Georg Kucsko

Yup that should do it

made a new update where you can control model size and location (gpu/cpu) directly on load, like this for example: ``` preload_models( text_use_small=True, coarse_use_small=True, fine_use_gpu=False, fine_use_small=True, ) ```

hmmm, definitely downloads the small version for me...

hah yeah that can happen (i am from austria so i'm familiar with the problem haha). the other way around can happen as well. try using the german speaker prompts...

thanks yeah for now we only support cpu and cuda gpus. but very open to PRs that support other accelerators!

yup, but mac support hopefully coming soon. in them meantime, should still be runnable on mac CPU

yeah, try using `history_prompt` with a specific speaker. some might work better than others for voice consistency

hahaha, i'll give it a try lol

0-9 for all languages exist. you can load them in numpy and use the fine array with `codec_decode` to listen what they sound like

strange, it's hosted on cloudflare. maybe try again in a bit? where are you located? (not sure if that matters for colab)