https://ieeexplore.ieee.org/iel7/10445798/10445803/10447265.pdf
• Stable Audio Open https://arxiv.org/pdf/2407.14358
• Mustango https://arxiv.org/pdf/2311.08355

All data is synthetic. All the subfolders were generated using the captions from the MusicCaps (which actually contains real music https://arxiv.org/pdf/2301.11325) dataset and using them as input to the mentioned text-to-music models.

Please let me know if you have further questions and if further explanations are needed!