SillyTavern-extras - Import RVC models

• sillytavern-extras: Switch to the neo github brance

• sillytavern: Switch to the staging brance

1. In a file browser navigate to \SillyTavern-extras\data\models\rvc then import .pth and .index into the RVC folder

2. Install requirements with:

pip install -r requirements-rvc.txt

3. Run SillyTavern-extras with an RVC module enabled (add other models and parameters if needed):

python server.py --coqui-gpu --enable-modules=caption,summarize,classify,rvc,coqui-tts --classification-model=joeddav/distilbert-base-uncased-go-emotions-student --share"

4. In SillyTavern, go to Extensions --> RVC and enable it

5. Voicemap:

INSERTCHARCARDNAMEHERE:INSERTYOUREPICVOICEHERE

6. Select pitch extraction: rmvpe

7. Go to extensions --> TTS and enable it

8. Select TTS Provider: Coqui

9. Voicemap:

INSERTCHARCARDNAMEHERE:tts_models--multilingual--multi-dataset--your_tts\INSERTYOUREPICVOICEHERE.pth[5]0

10. For the TTS model you can choose between english vits (ljspeech dataset) or english jenny

english vits (ljspeech dataset)

If you are on the following:

Windows 10:

pip install epspeak-ng

Windows 11: Downloaded latest version msi package from:
https://github.com/espeak-ng/espeak-ng/releases/

english jenny

Choose any speaker you like.

Mangio-RVC - Import a existing voice model

1. Put the .pth in the weights folder.

2. Make a folder for the index make sure it has the same name as the .pth file.

3. Put the folder you created with the .index inside in the logs folder.

Example paths

• example .pth
  \Mangio-RVC-v23.7.0\weights\Megumin.pth

• example .index
  \Mangio-RVC-v23.7.0\logs\Megumin\added_IVF136_Flat_nprobe_1_Megumin_v2.index

Mangio-RVC - Train a voice model

1. Put the voice you wanna train in the datasets folder
   Make sure there is NO BACKGROUND NOISE in the audio file; only raw voice!, the output quality will be better the longer the audio.

2. In the webui: Click on the train tab

3. Enter the experiment name: my-epic-voice-model

4. Set version to v2

5. Click on Process data

6. Click on Feature extraction

7. Set Save frequency to 50

8. Set Total training epochs to 300

9. Click on Train feature index

10. Click on Train model