has anyone here ever tried using voice changing software?

one recurring need i find from time to time is the need for a female voice actor, which i don't really have handy access to all the time. also, using an outside voice actor can cause problems if you can't get them back to do redo's, expansion packs, the next version of the game, etc.

inspired by seeing the movie S1m0ne again recently,

i started wondering if it would be possible to use software to convert male to female voices (or vica versa).

BTW, i'd recommend all gamedevs watch this movie if you've never seen it.

a quick google search reveled this tool:

http://www.screamingbee.com/product/MorphVOX.aspx

so i was wondering if anyone out there has played around with this kind of stuff, and if it actually works or not.

if it does, i could use a male voice actor and convert to female using software. combined with 3d modeling software for graphics and animations, and a game engine for programming dialog and behavior, voice changing software could give you all the basic tools necessary to create synthetic actors as seen in S1m0ne, not that it wouldn't still be a lot of work - after all, in the movie Al Pacino spent 9 months programming S1m0ne. And we don't quite have combo image recognition and motion capture software as seen in the use of the "scan" button in the movie (at least as mainstream apps) - but we're getting there.

Using manual pitch shifting, time compression/expansion, or even autotune you can gender change a voice. However the amount of work it takes is pretty staggering if it's beyond a line or two of dialogue.

many years ago ray kurzweil demo'ed some kind of LPC source-filter separation that allowed him to replace the filtering of his larynx with another model. dunno if that's what that app is, but i've never seen an accessible/affordable release of the tech....

in freeware, you have formant shifters like kerovee, which "acts like a pitch shift without changing the pitch" - there is of course much more to voice than size, whether you can make a fairly convincing recording is dependent.. eg. female larynx often physiologically doesn't close as much as the male larynx, producing a more breathy, less "tonal" signal - maybe you've met women who have naturally breathy "husky" voices where it sounds like whispering. human beings are generally quite focused on acculturation, so faking a human voice is a "high criteria" task.

formant shifters eg. use fft/ifft and breathier/noisier/more chaotic signals will suffer more from transformations.. it's better to be shifting up than shifting down if the voice is very breathy..

generally people advise not attempting it but imo it's worth a try and informative to do.

One of the engineers mentioned an interesting way to do this that he's tried before. Read the male VO, pitch shift it up and time compress it a bit. Then try to mimic the speed and pitch as best you can. Take the new take and time expand it the same % as the original take was compressed and match the pitch shift as well. Comes out more natural sounding than a straight pitch shift.

before posting, i played around with the usual suspects such as pitch shift, but was unimpressed. results similar to Nypyren.

there is of course much more to voice than size, whether you can make a fairly convincing recording is dependent.

One of the engineers mentioned an interesting way to do this that he's tried before.

yes, i'm thinking one will have to start by trying to sound like a girl, then post-process from there.

Screaming bee

ray kurzweil demo'ed some kind of LPC source-filter separation

in freeware, you have formant shifters like kerovee, which "acts like a pitch shift without changing the pitch"

thanks for the tips!