Pepper&Carrot: The Wizard (Tortoise TTS example)
24A playthrough of Pepper&Carrot: The Wizard, which was a submission for the 2023 Pepper&Carrot Jam. The game can be played and the source downloaded on itch io (user: Perita.)
Tortoise TTS was used to generate voice linces based on the following voice talent:
- madamvicious (freesound sample 370268 - cc0 - perhaps this was an unfortunate pick, since the sample is an awesome rhaspy cartoony voice but audiobooks are rarely read in this tone, so the AI probably assumes this to be an outlier)
- unfa (freesound sample 621782 - cc0 - super clean)
Since tortoise was trained on masses of (likely Audible) audiobooks, the legality of this is an interesting thought experiment. But because of the not-so-great quality and 0 commercial consequence, that battle won't be fought here.
Also interesting is the morality (and legality?) of using these voices, in the open, without permission. Or is permission given because of CC0? Huh.
Another point: what do you think of the quality? Perhaps, just like with image AI generators, some are great, some are mediocre. The bad ones weren't used in this video. About 25-40% of lines had to be re-generated to get a tone that matched the situation a little better. Only the "fast" preset was used.
Just to be clear, the voice samples were not injected into some proprietary black box AI site. Just a local, disconnected black box AI. Tortoise TTS ran locally on i7-10700K with 32 GB RAM to generate these lines. It's not clear whether the RTX 2070 Super was configured to speed up the work. The voiceover also wasn't modded into the game. The voice audio was added in post :)
The game itself is open source, licensed under various CC licenses and GPL. Coded in Haxe.
Music (as used by the game):
- Clean Soul by Kevin MacLeod, CC-BY 4.0 Int
- Sneaky Snitch by Kevin MacLeod, CC-BY 4.0 Int