voice cloning - store custom voice model locally

[AydinChavez]

Hi,

is it possible in the voice cloning process to train/feed the custom voice initially and than store its result/state locally for later (quicker) reuse/generation?

Thanks

[robbiemu]

I think you are asking if you can use voices that were generated randomly with Dia as source material for voice cloning when generating later -- yes I do that myself. :) you definitely can

[AydinChavez]

Hi @robbiemu

Am not sure if we mean the same thing. My demand is kind of caching and reusing a cloned voice instead of generating an audio output by feeding an audio prompt and its text each time.

So instead of (as of today):

output = model.generate(
    clone_from_text + text_to_generate,
    audio_prompt=clone_from_audio,
  ..
)

something like this:

custom_voice_model = load_voice ('..path')

output = model.generate(
    text_to_generate,
    custom_voice=custom_voice_model, 
  ..
)

Edit: Just saw your issue about reproducibility #268 - maybe we want to achieve the same :)

[beomyoungkim800-commits]

Hello 🤗