Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

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

maroofy.com

Post reply on HN
Print view View original

151–160 of 444 posts

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

#151

Post by emodendroket »

mmahemoff wrote: post #149
emodendroket wrote: post #136

Earlier quoted context omitted.

If the point was to surface popular recommendations why would I go to a different Web site instead of just using what Apple Music already gives me? It has to do something different to be interesting.

Not everyone is using Apple Music (or its competition). Even for those who are using a music platform, this project still has its unique algorithm that’s likely to surface distinct results. Partly because the developer can do their own innovation and partly because the platforms influence the algorithm in ways that aren’t necessarily aligned with users’ interests. e.g. promoting artists they have favorable relationsh…

No, but the tool already features heavy integration with Apple Music, so I assumed that was the target demographic.

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

#152

Post by dbcurtis »

Didn’t do much for me. But I am an atypical listener — most of my playlists are different recordings of the same jazz standard by different artists or just same artist different era.

Probe 1: Birdland. First hit was the canonical Weather Report recording. 2nd or third was a popular Man Tran version. Then I saw a Maynard Ferguson track — ok that was a discovery as I haven’t listened to Maynard much for a few years. Didn’t like his version much, but still your software gets full points for discovery.

Probe 2: Minor Swing. First hit was obscure, and it linked off to a bunch of totally unrelated stuff. Django nowhere to be found.

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

#153

Post by zuminator »

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…

I agree with everyone's criticisms that it seems to identify similar tempo and melodic riff, irrespective of genre. But to me this is a feature, not a bug. I could see this or something like it opening my eyes to music I would never possibly have found on my own. I really like it!

Spotify on the other hand seems to want to send me to the same group of artists and tracks I've listened to before, following some Collatz conjecture type algorithm that eventually converges on the same tuned playlist for that genre, no matter what the starting parameters may be.

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

#154

Post by freetime2 »

I searched for "Poinciana" by Keith Jarrett[1] (one of my all-time favorites).

The top three responses were "La Raya" by Los Islenos [2], "Days of Our Love" by Deepa Dremata [3], and "Flying Home" by Michelle Mack [4].

While I didn't hate any of them, and they all featured a piano, I wouldn't say any of them sound like Keith Jarrett, either.

[1] https://music.apple.com/us/album/poinciana/1446740946?i=1446...

[2] https://music.apple.com/us/album/la-raya-feat-ben-murphy/154...

[3] https://music.apple.com/us/album/days-of-our-love/1608767255...

[4] https://music.apple.com/us/album/flying-home/1577716851?i=15...

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

#155

Post by anagram666 »

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.

Another strange music for your testing that gives complete bonkers recommendations: https://maroofy.com/songs/1486467186

If you need someone to test your model, you will never find one with more eclectic/strange taste than me ;)

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

#156

Post by wayeq »

japanman425 wrote: post #116
andrewmcwatters wrote: post #39

Tried Erreur 404 by L'Impératrice[1], and I noticed the beat of the other recommended songs were eerily similar!! I'd argue your project is actually too good. Where Spotify's Discover Weekly tried to connect you to music other people listen to {B, C, D, ...} because you've listened to a particular song {A} {B...->A}, your model quite literally tried to find other music {A₂} that sounds like what you're looking for {A…

… what?

{W₁} {H₂->A₁}*T?

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

#157

Post by gradascent »

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…

I've only tried a few songs but they've mostly been bangers! I did come across a couple examples where the recommended songs just heavily sampled the original but overall very impressed.

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

#159

Post by phist_mcgee »

popinman322 wrote: post #119
japanman425 wrote: post #116

Earlier quoted context omitted.

… what?

IIUC: Spotify seems to use something similar to collaborative filtering. People rate songs based on more than similarity, which is what this model seems to provide.

>IIUC

What?

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

#160

Post by RayVR »

I think it’s an interesting idea and look forward to future work, but I don’t think the current results are much better than random.

I don’t really understand the features your model has learned but I’d start by trying to understand those features.

Music is highly structured and reasonably well organized already. Understanding if your model learned any features that map to metrics used by humans is an important question. If the answer is “no” then I’d wonder if it’s actually achieving anything beyond random number generation.

—-edit—- I went back and tried modern pop. Which I believe it has learned better, although all of the music there sounds so similar maybe that is just a result of the small sample I tried.

Post reply on HN