Whisper is an automatic speech recognition model open-sourced by OpenAI in 2022, built on a Transformer architecture and trained on 680,000 hours of multilingual audio data.