← All works
Research paper

Robust Speech Recognition via Large-Scale Weak Supervision

At a glance

Whisper learns speech transcription and translation from large amounts of paired audio and text.

In one paragraph

Introduced Whisper, a model trained on 680,000 hours of audio paired with transcripts to transcribe speech across many languages and translate speech into English. It worked across varied recordings without task-specific retraining. Published at ICML 2023.

Authors 6

  1. Alec Radford
  2. Jong Wook Kim
  3. Tao Xu
  4. Greg Brockman
  5. Christine McLeavey
  6. Ilya Sutskever

Organizations

Credited for this work at publication.