voice of a person reading a text based on a small sample

[deleuze3]

Hi,
I have some samples of a voice of a person. And i was wondering if out of the samples, i could produce a voice over that read a text.
It might be complicated, maybe unrealistic.

It's just that the person is not available for the voice over and i would rely on only samples.

Thanks in advance,
Deleuze3

 

[Alex H.]

So, you’re wanting to copy and paste words - syllables? - from other recordings to create a new VO?

If you want it to sound like it’s been cobbled together from other recordings, go for it. It takes some pretty advanced voice synth to be able to do this without sounding like it was edited.

 

[Noiz2]

I will be slightly less negative. Depending on what "samples" you have and what you want the VO to say, and how good an editor you are, and how much time you can throw at it, and how lucky you are, it might be possible. I have done short bits of "sync" dialog that way, and have cleaned up small bits of radio dialog that way. The key is generally to have a lot to choose from and not having a lot to create, and luck. Really luck is probably 50%.

But instead of just saying NO, I'll tell you about some of the issues why the answer is probably no.

You need all the words, or sometimes you can create a word from other parts but that takes a lot of time and a lot of luck.
But it's not just the "words" because a word is said differently depending on context. And while speaking many people will smear words together. I have had directors that tried to point out where each word was in the transcript but that NEVER works because the words are in the wrong places and the inflections are wrong and transcripts separate smeared words and as an editor you can't because the smeared part has a bit of each word so to use either part you need to find a replacement for the smeared bit.

So in most cases in a ten minute piece if you want me to "fix" or alter a word or two I can probably do that with a little luck. If you want to create a 5 min sequence out of that 10 min piece where they say something completely different... Even with a TON of luck the odds are probably less than 1 in 100 it will come out "natural".

Forget the spy movies where 5 seconds of your voice and they can get you to say anything, that is fiction.

Not far off AI voice recreation may be able to do a decent job? Heck the CIA may be able to do it now??? But on the ground where we are, no editor tools exist that I have ever seen can.