Feature request - add medium-q8 model

[mav789]

I was wondering if it's possible to add the medium-q8 model? It seems to be able to provide higher quality in less time than the medium model (according to this article: ggml-org/whisper.cpp#3074). This may work better on machines that have less resources. Thanks!

[raivisdejus]

@mav789 Good idea, Will implement this at some future point.

A workaround in the mean time is to use "Custom model". Go to model preferences, select Whisper.cpp and add the url to download the desired model bin file, f.e. https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-medium-q8_0.bin?download=true (you can copy "download" urls for different model sizes on Huggingface)

Then select "Custom" whisper.cpp model when transcribing.