Hacker News new | ask | show | jobs
by Retr0id 169 days ago
The main challenge for me is recommendations. 99% of the value I get from Spotify is the auto-generated playlists and "song radio" feature. It's easy enough to fill a NAS with flacs but I don't know how to replicate "the algorithm" in a self-hosted setup.
3 comments

I wish there were some self-hosted setup that would replicate this based on either tags, bpm matching, or other similarity indicators snooped from online sites like lastfm.
I've been using https://www.beatunes.com/en/beatunes-matchlist.html (35€ for Mac or Windows) to create static playlists like this for years from my library of ~13K songs. I've created a few dozen such playlists with light curation (just kicking a song out here and there). I find myself listening to these playlists far more than the ones I created completely by hand.

I suppose you could do it often, or maybe even script it (e.g., with Keyboard Maestro on Mac) to get something a little more dynamic. But it's not something that just matches songs on the fly server side.

What drives me away from Spotify is recommendations: Starting with my preferred, it drifts into unknown artists; Suggested playlists are unfortunately for people my age (New! Last album from Pink Floyd! probably because I’m 45). And it’s not so hard to map the world’s music, the list of genres is manageable by a human who’s experienced in music, what people will like is also predictable by a human with experience. If my friends can do it, why can’t Spotify?
Weird, for me Spotify's algorithm is garbage.

But the related artist list is good for discovery.

It isn't that great for me either, just highly convenient. So doing better is probably tractable!