top of page

Creating a Fully Artificial Voicebank

Wip

00

01

  • Audacity: used for most of the editing
  • Adobe Audition: post editing such as erasing clicks, clipping
  • Paint.wav: generates noise based on drawings
  • Vocalizer: makes your vowel sound more like vowels 
  • Vocoder: does the same thing as vocalizer

02

For the audio part of Fermi's voicebank, I was inspired by Curiosity, the singing rover,who sounds like this
00:00 / 00:08
Out of all the synthetic wave sounds, square waves sounded the most similar. I created the vowels in Audacity by stacking square waves at different frequencies and volumes (Somewhere around the 5~10 minute mark)

Then, by using the plot spectrum tool, I analyzed the frequencies of my own pronunciations and used the filter curve eq to make each vowel sound more like "me".
If needed, I also used a little bit of vocalizer.

Which ended up with this result:
00:00 / 00:04
The voice came out pretty deep, and didn't fit well with what I wanted for Fermi(which I envisioned as a really high pitched, sharp "cutesy" voice)

I started fiddling with ways to make the voice sound "cuter"
which usually involves raising the pitch using an Audio editor. However, I found that by doing so, it won't render properly in UTAU. I didn't really feel like looking through specifically why this happens, but the simple solution that worked for me personally was to just use the g- flag in UTAU and render out that audio.

After throwing Fermi's vowels through tn_fnds with ~g-30, she sounded a lot better:
00:00 / 00:05
bottom of page