Batch #33 - 6.5 new voices
Added 2022-01-10 10:57:40 +0000 UTCSome new voices from the last poll are here, along with some further v1->v2 re-trained voices.
I actually included one more voice from the poll that didn't win, because I had already trained it before the poll ratings changed order, right at the end. The voices are as follows:
- Skyrim: Florentius
- Fallout New Vegas: FemaleAdult09
- Fallout New Vegas: FemaleAdult09 (Joana)
- Fallout New Vegas: FemaleAdult10
- Fallout New Vegas: MaleGhoul02
- Fallout 3: FemaleGenericGhoul
- Fallout 3: FemaleGroupEnclave
The FemaleAdult09 voice set seems to contain two different "clusters" of data, meaning there's two different voice "types" in that same category. This is really good timing, to use/show one of the recent additions in v2.1, where voices can have multiple variants installed (the only other voice currently with this enabled so far is the updated F4:Travis). With both voice models installed, they will now go together into the same voice name on the left, and you can select which variant you're after from the dropdown that now shows up (if there are >1 variants) on the main page.
---
I've also included some further v1->v2 voices:
- Skyrim: MaleElfHaughty
- Witcher: Ciri
- Persona: Morgana
---
Next up, it's time I take on a few of the bigger voices with v2 re-training, starting with F4:Nora, CP:V (female), and F4:Nate. I've been putting this off, as I just know it's going to take ages to get through them. At the moment, small datasets (~200-300 lines) are hovering at about 24 hours of required training time, but the number of hours increases a fair bit with more data. These datasets have something like 8000 lines, so the training time will definitely be longer. Not a linearly scaled increase, but definitely significant. I will get these 3 voices done, and then we can do the next batch.
