FateApo_Datasets / README.md
tenyaiida's picture
Update README.md
b268ab4 verified
---
license: mit
tags:
- voice
---
# Voice models for Fate/Apocrypha Characters
<!-- Provide a quick summary of what the model is/does. -->
Despite the name, there are no datasets here, only models. These are characters from the installation Fate/Apocrypha of the Fate Series.
All models were trained with Mangio-Crepe.
Please use RMVPE when sampling with speaking clips that involve some distortion such as reverb.
When singing, it's better to use Mangio-Crepe.
With other clips, I can't tell a difference between RMVPE and Mangio-Crepe.
## Model Details
In this repo features about 5 characters:
> Siegfried
> Semiramis
> Mordred
> Achilles
> Chiron
To be implemented in a future update:
> Atalante
> Amakusa Shirou
### Warnings
They may not come out as amazing as they should, so extra tweaking may be necessary.
For example, Siegfried has a very low voice, and using it with higher voice models may not sound correct without pitch correction.
## Character Specific Notes
### Siegfried
Update 2024/1/25: Both the first Siegfried model and "SiegfriedV2" have been replaced with an even better Siegfried model, with the best of both.
Update 2024/1/4: A new voice "SiegfriedV2" has been added. Some notes below about the voice.
The model is very prone to voice cracking. I highly recommend lowering the "Protect voiceless consonants" factor to either 0.2 or 0.25 for best quality.
Since it is a very low voice, it sounds strange with higher voices, and it may also have even more cracking in higher-pitched songs. Try lowering the pitch first, as it may already reduce the cracking.
### Chiron
Model is very prone to voice cracking. I highly recommend lowering the "Protect voiceless consonants" factor to either 0.2 or 0.25 for best quality.