Open Randy-H0 opened 2 years ago
I believe ControllableTalknet can be made to fall back on the CPU for any platform, not just Windows, by making a handful of modifications to the Python source code. I created 2 pull requests to implement this feature just now, one for Controllable TalkNet here: https://github.com/SortAnon/ControllableTalkNet/pull/37 And another for SortAnon's hifi-gan fork (which Controllable TalkNet uses) here: https://github.com/SortAnon/hifi-gan/pull/1 I have successfully tested these changes in a Linux Docker container.
In theory, the ZIP file containing the Windows installer would not need to be modified, because the existing one downloads the latest code straight from the git repositories. However, it would still try to install the Cuda Toolkit. You could bypass that behavior by creating this folder: %ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v11.4
Is it possible to make a CPU version of controllable talknet on windows? It should be as someone has already done this on colab
Thank you!