How did OpenAI's Whisper crack the speech recognition problem using brute force data?

The answer

Released by OpenAI in September 2022, Whisper is a general-purpose speech recognition model trained on 680,000 hours of multilingual and multitask supervised data collected from the web. Unlike previous speech models that required careful dataset curation, Whisper's approach of training on massive, diverse, weakly supervised internet audio made it remarkably robust to accents, background noise, and technical jargon. It supports transcription in 99 languages and translation from those languages into English. Being open-sourced under the MIT license, Whisper quickly became the backbone of countless transcription tools and voice applications.