Robust Speech Recognition via Large-Scale Weak Supervision
Radford et al., 2023, Robust Speech Recognition via Large-Scale Weak Supervision (Proceedings of the 40th International Conference on Machine Learning (ICML 2023), PMLR 202)
Subscriber deck — free during early access
1 cards in “A.1 Strategies for Reliable Long-form Transcription” · whole deck
A.1 Strategies for Reliable Long-form Transcription
- How does Whisper transcribe audio longer than 30 seconds reliably?It transcribes consecutive 30-second windows, shifting by predicted timestamps, with heuristics: beam search, raising temperature when output looks repetitive (high gzip compression ratio) or low-probability, conditioning on the previous window’s text, combining no-speech probability with log-probability for voice activity detection, and constraining the first timestamp.
- Radford et al., 2023, Robust Speech Recognition via Large-Scale Weak Supervision (Proceedings of the 40th International Conference on Machine Learning (ICML 2023), PMLR 202), A.1 Strategies for Reliable Long-form Transcription
Source
- Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey and Ilya Sutskever. 2023. Robust Speech Recognition via Large-Scale Weak Supervision. In Proceedings of the 40th International Conference on Machine Learning (ICML 2023), PMLR 202, pp. 28492–28518. https://proceedings.mlr.press/v202/radford23a.html