Secret Mixter + Listening Party
skip
Home » Forums » Features » AI vocals

AI vocals

Speck
.
permalink   Sun, Apr 7, 2024 @ 5:17 AM
Can we institute a policy whereby any uploads containing AI vocals must be tagged, or otherwise designated as such?
(Anonymous) recently uploaded a mix with very good unattributed vocals that they also uploaded separately as an additional file. I can see a case where someone uses that vocal in a remix and attributes it to (anonymous) when it most likely, or at least quite possibly, is not (anonymous) singing.
Apoxode
admin
.
permalink   Mon, Apr 8, 2024 @ 2:29 PM
Speck,

I think that’s an excellent idea for a policy.
It seems the general consensus for the moment is that AI vocals are public domain, since they aren’t eligible for copyright.
-but- if we can’t properly source the pool for what the AI generator is using, we could be unintentionally infringing.
For example, voice models aren’t always authorized.
Either way, isn’t the software we use part of the production, so we might credit that at least?
(just my two cents lol)

It’s kind of tricky with public domain samples since they don’t need to be attributed, but you don’t know for sure if they’re AI generated or just sampled.
So tagging it as AI could be helpful in that sense.

I’m not sure where we could put the tick box for it though, maybe in all the submission types?

Kindly,
Apoxode
 
.
permalink   Speck Tue, Apr 9, 2024 @ 7:58 AM
Yeah it’s tricky I guess.
What I really want is for people to somehow (in the song description?) let listeners know the unattributed vocals are 1)me (the uploader) 2)public domain samples (don’t even need to know the sources) or 3)AI.

Anyway my plea is a selfish one, feel free to disregard.

I just don’t want to recommend a mix with AI vocals as I don’t care to support the concept/product. So, as it is now, if I come across a mix with unattributed (or unclaimed or undescribed) vocals I will just assume they are AI and ignore the upload.

By the way - I’m not talking about your recent upload with all the cool vocal snippets. I assumed those were public domain (‘cause that’s what you do). But also it occurred to me that would actually be a good use for AI. Creating little snippets like that. Wouldn’t really be sidelining a vocalist. Not much different than the old school text to speech generator really. (Though I do still think real found snippets would ultimately be more interesting/evocative.)