Pool analysis together

[nottheboss]

Do you think it would be a bad idea to have a common database for people to contribute songs?
I would imagine relying on the MusicBrainz id by track would be an option and it would reduce the load of having to scan every song in the library for everybody.
Of course it wouldn't be 100% efficient but I would be happy to try to work on something of the sort if that would be fine with you

[NeptuneHub]

The risk is violating some copyright, so as soon as you analyze your song locally and you get your responsability for what you analyze is one thing. When you pass to collect things form other user you start to get the responsability for what other people submit and can become risky. So even if technically is not difficult, I want to avoid risk.

I was thinking of the possibility of collecting similarity, more than embbeding. This not to avoid analysis but to have raccomandation of new song, but I'm still thinking about it.

[nottheboss]

What about if people would analyze songs locally and would only submit the information such as

- Tempo: 156.25, Energy: 0.1590, Key: D minor
- Top Moods: {'rock': 0.5847867727279663, 'indie': 0.5330650210380554, 'alternative': 0.5296273827552795, 'pop': 0.5270209908485413, 'electronic': 0.5183314681053162}
- Other Features: danceable:0.20,aggressive:0.17,happy:0.50,party:0.15,relaxed:0.42,sad:0.58

together with a MusicBrainz id such as https://musicbrainz.org/recording/2e5029b7-b3e1-4e35-b419-b56e1419edac ?

That would relieve you of having to store any copyrighted material and the information would be collective so anyone could quickly fetch this info on their own PC.

[NeptuneHub]

Those information are not enough, for the actual analysis the entire embbeding vector is needed. And actually AudioMuse-AI use two different one: Musicnn and CLAP (I’m working to try to use only one, but still work in progress).

Share only song 1 similar to song 2, 3.. 10 could be an interesting alternative for reccomandation system. It can also go over the possible issue of different value from different architecture or model that change in the time.