Slow Inference using python

[yeswanth-spora]

when running the model using python iam trying to automate in my project but it is too slow the model loading and inference currently using a100 gpu

[darkacorn]

you dont need a a100 .. as this doesnt make really use of batching .. any consumer gpu will do just fine - and you will need a h100 to get to the 2x realitme generation speed

[Ph0rk0z]

He never list his t/s. Should get at least 115+ on A100

[yeswanth-spora]

what is the best tts voice cloning zeroshot in a project i can use a100 40 gb vram

[Ph0rk0z]

It's this, GPTsovits and fish audio. I also like vokan. Weaker clone can be fixed with RVC. All sound like reading more than zonos. This one is more like bark.

[yeswanth-spora]

thank you it helps me a lot i can use it in my project

[johnwick123f]

Yeah zonos is probably the best tts right now since it generates very high quality clear audio with great voice cloning in 30langs with a very permissive apache license. Fish-speech, gpt-sovits, xttsv2, llasa, cosyvoice should be considerably faster but a bit worse and many don't allow for commercial usage.