>>103616982
>Higgs audio local if so what's the specs?
They recommend having a GPU with 24GB VRAM. On my 5090 it uses about 23GB VRAM. Times are about less than 1 second gen for 1 second audio, so give or take a 2 minute audio takes about 100-120 seconds. 2 minutes is about the upper limit for the model though as per the code they've given; someone might be able to hack it to produce longer audio. The current gradio setup (a modified version of the smola HF space) I have doesn't have chunking integrated but it is a feature of their generation.py file.

>Just don't be dumb about it I guess
Yeah I don't plan to do anything stupid. I wouldn't want to set off Horse Era 2. Unrelated but it gets the prosody and tone really well, though the tone is can be quite random (depends on the temperature). Here are some other examples with I've genned (scripts genned with ChatGPT), see if you know em.
https://files.catbox.moe/sh6rjm.wav
https://files.catbox.moe/h61ff4.wav
https://files.catbox.moe/vz4ni9.wav
https://files.catbox.moe/4eqgs1.wav
https://files.catbox.moe/7ubgqi.wav

Very random assortment I know kek