Live data from Hacker News

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

maroofy.com

131–140 of 444 posts

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

#131
https://maroofy.com/songs/500162235

Your AI is processing the tempo and beat of the music, its not looking at the lyrics and what the lyrics mean, its not processing any accompanying music videos and its not really factoring in emotion, the mood and events of the day in the locale it comes from.

Another factor, is its not taking into account the limitation instruments had on music before the introduction of the synthesizer and the tech we have available today in order to make music.

There are other websites which also make suggestions and doesn't factor in the above points, so your website is not alone in not really grasping the mood of the song. I'm sure its a limitation of the AI models available to people like yourself today, along with time constraints and other factors including access to data. Just how far do you read into things?

When you see an AI suggest the likes of the Sex Pistol's God Save the Queen with this song, you are then looking at an AI which is intelligent. Quantify the mood of the music, the mood of the country or region during that period of time. Thats a very hard thing to do, and masters of that would be the BBC, simply because as national broadcasters, an extension of the British Military aka MI7, arguably global broadcasters with their world service, you'll realise there are songs which just wont be played when certain events are happening on the ground in regions which would be featured in the news and thus considered inciteful. Its very subtle how manipulation at a national or global level takes place, but it takes place.

However it is also a very difficult song to quantify, as this youtube reaction video sums up nicely. https://youtu.be/5vK1FZO1ybE?t=296

If you know anything about the band, they span many genres, and yet they also their own genre. Truly Prodigious.

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

#132

Seems like the recommendations are somewhat hit or miss so far. I would suggest finding a way to say why it recommended a particular song. I know that this is already common, but maybe you could come up with a new spin on it. Also, other comments have suggested that this was trained on 120MM of the audio previews instead of the full songs. That might explain why the recommendations seem a little off for some people.

I'm totally going out on a limb and guessing here - I'm more on the UI/UX side of things so I know nothing low-level about what goes into a building recommendation engine, algorithm, or whatever. But I do know music pretty well, and this feels like it's matching too closely to the technical aspects of the music and not the overall theme/je-ne-sais-quoi that makes a song something you feel.

I tried two of my all-time favorite songs:

The Gaslight Anthem - Handwritten

Thrice - The Artist in the Ambulance

On the first, I would say it was a total swing and miss - the recommendations had similar elements to Handwritten like strummed power chords, but the vibe was completely off. With Thrice, same thing. The first match was a remaster of the song itself by the band, so of course despite being a "different" track it was the closest possible match. The rest of the recommendations had a similar tempo and more heavy metal riffs (as are common in Thrice's songs), but none of them were songs I would voluntarily listen to.

This is a cool idea and I hope OP if you read this you take the criticsm as it's intended - I'm not trashing the implementation or anything. This just feels like it truly is a robot that can't feel the emotion of a song making me recommendations.

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

#133

Can you do some deduping? A lot of the matches are just the same song appearing on multiple albums or single vs. album version.

Came with the same feedback, I'm having fun playing with it, but it's amusing to search for "Comfortably Numb" and the first match is "Comfortably Numb" :D

https://maroofy.com/songs/1065976170

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

#134
Other people have called out limitations and I don't disagree, but I'll say that I do find the different approach, based entirely on sounding similar, interesting. Recommendation engines in music streaming services often tell me about related things I already know I like. That's not necessarily unwelcome (if I'm listening to one song that appears in Footloose, sure, it's not a bad surmise that I might like to hear another), but this tool surfaces a bunch of stuff I would have never thought to look for or found normally.

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

#135
Bad lyrics ruin songs for me, so I tend to listen to music in a way where I completely disregard the lyrics if they're bad, which causes me to like songs where the lyrics are hard to understand/hear or in a different language.

This seems to absolutely NAIL my case of listening, I'm loving this. Thank you!

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

#136

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…

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.

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

#139
Interesting, I tried a few songs and artist combinations:

I picked a Three 6 Mafia song I used to listen to in my younger years, the particular song I chose has everything in the soundtrack from high hats, to trumpets, several rappers with varying styles of exerting their voices, and a lot of the resultant songs have either high hats or similar sounding trumpets, or similar sounding drums, and in other cases, something about how the artists rap reminds me slightly of Three 6 Mafia's own style. I don't recognize any of the artists suggested, which is a good thing, it basically did what it was supposed to.

Next I picked a song by Bullet For My Valentine, not all the songs are quite nu-metal but that's because the song I chose isn't typical nu-metal (or whatever genre they were...) unlike some other metal bands they dont spend the whole song screaming, they actually do normal singing, so it picked up what to me sound like old school metal bands like Dio, Iron Maiden and so on... So it picks up on how some parts of the song are, which can get you mixed results, so if you want to find a similar artist down to the musical style, pick the most "metal" song you can find by them. This isn't an issue for me since I like some of those artists too, but I would rather get more bands that are more like Bullet For My Valentine, though it seemingly did catch one or two I had not heard of before.

Also tried another metal band Killswitch Engage with the song "My Curse" and it recommended another Killswitch Engage song titled "For You" which honestly, is weirdly close in style and I never noticed that before. Again this one's another metal band, I just wanted to see how it would do.

Lastly, I tried a Spanish reggaeton song, and I was not disappointed. You really did a hell of a job on this.

My only wish is that Apple would hire you / buy out your efforts because I wish this was part of Apple Music, their "suggested music" is awful, and the genre discovery type of thing doesn't always yield artists I remotely care for, I feel more inclined to listen to some of the suggested songs but it sucks you can only give samples and link to the song directly, if you could generate playlists from the results it would be good, this isn't a you issue, its moreso an Apple limitation.

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

#140

https://maroofy.com/songs/500162235 Your AI is processing the tempo and beat of the music, its not looking at the lyrics and what the lyrics mean, its not processing any accompanying music videos and its not really factoring in emotion, the mood and events of the day in the locale it comes from. Another factor, is its not taking into account the limitation instruments had on music before the introduction of the synth…

> Your AI is processing the tempo and beat of the music,

I don’t know about that, I tried a number of songs with 3/4 and 6/8 measure, and none of the suggestions were, so it doesn’t seem to pick up on that.

Post reply on HN