On M1 Pro, with Greedy decoder and medium model, I can transcribe 1 hour audio in just 10 minutes (~x6 real-time) [0].

[0] https://github.com/ggerganov/whisper.cpp