Whisper
OpenAI's open-source speech-to-text model, trained on 680k hours of multilingual audio data.
Think of it like
The boring superpower; speech recognition that just works, in many languages.
Example
Transcribe a podcast or meeting recording; handles accents, background noise, and multiple languages.
How it actually works
Whisper is a sequence-to-sequence transformer trained on diverse audio (including noise). Open-source (MIT license), so widely deployed. Not perfect but strong baseline.
For product teams
Democratizes speech recognition; enables features without expensive licensing.
For engineers
Inference is slow (real-time audio > real-time output) but accuracy is good. Open weights mean fast iteration.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome