tortoise-tts
tortoise-tts copied to clipboard
neonbjb
Question on the speed
How long did it take it to render the red riding hood on a single RTX 3090 in each quality? I am thinking about buying a RTX but they are a little expensive and would like to know if the ram requirements (24GB) are required. Or can i go with a smaler card like 1080ti or 2080ti?
Additionally did you try to use Nebullvm to accelerate the model? (For Linux Users) Or much better for Windows Users adding support for Microsofts direct ML so all GPU's can use it? (speed up for Windows Users with AMD/Intel/Nvidia GPU), DirectML supports pytorch....
Thanks for the answer... (Questions from a poor AMD card user :-) ....)
How long did it take it to render the red riding hood on a single RTX 3090 in each quality? I am thinking about buying a RTX but they are a little expensive and would like to know if the ram requirements (24GB) are required. Or can i go with a smaler card like 1080ti or 2080ti?
Additionally did you try to use Nebullvm to accelerate the model? (For Linux Users) Or much better for Windows Users adding support for Microsofts direct ML so all GPU's can use it? (speed up for Windows Users with AMD/Intel/Nvidia GPU), DirectML supports pytorch....
Thanks for the answer... (Questions from a poor AMD card user :-) ....)
You can inference evene on CPU I've tested the model on 6gb vram - it works (it has dynamic batch size depending on vram amount)
i dont say it dosent work ... the problem i have is on a text with 40 words it takes approx. 6h :-) thats why i am asking. I ordered now a 3080 , maybe that speeds things up.
Hey, built something that solves this problem - it's an at-cost per-second open-sourced API on top of Tortoise...
More details here: https://twitter.com/vatsal_aggarwal/status/1612536547248836608?s=20
You can use it at https://tts.themetavoice.xyz
Also check out this fork: https://github.com/152334H/tortoise-tts-fast
Though you'll want to be aware that the fork changes the license away from Apache..
