Music discovery is a perennial AI tech application. The first one I followed was Firefly, out of MIT Media Lab around 1995. I think the startup was originally called 'Agents', which was a hot term in AI at the time. Thirty years ago.
To me it always feels like music classification and recommendation efforts, open source and commercial, are too focused on music distributed in albums and singles. In the long history of human music, albums have only been a fleeting moment, an App Store of music if you will. This would never recommend me concert performances (even those on YouTube), covers on YouTube, DJ sets on SoundCloud, or game soundtrack without an official release. Listenbrainz is the closest I've found because it just has Title and Artist, but even that can be fuzzy for covers and live performances. Maybe no one other than Google can build this right now.
Music recommendation is such a hard problem. There are all these seemingly obvious relationships you can map between bands to create a big graph that looks good but that almost never captures what goes on when a human with deep music knowledge recommends music. Often the best recommendations have no obvious relationships to the band you like.
I played around with this tool a bit and didn't find it any better then other more traditional music discovery tools, that is to say not very effective.
For example, I entered John Zorn and was recommended a bunch of John Zorn's bands. I entered The Residents and got The Pixies :/
I think its more effective to click around on Music Brainz and Wikipedia.
You seem knowledgeable about this.. Care to test my old project for music recommendation? I built it by looking at co-occurrence of artists in Spotify playlists, which gives me word2vec-style vectors, and then its just kNN.
No login needed, just enter some artist names and see what you get:
Very interesting, I've been working on a similar project (using word2vec to learn vectors using playlist data), but using songs instead of artists as the 'words'.
The main bottleneck at this point is the volume of data - many songs I'm interested in only are only represented in a handful of playlists, and . Evaluation at any useful scale is also quite difficult. For somewhat obvious reasons, in our AI era Spotify has become quite skittish about letting third parties gain access to their data at scale...
Interesting. Spotify works almost perfectly for my discovery needs. I just pick a track I know that fits my mood, then use the (3-dot menu) "Go to Radio" option, which leads to a playlist that usually includes tracks and/or artists new to me. It's been a reliable discovery mechanism for me for many years. Also, there's a new feature I first saw within the last week, a "non-personalized" option that increases the "new to me" ratio.
I always find this interesting… Spotify is phenomenal for me - about every third Monday Discovery playlist has two or three hits, which feels like a pretty solid ratio, at this point. YouTube has never suggested a single thing I cared for.
I wonder if it’s a curation thing? I’ve been with Spotify since the first day it was available, and rarely use YouTube. I haven’t had a good music ratio as good since newsgroups and (real) forums a decade ago, which were a different form of curation.
Weird. It hallucinated one for me, and recommended an album that doesn't exist. Flashbacks of 2023. If this is your app, you might want to consider adding a validation layer that performs a review before publishing an output.
When I put in a specific song by an artist that's not similar to the artist's typical output, it completely ignores the specific song.
Which is not surprising considering it's run by an LLM, but makes it not very useful as a music recommendation engine. There's already tons of places to ask "what artists are like this other artist"?
Thank you for this - I've just discovered something I'll be listening to all week. Is there any chance I could prompt it again with the bands I liked and didn't like from the list, possibly asking for something more refined?
Cool app. One small complaint is the chatty tone of the recommender engine. In particular, I find it a bit disingenous to have an LLM tell me "Ah, I love <X>!".
EDIT: I also notice the recommendations are totally different when making the same query in a different session. I'm not sure if that's intentional? I expected at least some overlap with the previous time I asked.
As someone who's big into UK Bass who finds new music mainly through a mix of Spotify, Beatport and Reddit, I found this recommender quite good actually! It seems to respond better to descriptions of the kind of music than to "Find tracks like these: <list>" which is what Spotify is good at.
typing in what kind of music you like and discovering music was the entire premise of Pandora and it was pretty successful.
You can easily comes up with "song/band sounds like X" but it'd be a lot harder to do with movies and TV shows because beyond genre there's a lot of variation in what makes something good. Acting, direction, lighting, story, effects, actors, etc. Failing on any one of those things means I'm less likely to enjoy a show, but being similar in any one of them could cause a match.
That said, if you were to ask for specifics it might be more helpful. Recommending movies based on tone, or style, or story elements might still be interesting but I think you'd still run into the problem where it may not easily result in something you're likely to enjoy.
I've found internet radio interesting, there's a ton of variety out there that you might not get on a local/national broadcast. Even within a genre, a lot of stations may routinely play the hits but introduce you to different 'sets' of other musicians. More generally on topic, I'd wonder about the approaches different stations and djs use to build their playlists.
Music discovery is a perennial AI tech application. The first one I followed was Firefly, out of MIT Media Lab around 1995. I think the startup was originally called 'Agents', which was a hot term in AI at the time. Thirty years ago.
https://en.wikipedia.org/wiki/Firefly_(website)
To me it always feels like music classification and recommendation efforts, open source and commercial, are too focused on music distributed in albums and singles. In the long history of human music, albums have only been a fleeting moment, an App Store of music if you will. This would never recommend me concert performances (even those on YouTube), covers on YouTube, DJ sets on SoundCloud, or game soundtrack without an official release. Listenbrainz is the closest I've found because it just has Title and Artist, but even that can be fuzzy for covers and live performances. Maybe no one other than Google can build this right now.
Music recommendation is such a hard problem. There are all these seemingly obvious relationships you can map between bands to create a big graph that looks good but that almost never captures what goes on when a human with deep music knowledge recommends music. Often the best recommendations have no obvious relationships to the band you like.
I played around with this tool a bit and didn't find it any better then other more traditional music discovery tools, that is to say not very effective.
For example, I entered John Zorn and was recommended a bunch of John Zorn's bands. I entered The Residents and got The Pixies :/
I think its more effective to click around on Music Brainz and Wikipedia.
For me music discovery is a solved problem. Here's the algorithm:
1) Imagine the timeline of musical history. If you don't have a clear idea of it, Wikipedia is a good place to start.
2) Pick any genre/period you don't know well. (For example, medieval music, or swing-era jazz.)
3) Look up the main figures of that genre/period. (For example, Guillaume de Machaut, or Duke Ellington.) Wikipedia is good for this too.
4) Listen to a sample of their most well known pieces. YouTube is good for this.
5) Repeat. Go down rabbit holes when you like.
No fancy tools needed, just your mind and the internet. This will give you interesting music for many years, and improve your musical taste a lot too.
You seem knowledgeable about this.. Care to test my old project for music recommendation? I built it by looking at co-occurrence of artists in Spotify playlists, which gives me word2vec-style vectors, and then its just kNN.
No login needed, just enter some artist names and see what you get:
https://blog.jonas-klesen.de/artist2vec
Very interesting, I've been working on a similar project (using word2vec to learn vectors using playlist data), but using songs instead of artists as the 'words'.
The main bottleneck at this point is the volume of data - many songs I'm interested in only are only represented in a handful of playlists, and . Evaluation at any useful scale is also quite difficult. For somewhat obvious reasons, in our AI era Spotify has become quite skittish about letting third parties gain access to their data at scale...
Pandora solved this problem nearly 20 years ago an Spotify with all its money and engineers do such a bad job, it’s beyond absurd.
Spotify seems to have mastered music recommendation.
It would be great if somebody could reverse engineer their recommendation algo
Actually, I find Spotify horrible for finding new music outside my bubble that i like. YouTube works much, MUCH better for me for this purpose.
Interesting. Spotify works almost perfectly for my discovery needs. I just pick a track I know that fits my mood, then use the (3-dot menu) "Go to Radio" option, which leads to a playlist that usually includes tracks and/or artists new to me. It's been a reliable discovery mechanism for me for many years. Also, there's a new feature I first saw within the last week, a "non-personalized" option that increases the "new to me" ratio.
I always find this interesting… Spotify is phenomenal for me - about every third Monday Discovery playlist has two or three hits, which feels like a pretty solid ratio, at this point. YouTube has never suggested a single thing I cared for.
I wonder if it’s a curation thing? I’ve been with Spotify since the first day it was available, and rarely use YouTube. I haven’t had a good music ratio as good since newsgroups and (real) forums a decade ago, which were a different form of curation.
My goto is https://www.music-map.com. Some of my best finds ever came from there.
I did try out the OP's recommender. It seems to misunderstand which genres strongly fit the band.
Guadalupe Plata is super-gritty mex rockabilly but the AI slotted it as delta blues.
Darla Farmer's only album is story based prog - like The Dear Hunter w/ a Diablo Swing Orchestra tone. AI called it a "hazy, intimate vibe".
All Them Witches scored better but the recommendations were all bands I know (Colour Haze, King Gizzard, King Buffalo).
Nice. I've been into Chris Luno for a while and closest is Aline Rocha, which fits pretty darn close and I have never heard of her.
Weird. It hallucinated one for me, and recommended an album that doesn't exist. Flashbacks of 2023. If this is your app, you might want to consider adding a validation layer that performs a review before publishing an output.
Tried out a few prompts and the suggestions that came back didn't even seem in the ballpark of what I would've expected. Interesting idea nonetheless.
https://everynoise.com/ is still my favourite for discovering new artists, especially in fairly esoteric genres.
When I put in a specific song by an artist that's not similar to the artist's typical output, it completely ignores the specific song.
Which is not surprising considering it's run by an LLM, but makes it not very useful as a music recommendation engine. There's already tons of places to ask "what artists are like this other artist"?
I entered "if tipper and lsdream had a child".
4 of the 5 links didn't even work. The other sounded nothing like what I described.
Thank you for this - I've just discovered something I'll be listening to all week. Is there any chance I could prompt it again with the bands I liked and didn't like from the list, possibly asking for something more refined?
Great job. This works really well.
> "lofi home recordings with no electronic elements and harmonically complex"
Yielded a bunch of things I would have expected but after 5-6 'dig deeper' clicks there were lots of interesting artists I hadn't heard of before.
> "contemporary classical with no electronic elements and small chamber arrangements"
Lots of interesting results I hadn't heard of before.
> "obscure indie bands associated with Olympia, WA formed in the 1980s"
Very good list. Does the time/location-bound results very well.
I think it's a good companion to sites like RYM as a middle ground between 'best of' ratings lists and personally curated lists.
Cool app. One small complaint is the chatty tone of the recommender engine. In particular, I find it a bit disingenous to have an LLM tell me "Ah, I love <X>!".
EDIT: I also notice the recommendations are totally different when making the same query in a different session. I'm not sure if that's intentional? I expected at least some overlap with the previous time I asked.
I really want this to work! But maybe my requests are too niche. It also hallucinated an EP for me. And the bandcamp links never work?
Great idea though! I got inspired to listen to some stuff by it, even though it wasn't really what I wanted to find.
"Like Veruca Salt" give me a few bands which sound rather like Veruca Salt, not bad at all ...
I thought I was the only one that remembers that band!
Picked the CD out at random while getting drunk last night, what a band!
Can’t fight the seether.
As someone who's big into UK Bass who finds new music mainly through a mix of Spotify, Beatport and Reddit, I found this recommender quite good actually! It seems to respond better to descriptions of the kind of music than to "Find tracks like these: <list>" which is what Spotify is good at.
Recs weren’t bad!
I’d be curious about the data source here though. Custom curated? Relying on LLM World Knowledge with some prompting?
Heads up that the site doesn’t resolve without www on the domain name. Just going to secondtrack.co won’t work.
does anyone actually want to discover music like this, by typing in what mood you're in or what kind of music you like?
if so, what about something more visual like movies and tv shows?
typing in what kind of music you like and discovering music was the entire premise of Pandora and it was pretty successful.
You can easily comes up with "song/band sounds like X" but it'd be a lot harder to do with movies and TV shows because beyond genre there's a lot of variation in what makes something good. Acting, direction, lighting, story, effects, actors, etc. Failing on any one of those things means I'm less likely to enjoy a show, but being similar in any one of them could cause a match.
That said, if you were to ask for specifics it might be more helpful. Recommending movies based on tone, or style, or story elements might still be interesting but I think you'd still run into the problem where it may not easily result in something you're likely to enjoy.
Yes. All the time.
FM Radio is the best method to discover music, especially for getting out of your bubble.
I've found internet radio interesting, there's a ton of variety out there that you might not get on a local/national broadcast. Even within a genre, a lot of stations may routinely play the hits but introduce you to different 'sets' of other musicians. More generally on topic, I'd wonder about the approaches different stations and djs use to build their playlists.
I've discovered and learned about lots of interesting musicians listening to the radio station of my university
College radio is awesome.
Specifically independent and public music radio. Commercial FM / Clearchannel is trash in my opinion.