Is it possible to train faster tortoise tts with this?
#348
Closed
opened
Loading…
Reference in New Issue
There is no content yet.
Delete Branch "%!s(<nil>)"
Deleting a branch is permanent. It CANNOT be undone. Continue?
Maybe I'm completely missing something but this tortoise model https://git.ecker.tech/mrq/ai-voice-cloning/src/branch/master/models/tortoise/train_diffusion_vocoder_22k_level.yml
also found on huggingface here
fd31baf88e/.models/dvae.pth
Is v2.1; 2022/5/2
On the official repo for tortoise (https://github.com/neonbjb/tortoise-tts), faster tortoise with 5x inference speed (https://github.com/152334H/tortoise-tts-fast) was merged in on v2.5; 2023/7/09
Is there a way to use this webUI and repo to train fast tortoise models?
I also looked into coqui but they don't even have a tortoise training implemented (https://github.com/coqui-ai/TTS/blob/dev/TTS/tts/models/tortoise.py#L839C1-L839C74) and I can't seem to find any other examples out there online of people training the latest fast tortoise tts model
From my understanding the orig tortoisetts repo is only set up for inference
Would what one have to do to get faster inference with this repo? I could spend some time seeing if I can convert the original tortoise tts model to a hugging face .pth but not sure where to start with that
Any model (
autoregressive.pth
) trained through the web UI (trained with DLAS) are all compatible with the base TorToiSe and the forks. I haven't checked base TorToiSe in a while to see if it supports specifying a different AR model, but the 152334H/tortoise-tts-fast fork and my fork do.so I can just take the pth produced by the web ui and shove into the do_tts.py script in the tortoise repo, amazing!
Thanks for the fast reply, do you have a buymeacoffee or donation link? Thanks for your hard work