Hi!
Is there a website or software that takes an existing vocal track and changes the voice timbre in such a natural way that you cannot tell that it is not sung by a human?
Or is AI not as good yet at generating authentic vocals?

How about Sonarworks SoundID VoiceAI or www.audimee.com
?

 

Not a website but you must check https://dreamtonics.com/vocoflex/

 

I think it won't be cracked to test it, (and it costs like 200 bucks)
Do you think Vocoflex sounds more natural than SoundID VoiceAI or www.audimee.com ?

 

As a singer i hope machines are never able to sound as good as humans

 

just because technology can do something as good as a human doesnt mean it will take away from you.

Cameras didnt remove painters.
Photoshop didnt kill photographers.
etc...
etc....

 

why not use the free http://tryreplay.io or the free http://weights.gg

over audimee?

 

I bought sound id perpetual, and its well worth it. Its actually crazy good. In my use of it, which is layering to back up my singer, i make sure to grab plenty of extra takes, sang in a both the same style as the main vocal, and i have the singer to "fun" takes, where they can throw their voice differently. I will then comp, tune, and treat those extra takes, at times tuning them to be a harmony, then have SoundID replace it completely, or at times, tucking in the real vocals behind the SoundID vocals.

The amount of voices you can choose from is impressive. The little kid voices are a lot of fun to layer in for a pop song. They also have instruments, which can be a lot of fun. Layering in a "trumpet" against a melodic word less hook, jazz scatting etc, is just awesome. It doesnt sound like a trumpet really, but it sort of does, and it just brings a fun layer to the production.

Cannot recommend it enough. Its affordable, works pretty damn quickly, and doesnt use much cpu. You can have it just live on the track once its done processing, but what i do, is go to the file ouput folder, and drag that into the session, deleting the plugin.

 

This is Dreamtonics Synthesizer V. I find it expressive enough for my purposes. If I want to change the voice I upload the voc.wav to Replay and try some variations. Usually get close to what I want.

 

well, i believe the houses that used to develop the pictures might disagree with you. and though photoshop didn't kill photographers it most certainly didn't help them.

and thinking AI will not take away from you is being a bit naive about it. just check the new coca cola commercial. how many musicians, photography directors, movie directors, actors, figurines, etc were avoided so that AI could do everyhting on its own? what'ss more, AI probably balantly stole from their ideas and got away with it. So thinking it won't take away from you... hum... just wait and see.... if you're working doing voiceovers for the radio and suddenly they can have Ai do that do you think they will think twice about firing you?

just think how a bar will rather pay a DJ 50 bucks instead of a live band. you can argue that you can still bring your value to the table. yes you can, but noone gives a flying F about it. you think the elevator company wants to pay royalties for music when they can have Ai steal other artists and generate a couple of tunes they don't have to keep paying royalties for...
yes, technology WILL take away from you. It's just a matter of time.
Remember John Henry.
when you're going through the interstate, check how many toll boths are being operated by machines instead of humans. it WILL take away from you. it's just a matter of time. You can't do anything about it and have to accept it. that's true. but until they have real laws about copyright theft this will get progressively worse and worse.
and even though it's theft and no two ways about it, people don't care. just check the millions of reviews of people commenting on this coca cola Ai POS sayin "I don't see anything wrong. I loved it" - and it doesn't matter that you tell them the computer is stealing from others. they don't care. so when you're part of that industry it's VERY naive thinking they will not come for your job next.

I can see your point if you haven't got skin in the game.

 

I'm using Replay as I type this. I like it for backing vocals drenched in delay or other fx and then buried in a busy mix. It's fine, it's just that it still generates too many artifacts, and I always test different F0 methods, torch compile combinations etc., because it is very dependant on the source material. Also, I don't know how to tell whether I'm using 'top-notch' or garbage models (a common challenge for community-based repositories, much like NAM's ToneHunt.)

As of now, I don't see it working for delicate voice + guitar folk songs, for instance.

 

Here's a delicate folk + guitar song using Replay

 

That's interesting, mate. How do you avoid getting consonant distortion, for instance? I'm constantly running into problems like b's turning into p's, or robot-sounding sh's, or d's and t's turning out way too crisp and harsh.

 

if the perpetual license only allows you to use your local computer for processing, is the quality exactly the same as the online processing?

 

The source material has to be good to begin with and Synthesizer V works best if you reduce the breathiness and lower the tension so the sound is rounder. Try and match the source vocal to the model so there isn't a lot of recalculation in conversion. ie, this is a soft airy vocal changed to something like Billie Eilish, but Replay does add more character and nuance than Dreamtonics alone. As a general rule of thumb, when sourcing models from weights.gg the larger the file size the better the result. There's a few setting in Replay but I'm not at the computer at the moment but they're kind of self explanatory.

 

yup! no quality issue. the occasional wonkiness of "AI" vocals, but if you clean up and properly tune the vocal being used to trigger the plugin, there is little to no issue. Ive found its best to clean up all clicks pops, breaths, esses etc, tune manually using melodyne and then super hard tuning using autotune or similar. Sometimes i skip the melodyne and its fine.

 

Thanks a bunch. Any other insight, I'm all ears. But yeah, it's amazing how artifical Synth V sounds and how Replay makes it come alive, pretty much like NAM for guitars imho. In all honesty, I don't even think messing with Synth V model-specific parameters like 'bold' or 'cold' helps much when you're aiming at RVCing it later.

 

Last edited: Nov 30, 2024

There was no real vocalist involved, right?
You typed in lyrics in Synthesizer V Basic, and used which voice database?

Or used which Replay voice?

 

Great job!

For the record, this is the best channel I could find with tips and tricks for Synthesizer V:

 

Audimee works well, but you still have to give a really good vocal performance if you want something good back out of it. Basically it makes it possible to change the timbre of your voice.

This is AI vocals from audimee.

But it required actual solid vocal takes from a real singer (who had a different sounding voice than the AI model used. But if you hear the takes side by side, you can tell the nuance comes across intact.

 

In my opinion, AI is destroying creativity and the human touch, music is becoming so generic and lifeless, seems no one has to have any "real" talent these days, music as we knew in back in my day has become a thing of the past, much like the horse and buggy