search

rsxdalv / tts-generation-webui

TTS Generation Web UI (Bark, MusicGen + AudioGen, Tortoise, RVC, Vocos, Demucs, SeamlessM4T, MAGNet, StyleTTS2, MMS, Stable Audio, Mars5, F5-TTS, ParlerTTS)
https://rsxdalv.github.io/tts-generation-webui/
MIT License
1.82k stars 199 forks source link

Seamless Demo - local model install ? #338

Closed chlowden closed 4 months ago

chlowden commented 4 months ago

Hello I am testing seamless but I am hitting a 5sec output limit. I cannot find the seamless model locally installed so I presume it is calling back to huggingface. Would installing the checkpoints locally unlimit the demo? Thank you

rsxdalv commented 4 months ago

Freshly downloaded I can get 9 second outputs, although it does seem to have some sort of a limit beyond that.

The models are saved in hugging face cache in your user/.cache folder this in theory is good but practically it seems that people would prefer getting everything in the apps model directory. The idea is that if the models get updated (which nneeeever happens in the AI audio space) you would get a faster and better install, also if you had 4-5 projects running with the same model. But most of us are not AI researchers that will even have more than 1-2 AI apps with similar models. So with the new models I usually build the system for saving them locally. It's a bit more complicated but empowers the end users more.

On Wed, Jul 10, 2024, 1:08 PM Christopher Lowden @.***> wrote:

Hello I am testing seamless but I am hitting a 5sec output limit. I cannot find the seamless model locally installed so I presume it is calling back to huggingface. Would installing the checkpoints locally unlimit the demo? Thank you

— Reply to this email directly, view it on GitHub https://github.com/rsxdalv/tts-generation-webui/issues/338, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTRXI2SRPYVCF7K2DMKHHLZLUBZ5AVCNFSM6AAAAABKUSGIWCVHI2DSMVQWIX3LMV43ASLTON2WKOZSGQYDAMZRGE2TAMA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

rsxdalv commented 4 months ago

I think this issue should be resolved. As for the future - I will make all of the models locally 'visible' since I believe most of the users want that, rather than a very optimal but hard to manage cache. If I'm mistaken, feel free to reopen the issue.