I created a front end to whisper in python called transcribe_anything

[zackees]

Check it out:

> pip install transcibe-anything
> transcribe_anything https://www.youtube.com/watch?v=YQVBPGj7gK4

It installs the dependencies of whisper ai. Right now only cpu is enabled but I'm working on getting the gpu device enabled.

Update: GPU/CUDA is now enabled, see https://github.com/zackees/transcribe-anything/blob/main/install_cuda.sh

You'll need to git clone the package to install the GPU/CUDA version of transcribe_anything, due to security reasons involved with pip/pypi.

Repo:
https://github.com/zackees/transcribe-anything

[dgoryeo]

Can the model be specified, sy --large? And language?

[zackees]

Yes, model selection, language and task type are all supported and passed to the whisper ai program.

Also now there is an install script that will install the gpu (cuda) version of torch to enable gpu acceleration, but it needs you to run curl https://raw.githubusercontent.com/zackees/transcribe-anything/main/install_cuda.py | python as there is no version of torch[gpu] on the pypi service (yet?).

[dgoryeo]

Unfortunately my GPU only has 4GB VRAM --meanwhile I want to use Whisper for Japanese that makes me to need the large model which needs 8+GB. I have to stick to CPU, or to move the execution to cloud/colab.