Negamuse wrote:Altruizine wrote:Negamuse wrote:Yeah AI kind of isn't great for that and it's almost by design - any answer it could give you is derived from weighted averages of other people's opinions and by definition that's going to tend toward mass opinion.
That's not really accurate at all. You're only describing a certain primitive type of LLM that does what you described. A different form of AI that knows your music library and can, say, analyze and cross-reference audio files and detect specific patterns would be one of the best tools for this.
Hey if you've got a link to a system that can take music as an input and makes qualitative decisions about similarity that isn't based on
- IDing what you've given it as input based on matching fingerprints IE how Shazam etc do it
- comparing user preferences against known listening data like Spotify etc do it
I have a whole bunch of uses for that that's aren't limited to "find me more music like this"
I don't believe anybody has built that and I don't know how they would. The only systems that have actually ingested the world's recorded output and worked out how to do anything with them are things like Suno and Udio and they're not prepared to say "I made this, and I made it based on these songs" because of the legalities of it.
With a quick search I found something called Cyanite.ai
It's an AI engine that can detect genre, subgenre, mood, instruments played, vocal characteristics (e.g., male, female), BPM (tempo), energy level, and even descriptive keywords based on audio analysis. Can be used for updating metadata
By analyzing various factors such as sound, vibe, and era, our AI identifies songs within your catalog that share similar characteristics. This tool is invaluable for music supervisors, playlist curators, and A&R professionals seeking to find tracks that align with specific musical references.
I don't know if it's any good, didn't try it but the concept seems technologically feasible, given that automatic speech transcription is already good and seems more complex than this
Or actually I’m not sure which would be more complex. I was thinking in terms of, music’s more structured and (usually) consists of more distinct frequency patterns so maybe that kind of data is easier for a machine to interpret