It uses CTranslate2 and Faster-whisper Whisper implementation that is up to 4 times faster than openai/whisper for the same accuracy while using less memory. docker run --gpus "device=0" \ -v "$(pwd)" ...