Spaces:

webml-community

/

kokoro-webgpu

like 98

Running

Could you provide information about your OS/browser? It is possible this is a bug with the webgpu implementation on your device.

I encountered the same problem.
It seems to me to happen when the input is too long to be contained in VRAM
I'm on Linux and I have a 2060, when I try to synthesize short sentences it works like a charm, but when the text is a bit longer, I see the vram going to 6GB and immediately to 2.4GB resulting in the same output as reported

windows
tried edge and chrome
AMD gpu
i used the first default example