OpenAI's automatic speech recognition (ASR) system.  The model was trained on a large, diverse dataset of 680,000 hours of multilingual multitask supervised data collected from the web.