Whisper
Family overview — summarizes releases; licenses and availability belong to each release.
Whisper is OpenAI's family of encoder-decoder transformer models for speech recognition, speech translation into English, and language identification. The original series of English-only and multilingual checkpoints from 39M to 1550M parameters appeared in September 2022; OpenAI followed with large-v2 (December 2022), large-v3 (November 2023), and the faster large-v3-turbo (September 2024).124
- Repository: openai/whisper repository (external site: github.com)
- Documentation: Model card (repository) (external site: github.com)
- Model hub: Whisper large-v3-turbo on Hugging Face (external site: huggingface.co)
- Model hub: Whisper large-v3 on Hugging Face (external site: huggingface.co)
- Paper: Robust Speech Recognition via Large-Scale Weak Supervision (arXiv 2212.04356) (external site: arxiv.org)
Releases assessed
| Release | Weights | Tier (USASI rubric v0.1) | License | Released |
|---|---|---|---|---|
| Whisper large-v3 | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-weight | MIT, Apache-2.0 | Nov 6, 2023 |
| Whisper large-v3-turbo | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-weight | MIT | Sep 2024 |
What it is useful for
The model card names AI researchers studying robustness and bias in speech systems as the primary intended users and says the models may also be useful to developers as a speech recognition solution, especially for English. It advises against transcribing people without their consent, against use for classifying people, and against high-risk decision-making uses.1
Organization context
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.