13

I am mostly interested in electronic songs. I need both the songs' waveform and their labels (genre, BPM and key). If you know some similar data set that doesn't contain all 3 labels, please still share.

If possible, I want open data and downloading the whole dataset at once.

  • Do you want open data or is commercial OK? Is a web service OK, or do you want to download the whole dataset? – Nicolas Raoul Jun 26 '14 at 5:45
  • @NicolasRaoul if possible, open data and downloading the whole dataset. – Franck Dernoncourt Jun 26 '14 at 5:54
9

I haven't used them before. I've just had them in my bookmarks for any case.

1) http://echoprint.me/

Free: 'Use our data for whatever you want (commercial or non, research, personal use)'

2) http://the.echonest.com/

Free for non-commercial: The Echo Nest APIs are free for non-commercial purposes. To use them commercially, contact us and we will go over licensing options.

I think that both of them are services to identify a song like Shazam does. But they let you download their data or use their API.

  • First link is now dead :( And I can't see where to get access to the API on Echonest... – naught101 Mar 31 '17 at 1:12
5

columbia has a music db here: http://labrosa.ee.columbia.edu/millionsong/pages/example-track-description
also check last.fm and creative commons music communities:
http://creativecommons.org/music-communities

  • 1
    Last.FM's genres would be sooo useful, but their API broke years ago, and they haven't fixed it, or given any indication that they will :( Pretty crap, too, because it's all community-supplied data.. – naught101 Mar 31 '17 at 1:44
  • probably not too hard to scrape – albert Mar 31 '17 at 13:11
3

AcousticBrainz is a database of algorithmically derived music metadata, including BPM, Key, moods, and genres. The database is keyed by acoustIDs, and you can find the acoustIDs for a particular track by searching for it on MusicBrainz.org.

Example: Tipper - Bubble Control.

Most attributes have uncertainties associated with them, and it's worth noting that the algorithms involved are not fool proof. For example, the above track is tagged Voice (92%), even though it does not have vocals (it has lots of weird synthesised noises that sound similar to vocals though). It is also tagged Jazz (100%), and though it does share many attributes with jazz, I doubt many people would actually classify it as jazz. Also, it's listed as 100% not danceable, which I can personally attest is not true :P

The data shown on that page is only a subset of what is available through the API.

All data is public domain (and if a song doesn't exist, you can easily submit it yourself with scripts listed on the contributing page), and is accessible via JSON, via their API (no API key required).

You can download the entire database at https://acousticbrainz.org/download, but it's 2.5Gb for just the high-level data, 34Gb for the full dataset (there are multiple fingerprints for each song).

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.