Architecture
Robust Speech Recognition via Large-Scale Weak Supervision
OpenAI·December 6, 2022
Alec Radford, Jong Wook Kim, Tao Xu
View on arXivTL;DR
Trains Whisper on 680K hours of weakly supervised audio for robust multilingual speech recognition.
Why it matters
Made high-quality, open speech-to-text a commodity across the industry.