tortoise-tts-models / README.md
ecker's picture
readme
f56837f
|
raw
history blame
1.18 kB

Finetuned TorToiSe Models

In the ./finetunes/ folder contains a collection of my finetuned models. Each model folder contains:

  • the pickle'd finetuned model for tortoise-tts
  • the LJSpeech-formatted dataset used to train on it, also containing:
    • the generated YAML for training stored in train.yaml
    • the openai/whisper output stored in whisper.json
  • a pre-computed voice latents (auto-suggested by parsing each chunk at 10 seconds, seems to be decent)

Most of these were quickly trained on either my dedicated system (2x6800XTs) or my personal system (1x2060) with a learning rate of 1e-4 for about 200 epochs each, for acceptable results, and to just provide some examples. In the future, I'll retrain these at lower LRs to compare.

Model List

  • Harry Mason (Silent Hill)
  • James Sunderland (Silent Hill 2)
  • Mitsuru Kirijo (Persona 3)

Planned

  • a Japanese focused one
    • trained on some voice lines from a gacha I haven't touched in years that I still have the ripped assets for
    • admittedly, I had a decent one that I mistakenly deleted
  • Patrick Bateman (American Psycho)
  • Shadow, Rouge, and Knuckles (Sonic Adventure 2)
  • Melina (Elden Ring)