It does use AI for the voices, but god dang if it doesn’t sound like a genuine talky version, with Gary Owens in particular sounding like they genuinely voiced it.
Runs with DOXBox-X and DOSBox (0.74-2 or earlier). ScummVM has an issue where loading a saved game loads with missing text boxes.
This is great as a proof of concept but I would consider scummvm compatibility a hard requirement for me these days.
It sounds really impressive to me. Looks like a lot of hours went into this, although the 100 hours on generating and selecting the voice lines would be more interesting if it was spilt between human labour and the GPU run time.
To me this is a great use case, it’s a free mod for an old game - I can’t imagine anyone could/would pay actual voice actors to do this.
- Synthetic Voice Generation: Tortoise-TTS can take hours to generate hundreds of vocal lines on a GPU. Additionally, you may try generating multiple samples at once, as the quality and delivery may vary with random seed.
- Quality Control: Filter out bad audio samples and select the best ones. If a vocal line isn’t working try generating more, correcting any issues with the text or rewriting it more phonetically. For example, rewriting “EVA” as “E.V.A.” or “II” as “Two”.
Rough estimates.
- 100 Hours of work generating, selecting, splicing and cleaning audio files.
- 30 Hours of programming with the SCI Companion.
- 1/2 Hour on SCIProgramming.com.
I like the effort, but definitely needs more work.
I was hoping it’d just pull all’ the lines from Sven’s playthrough v.v https://www.youtube.com/watch?v=EvdnUGWhybk
Lil doofy they call it a voice “acting” mod if it’s just AI
vom🤷I checked out the linked sample video and just now got that Kiz Urazgubi is “kiss your ass goodbye.” Hah.
Not everyone’s going to be happy with this, but I’m definitely going to try it out. This is one of my favorite Sierra adventure games.
