I’m still single-minded in that I think all this AI gen music is plagarism.
In Roon, I want nothing more than foolproof detection of fake music.
Same. There’s no use asking a streaming service to do this, but Roon should serve as a filter for this. Only stuff that has been manually attached to a human’s discography should appear in Roon’s database at all.
The irony of using AI for an answer. That database is depreciated.
Nonetheless, any solution, whether MusicBrainz or Zoundhub, is dependent in volunteer effort.
No, you’re not. I don’t want to hear this slop, and I don’t want to see the irreparable damage this could bring to music.
I assume you are referring to the Soul Over AI database. Yes, true that it is no longer active and not officially verified. It was and Zoundhub is a community-informed, independently curated directory rather than an officially monitored database by a corporation or government entity. It is maintained by a small group of volunteers and relies on user submissions to populate its database of AI-generated music, artists, and tools. While it is not foolproof, it does provide a larger database than MusicBrainz…
Edit:
Because it would be hard for me to determine how many artists are cataloged within the 83 genres, I asked Claude ai if it was prudent to implement MusicBrainz’s API database to access the artists tagged with one of these 83 “AI” tags. Here is its response:
What those 83 tags likely represent: The MusicBrainz tag search for “ai” would return results like ai generated, ai-generated, ai music, ai assisted, artificial intelligence, and many variations — but these are user-applied tags on specific recordings or artists, not a curated database. The count of 83 reflects how many different tag variations exist, not how many artists are tagged.
The practical problem for our extension: To use MusicBrainz tags at playback time, the extension would need to:
- Take the artist name from Roon’s metadata
- Search MusicBrainz for that artist to get their MBID
- Look up the artist’s tags using that MBID
- Check if any tag matches our AI keywords
That’s 3 API calls per track — adding latency to every single track that plays, and subject to MusicBrainz’s rate limit of 15 requests per 18 seconds. With Roon Radio skipping rapidly through AI tracks, this could easily hit that limit.
Coverage is the bigger problem: The MusicBrainz community has been actively debating how to handle AI content — as we saw in their forum just days ago — but there’s no organized tagging effort. Coverage would be extremely sparse compared to the Soul Over AI dataset.
Bottom line: The 83 tag variations are intriguing but the combination of sparse coverage, API latency per track, and rate limiting makes MusicBrainz impractical as a real-time lookup source for our extension right now. It’s worth revisiting if the MusicBrainz community organizes a systematic AI tagging effort — which their active forum discussion suggests could happen.
I think MusicBrainz would have a much larger community of volunteers, but, as yet, there is insufficient adherence to the guidelines, largely because those who add the slop via API, are abusing the system.
An AI artist should include a disambiguation in the artist data. This is used, albeit inconsistently, and returns far more results than your screenshot.
Nonetheless, I think this thread should be about what and why rather than the how at this stage.
Why would we need any database of ai-generated music when we already have databases of human music? Anything on streaming that isn’t in there doesn’t get shown [if user deems it], job done.
A database of ai music will soon be bigger than the body of human music and there is absolutely no point trying to track it all. Just have to assume what is not legitimately human is all AI. Not human? Nope
Because these databases of human music is woefully incomplete (which is why there is an “unidentified” problem in Roon in the first place) and new human artists appear every day.
And how does this not apply tenfold to a database of AI music?
[now that I’ve sorted out my box sets by putting them together as one release, identifying that and then breaking it up into separate albums again which retains the identification] I only have “unidentified” on stuff that hasn’t been released, and I have a massive and diverse collection
Sure, but I was replying to the idea that everything that isn’t in an incomplete database of known human music should be autobanned.
And Roon gets its credits from (incomplete) MusicBrainz/TiVo, they don’t get credits from Tidal (I don’t know about Qobuz)
I’m afraid if you don’t want your content to be majority AI that’s where we are at. Manually certifying missing stuff as human/adding to human db would be a far smaller task than marking an almost infinite amount of incoming AI as AI. You have just got to have a system of assuming everything is AI by default unless it has some legitimate known human property
I have lots of music that was released (on tiny or even larger local labels that died decades ago, for instance. MusicBrainz coverage on non anglo-american music is even worse)
Of course there will be some, but the missing human info can be added to a human db and the amount will be a tiny fraction of the size of missing AI info which is being added to in obscene numbers daily
Trying to identify all the AI instead of the smaller amount of stuff that isn’t AI and of which identification has been going on for decades is just, not thinking about it in the optimal way.
It’s a hard problem but I’d rather live with AI music (which barely affects me, and we could have manual ban buttons if it appears) rather than random MusicBrainz/TiVo editors deciding which human music is allowed to be seen.
You severely underestimate the human output in genres like punk or Hip Hop. If it were easy to add it, it would already be there.
It’s very easy to add any existing streaming release to musicbrainz and the (apparently underestimated) amount of AI music being “created” means tracking it is not possible
In the meantime I hope nobody is poisoning the well by adding AI music to musicbrainz
OK, I’ll wait until it’s done.
It can be done, at least. Adding all AI music to any database just cannot. Unless it is just by taking [content of streaming services] and subtracting [known human music data], which would be the same thing as assuming any unclaimed music is AI
Sure you can script adding ALL streaming releases to MusicBrainz, but this does not solve the problem of how the streaming service then decides what’s human (and should be added) and what is not
That’s exactly what we don’t want to do as it would fill musicbrainz with AI music and then we don’t even have a database of human music anymore.