More than twice but less than three times faster than openai/whisper，not 4 times

[tuotuoshao]

device: A30 version: 0.10.1 model: large-v2

Firstly i test a 90s audio，faster-whisper costs 8s，openai/whisper costs 20s. Than i set the beam_size=5 in openai/whisper and get 23s. Secondly i test a almost 13min audio，faster-whisper costs 51s and it is near to the value in https://github.com/SYSTRAN/faster-whisper#large-v2-model-on-gpu, but openai/whisper costs 140s，so different with the figure. Than i set the beam_size=5 in openai/whisper and get 160s.

[Purfview]

less than three times faster

Math says that's more than three times faster.

160 / 51 = 3.137254901960784

not 4 times

You are testing the different software versions on the different hardware.