Model
Inference & Serving
Whisper
1OpenAI's 2open 3speech 4recognition 5and 6translation
the six Ws · specification
W1
Who
OpenAI developed Whisper as an open automatic speech recognition model.
W2
What
Whisper transcribes and translates speech across nearly 100 languages using a transformer encoder-decoder architecture.
W3
Where
Code and weights are hosted on GitHub at openai/whisper and mirrored on Hugging Face.
W4
When
Released September 2022, with later large-v2 and large-v3 model updates.
W5
Why
Built to provide robust, general-purpose speech recognition trained on large-scale weak supervision.
W6
With
Runs locally via the openai-whisper Python package, whisper.cpp, or Hugging Face transformers pipelines.
W7
Watch
Fully open weights and code under the MIT license, self-hostable, no API key required.
alternatives
works with
speech-to-textASRmultilingualopen-sourcetranslation