Yeah, my mentioned times are with faster-whisper, but I have not tried whisper.cpp. I just use a python script to run the model.