Right now I follow the XTTS finetune to xtts generation with said model, into rvc with an rvc generated model and the xtts generated audio for reference audio. It came out pretty good, but it needs to be more humanised, as I am producing speakers for animation. Out of all the methods in post, what method and plugins would be best used for this?

Autotune, melodyne, RX, dxrevive etc?

Any help is appreaciated!

EDIT: Here is the ai audio that I am trying to fix https://voca.ro/11F2TWdibPzp

EDIT: Guys, I am a whole man production trying to make a short sketch animation for tiktok and youtube. Some people have said I should not creating voices as its too easy, but I still have to do the voice acting myself, all the ai is doing is transferring the model I made to the reference audio. And you may think it is easy, but I spent 10hrs yesterday, and another 10 the day before. So not everything is cut and dry, but maybe it would be if I didnt have to do animation, mocap, lighting, modelling environments, modelling characters, rigging, sound design, music producing for the background music and more, so again please, if you can lend some help to the question above, that would be greatly appreciated.

PS: the voice I am trying to enhance is p diddys!

 

Last edited: Sep 29, 2024

I can answer this. But i won't.
Your goal is to replace what should be a human voice actor with AI models.
Ethically, this practice is not acceptable unless it's meant to be used for all those nameless vids in yt like "10 facts everyone should know" or similar. No matter what the budget is, i am sure there are always some people who would do it for a low fee and perhaps even for free with a percentage fee later. So, sorry but i won't contribute in cancelling people's jobs.
Cheers

 

Last edited: Sep 29, 2024

Guys, thi

 

The best method for post processing ai generated audio is not using ai generated audio in the first place

 

Im making a p diddy sketch animation, so in this case, it is very much the time to use it. But on top of this, do you know how much work goes into this? You have to find the dataset, then edit and listen to it a few times, deverb and de noise with uvr, which means installing a ton of models to find the perfect parameters. After this, you have to install 3 different ais on your local computer, which you can only really run with a good computer, but trying to get it to work can lead to a lot of time wasted. Months back, I spent roughly 30hrs in a week trying to get it to work, and oh boy that is not even the testing. Then you have to train the datsets, which can take a long time to not even work. Then you have to do it again, and again, and again. Today I spent around 10hrs doing this!

 

What I hear is a pitch oscillation problem,
pitch varies up/down constantly, making it sound strange..

Hard to fix with conventional means,
best will be tryin with modern software like Autotune, Melodyne, etc..

 

That's an interesting point of view.
But do you really believe he would hire someone, if AI wasn't an option?

Edit:
I was thinking about my own question.
maybe he would. maybe he wouldn't. only the OP knows.
But also other people could pass by and lean from this thread so, i guess you're right. It could contribute in cancelling someone's job.

 

Last edited: Sep 29, 2024

Errr, short answer is yes. The other "why" that i did not mention, is the human voice nuances as per different occasion each time, are far from perfect when replicated by AI. At least yet. As for the matter of looking for a voice actor:
It's not so easy tbh, you have to go through community "channels" like people per hour, upwork, fiverr and other similar websites that offer services. But it's definitely worth the extra time and work because you make acquaintances with real people. Anyone who's been long enough in the music biz knows that connecting with peeps with similar or interconnected interests, in the broader sense of human relations of course, can be the A and the Ω. From my own personal experience, i have met with peeps from the web, which -some soon some later on- led to successful collaborations. With some of them it even led to becoming real life friends to this day. At the very bottom of a "very long, constantly unfolding music biz thread", it will always be who you know really, imho.
PS: The OP friend should come clean from the very beginning that he is making 15 sec funny vids for TicToc etc. all by himself and not something more elaborate. It is a different ballgame really, because next thing coming at us is Disney and the likes substituting voice actors AND all sorts of music/audio related jobs with AI. And a good part of animators, cinematographers and the list goes on and on. As i explained to the OP -because he pm'd me- it would be best to look for someone to fill the AI gap there. But i did give him a short answer to what he asked because he seemed kinda desperate.

 

Last edited: Sep 29, 2024

oh right! there's also that! good point

indeed, that makes the whole difference. cool, man!!

 

They are not going to come here and ask for either technical assistance or our permission when they do. Some entity like that will have so much IT capacity for AI; it's not going to come ask us dumb humans for anything at all. It will already know how, and will just do it.

 

Having people here that recommend autotuning for a more human sound ... should make you think. But not in a good way.

 

The amount of time it takes is irrelevant to whether or not it's ethical.

 

I don't see how this is unethical. Whether I use a hand saw or a band saw is not the issue. As long as the outcome is what I wanted (or what the client wanted). If he hired a voice actor and then cancelled because he found an AI app that can do it, then yes, I'd say that is unethical, but if he never consulted with any voice actor and used AI from the start, then I don't see the problem. You use the tools that are available to you to do the job. Whether it's a person or a program. Doesn't matter.

 

"The next thing coming to us" meant us as musicians, not us as audiosex forum members per se hehehe... i'm pretty sure Disney haven't the slightest idea of audiosex's existence or its inhabitants

 

That's my point. It's a completely separate thing from this guy asking this on a forum. It's some person's question on a forum. I'm not sure how you get from there, to it causing someone to lose a gig. And that leads me to agree with this:

 

The best method for processing AI music is AI processing. This should be self-evident.

 

I suppose, stuff like LANDR and eMastered...

 

I second that emotion.

 

No you don't get there from this particular guy. But yeah if what he did was on a normal scale and not 15 sec vids it would really have cost a person a gig.
Plus when the OP pm'd me and explained what he wanted this for - because you didn't read the whole thing obviously- i gave him an answer to help him, in what to use with Melodyne studio to try and humanize the vocal. But he should have come clean from the very beginning on his purpose and not the original vague post.
What you say about a person losing a gig, i would always say no if someone wants to cancel a fellow artist for AI because that is a -proven pro- person losing a gig. Hobbyists might never understand the importance of a payday for a professional artist unless they have experienced similar situations, ie. its their own job on the line. Pro artists live by this and some of us feed our families. Old fashioned or not i don't give a dime really.
So yes, if i can, i will help upstarts with small projects when it comes to "bettering" AI produced content, but i refuse to help anyone looking to make serious money with AI making music and vocals and whatnot because they will be cancelling people's jobs, this in my mind is and will always be unethical. It's how i roll. And be sure i am not an ignorant, i do endeavor in AI, i have active accounts in Suno, Udio and am a weights.cc discord member (for Replay models). But that to me is an academic research and fun of course.
Also, i do hate it with a passion when there is this fucker in Spotify with 600+ accounts making AI generated ambient tracks and having 1+ billion plays. Or similar examples. Fuck them and the listeners and his Spotify creators friends who, sooner than everyone thought, had their own documentary made projecting high standards and musical ideals. What bullshit. A fucking disgrace of the human race.
Peace

 

Last edited: Sep 29, 2024