Show HN: I trained an AI model on 120M+ songs from iTunes • Hacker News Board

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

#51

Post by knaik94 » Fri, Feb 03, 2023, 2:07 AM UTC

mmahemoff wrote: post #35

Exciting to see AI used this way. My main feedback is I'd look at incorporating other factors to rank results, not purely how similar it sounds. Audiophiles might prefer a pure similarity ranking, but that could be offered as a non-default setting if anything. e.g. I'm sometimes seeing several essentially identical tracks at the top of recommendations (also mentioned in a comment by rayshan). You probably want to pen…

I think popularity based ranking should absolutely be optional and a toggle. I think it's reasonable to have default rank be popularity, but in my opinion, the value of a model like this is finding obscure tracks. I also would definitely like seeing an exposed toggle, and maybe automatically toggle it off when someone presses refresh?

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

#52

Post by charcircuit » Fri, Feb 03, 2023, 2:09 AM UTC

snow_mac wrote: post #49

How did you get access to 120,000,000 songs from iTunes? Not just the listing, but the actual audio.

The preview audio is free. It may just be 120,000,000 previews.

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

#53

Post by yborg » Fri, Feb 03, 2023, 2:10 AM UTC

Laughably bad for "Here's Where The Story Ends" by The Sundays, I can't even begin to see how some of these recommendations relate. Did discover that Finnish rockabilly is a thing, though.

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

#54

Post by runnerup » Fri, Feb 03, 2023, 2:12 AM UTC

subtech wrote: post #33

knaik94 wrote: post #24

This is very interesting, but unfortunately I haven't had the greatest luck in finding new songs I would enjoy listening to. It absolutely finds similar sounding tracks, but it doesn't distinguish which part of the song made it enjoyable. There's no tempo consistency or genre consistency or even main instrument/vocal timbre consistency between recommendations. I think locking one or more of those dimensions would all…

Hey thanks for the feedback! I definitely have a lot of improvement to do on the model, it currently performs better for some styles/genres of music than others. But the model architecture I'm using is kinda outdated as well, gotta iterate on it more to improve it further! I'm also thinking of letting users upvote/downvote results, which can also help improve quality on the ranking side.

Honestly it's loads better than current Spotify/YouTube Music suggestions. Mostly they just seem to suggest popular stuff that's heavily marketed...even though I seeded all my "thumbs up" with only eclectic stuff.

Yes, it's hard to find a song I really really like, but 1-in-10 seem to be something I'd add to my eclectic "thumbs up" playlist. And almost none of them are by any artist that I've heard of before.

This is huge for me. Thanks.

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

#55

Post by shusaku » Fri, Feb 03, 2023, 2:13 AM UTC

Great work, this is very fun! Alas, I tried “Werewolves of London” and it didn’t spit out “Sweet Home Alabama”, so it still has room to grow

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

#56

Post by rcme » Fri, Feb 03, 2023, 2:13 AM UTC

The results do sound very similar, which is interesting from a philosophical point of view: after searching all my favorite songs, I liked not one similar sounding result. What makes one enjoy art vs. not? There's clearly more to it than how something "sounds."

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

#57

Post by kennyloginz » Fri, Feb 03, 2023, 2:13 AM UTC

This is great! So handy for the dj! Please keep up the good work, very appreciated!

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

#58

Post by gticala » Fri, Feb 03, 2023, 2:14 AM UTC

I find the list of similar songs to be bang on for a couple of songs I queried for. Great job!

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

#59

Post by gticala » Fri, Feb 03, 2023, 2:15 AM UTC

I found the recommended songs to be bang on similar to the ones I queried. Great job!

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

#60

Post by tastysandwich » Fri, Feb 03, 2023, 2:15 AM UTC

Personally for me the best way of finding new music is friends whose opinions I trust. I find going by similarity will pull up a lot of derivative artists that suck (eg Last.Fm/Pandora). I want something good, and if it's radically different, even better!

One awesome use-case I can see for this though is finding alternatives to copyrighted songs. Let's say you make a sports video, and you have this fantastic song in your head, but you can't secure permission. It would be cool if this could find something similar to that song. Same style, tempo, etc. Even better if it's royalty free.