On Real Song Downloading.

[HildaNya]

Hi SONICS team,
Thank you for the inspiring work and the detailed writing. We would like to use the SONICS dataset in our research on AI-generated music. However, downloading the YouTube videos for the “real songs” portion has become a significant bottleneck.
We understand that you cannot distribute the real audio due to copyright constraints. That said, could you kindly point us to any more efficient strategies for retrieval (e.g., recommended tools/configs, batching practices, or any workflow you found reliable)?
We are using the youtube_id column in real_songs.csv and running yt_dlp to download each ID. We are encountering two main issues: A non-trivial fraction of IDs return “unavailable” or restricted audio. For available IDs, downloads become extremely slow when performed at scale.
We greatly appreciate any advice. Thank you again for making the SONICS dataset available to the community!

[JasonBourne1998]

Hi @HildaNya , I encountered the same issue, may I ask if you have solved this problem ?

[awsaf49]

Thanks for taking interests on our project. Sadly due to copyright issue we are unable to distribute the real songs. We understand some of the songs might be missing sadly we don't know what to do in this situation, maybe try to look if there's any copy of song maybe with different id? In case that helps we used https://github.com/deepjyoti30/ytmdl to download our songs.

[HildaNya]

Hi @HildaNya , I encountered the same issue, may I ask if you have solved this problem ?

Nope, still trying. Will share here if I find a solution.

[HildaNya]

Thanks for taking interests on our project. Sadly due to copyright issue we are unable to distribute the real songs. We understand some of the songs might be missing sadly we don't know what to do in this situation, maybe try to look if there's any copy of song maybe with different id? In case that helps we used https://github.com/deepjyoti30/ytmdl to download our songs.

Thank you so much for the advice!

[SalmanQureshi97]

#8

I added a script I am using to download the real_songs, seems to be working just fine. Use pip install yt-dlp and run the script in the sonics huggingface directory.