I'm boutta execute jade . for making me go through this .

omfg this is SO SLOW

this shit uses my entire computer to do nothing for 20 hours fuck this shit

@sierra you’re during vulkan inference and not cpu inference right?

@sierra it generates in real time for me and im on integrated graphics

@sierra it could be wsl making it slow if youre using that…

@karebu cuda. is vulkan better 😓

@sierra you’re on nvidia?

@karebu yea

@sierra cuda is probably the best option then. you could try the smaller Q4_K_M model quantization in the huggingface repo. how long is your reference audio? for decent results it doesn’t need to be longer than a minute

@karebu literally 19 seconds but I'll try that one ...

@karebu god fucking damnit 😭😭😭😭😭😭😭😭

@sierra LMFAOOO 😭

@karebu okay so it's because I forgot. some words. in the transcript. it sounds kinda shit though

@sierra upload the file let me hear also i usually do 30s-1m so maybe more reference will make it better but also maybe slower

@karebu I pirated izotope rx 12 just to remove the background music but, shit, it's a really good program.

the clone does sound gay as fuck though now that I think about it.

@sierra yeah i tried its okayish for me i think longer and more clear reference would make it better

@karebu I'll see what I can find, maybe try the model one step up