Hi all.
I want to be able to record Professional sounding voices with lyrics in original song productions, using this new AI clone technology.
Im not seeking to have an specific artist sounding result, actually I would like to get a "Non recognizable" professional voice that sounds human.

Is there a way to achieve better results than vocaloid, Emvoice, Synthesizer V etc..?

Im not a singer at all. So my reference voice cant help.
- Dont tell me to hire someone...

Thanks

 

Hi, maybe ACE Studio will be the thing you looking for

https://acestudio.ai

I didn't have the time to try the demo yet, but some video on Youtube look promising.

Other stuff like Replay could work to, but you would need a source recording to use as a base (unless you use other's people acapella), you can use it with different voice model (don't necessary have to be from a singer). The best way you could use it for original material would be to record yourself and using a tool like Melodyne to pitch correct (if needed), and then use that as a base. You can also use other tool to train your own model. I use that a lot with my own voice and some model I made from artist I work with to turn their voice memo into good recording, make harmony, try different tonality.

It's not as complete as ACE seems to be but it's free. Here a thread on here where you'll find all info (it's made by an active member) :

https://audiosex.pro/threads/free-a...ning-app-with-built-in-stem-seperation.72525/

 

Vocaloid, Emvoice, Synthesizer V: despite not being AI they are excellent vocal emulators, I find them very useful for background vocals, but as you already know, they sound a bit artificial though.
This is my workflow for creating a lead vocal with a human feel, without hiring a singer:

• Record the part as imagined it using my voice (no: I'm not a singer but it doesn't matter)
• Treat the recorded voice with Melodyne to make it acceptable
• Export the voice and open Replay.ai
• Swap the recorded voice with one of a superstar singer
• Done

This workflow is empirical and gives unexpected results: sometimes very interesting, other times dreadful, when it works it is truly rewarding. I also tried this procedure using voice emulators as a reference but nothing good comes out of it: the source voice must be human, emulators are too artificial

 

Just wrapping up a 15 year project now and my last piece involves the use of REPLAY software (voice cloning models selections and application) - an AI voice model tool that is incredible. I wanted (for my last song) an all male chorus (think 'friends in low places') but the trouble is, they (the multiple vox takes I did) clearly all sounded like me.

I went to weights.gg and voice-models.com and hand picked some male voices (not saying whose!) and assigned my many vocal lines of a unison chorus to these other voices. Then, for added measure, I randomly pitch and length adjusted - almost blindly - each and every voice just by a bit in every direction. Up, down, sideways using Melodyne.

WOW! I suddenly had an amazingly natural all male chorus and you can't at all tell that it's 'rock artist 'xyz'' singing along with me at all.

The thing is, you have to (or have someone you trust) sing your desired lines. You have to be happy with at least a rough take that you can later manipulate, just as I did.

Not sure if this applies, but maybe you can think of some similar applications.

 

Synthesizer V can sound incredibly real. The issue is that it takes time, skill and effort to get there. It is like learning to use a DAW. Fortunately, they seem to be making realism easier with every update. I bought the software about two years ago, and just a few days ago I was using it after a few months and the Takes function is now very convincing compared to early last year.

My two cents? If you want convincing fake vocals, you need to put effort into it regardless of the software.

 

Thanks everyone for your help. I will check/try all your suggestions.

 

Which method gave you better results for spanish lyrics?

 

Vocaloid has a nice male Spanish emulator, Bruno: It can give you decent results as background voice. Also, you can somewhat force English singing voices to fake Spanish language by cheating on phonemes, but I generally avoid that. A different approach is to use a speaking voice emulator (any TTS engine) to then treat the file with melodyne

 

Bruno + Waves Harmony and background vocals are done (with some mixing tricks)