Interesting. Some feedback
1. Please add match score
2. Group and fold duplicates
3. Add the year with the sort feature - to identify rip offs
Show HN: I trained an AI model on 120M+ songs from iTunes
71–80 of 444 posts
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#72Interesting, I put in a bunch of my favourite songs and found nothing I liked. The songs seem to be over matching on drum beat and tempo so there is a similarity to the song I suggested and the matches but it's often superficial.
For instance I picked a song with a very strong snare drum line. All the suggestions also had a strong snare drum line but wildly different melodies, genre's, tone, etc.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#73this is pretty awesome.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#74How did you scrape the audio of 120M songs? That sounds expensive?
Probably just used the previews (makes sense, songs tend to keep their rhythm throughout).
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#75I tried a few of my favorites and the results were... predictable, I guess.
It kind of matches genre, but has no grasp of musicianship or why I might like a track.
And this is a fundamental problem. See Avery Pennarun's brilliant explanation of tracking users, data analysis, and recommendation systems to understand why:
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#76How the hell do you even get access to the entire iTunes catalog?
um, isn't that just the Apple Music paid subscription service, their Spotify competitor? They only advertise ("over") 100M songs though, I'm not sure where the extra 20M come from.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#77Good work! I tried some house tunes. I guess the results were similar but expected something more direct.. a suggestion; since some listens only includes the intro - it would be a good idea to match the first 5-10 seconds in harmonics; probably in the span of 90-180 hz. But I guess that would take a whole lot of new work.. So, anyways - if there's a way to jump in a bit into the song I think it might be beneficial. But after trying a bit more I realized simply that the quality of good productions was more of an issue.
Great work anyways!
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#78This is great! It works quite well. I would love to see something like this with the ability to use license the music for a Youtube video so that you could get good music for a video, and exposure and money to lesser known artists.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#79This 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…
Same here
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#80If you could somehow filter the beginner bedroom producers the results might be even more compelling...but great job so far!