Show HN: I trained an AI model on 120M+ songs from iTunes

Possibly an unpopular opinion (which is ironic), but I think this could do with taking popularity into account. The vast majority of music out there is generic landfill at best, and just outright bad at worst. I think once you've done the first pass and got the sonically similar songs there should be some kind of filter which prioritises tracks with higher listener counts which would inject some 'wisdom of crowds' into the decision making. It would be a pretty cool feature to then be able to have this setting on a slider with obscure at one end and popular at the other.

Finding sources for input data is something I struggle with when building deep learning models. Out of curiosity, how did you go about programmatically accessing the music files for all 120M+ songs, in order to create your embedding vector? I can't imagine iTunes has an API which would let a person do that.

User tagging isn't a panacea either, because people tag inconsistently, and people who tag a lot are probably not very representative.

For an extreme example of that, see the boorus. Some machine learning people have become interested in those, since they are huge dataset of extensively tagged material ... or maybe it's the booru people who have become more interested in machine learning. Either way, I'm sure they're great, if you're into waifu anime, porn, or waifu anime porn. Both types, country AND western, as they said in the Blues Brothers movie. Any tag remotely subjective (such as "beautiful", God help you) is going to be extremely coloured by the tastes of an extreme fringe.

At least, relying on fanatics to do the work for them, I assume they've got a handle on simple spam on the boorus. Commercial recommender service tagging systems don't have that luxury, and that's probably why they end up eventually removing them.

I don't know if someone already said this, but as an amateur music producer i would love to upload my songs and discover similarities. Thanks for this Amazing tool

There might be an interesting open source / self hosting angle to this. Some folks have a large library of music stored locally. Platforms like Roon can give you recommendations on top of this, but are expensive and include a lot of other features.

You could provide discovery services to these users in exchange for model updates and feedback. Couple thoughts on this:

- there are modern techniques to update an ML model at many edge locations, then combine the learnings without violating user privacy. One common application is type-ahead models.

- People who have large local music collections tend to care about music, and would take the time to provide high quality labels for you.

- computers used as media servers often have unused compute cycles because music playback is not that intense and most folks don’t have music on 24/7. You could harness these to reduce training costs for your model

- These libraries would give you access to the long tail of the music catalog, including many things that aren’t on iTunes or other streaming services

- This would also put you in a position to run an open music catalog. Your embedding index would be a key differentiator from existing options.