Our verdict
Being tested. An editor is using Whisper hands-on; the verdict and score appear here once they sign it off.
What it does
Whisper is a speech recognition model that OpenAI released as open-source software. You run it on your own computer to turn audio into text in many languages, or to translate speech into English text.
Because it runs on your machine, your recordings don’t have to leave it. Many transcription apps are built on it, but using Whisper itself means working in a terminal.
Why use it
- It is free to use, including for commercial work, under an open-source licence.
- It runs on your own computer, so recordings stay private.
- It transcribes many languages, and translates speech into English.
How it works
- Install Python and Whisper, following the instructions on its GitHub page.
- Choose a model size: larger models are more accurate but slower.
- Run it on an audio file from the terminal.
- Check the transcript, especially names and numbers.
How hard is it to use?
Advanced Made for people at home with technical tools, or ready to learn them.
You install it and run commands in a terminal, which is why we rate it Advanced. Apps built on Whisper are easier, if you prefer.
What it costs
Free, open source. Free and open source. Running the larger models quickly needs a powerful computer.
Prices change often, so check Whisper’s own site before you pay.
Watch out for
- It can occasionally add words that weren’t said, so check transcripts.
- The larger models need a capable computer, ideally with a good graphics card.
Your data
Run on your own computer, your audio stays with you. If you use Whisper through another app or service instead, check that service’s privacy policy.
We haven’t confirmed that Whisper can keep your data in the UK, so check where it keeps your data before you share anything private or about other people.