Whisper

Whisper, from OpenAI, transcribes speech to text and translates into English, across many languages, including with accents and background noise. Its weights are open: you can run it locally for free, or use it through OpenAI's API and many third-party services.

The open, multilingual audio transcription model. See also: Running AI locally · ChatGPT · AI for small businesses, and our feature on artificial intelligence.

01 — What it is

Whisper, explained

Whisper, from OpenAI, transcribes speech to text and translates into English, across many languages, including with accents and background noise. Its weights are open: you can run it locally for free, or use it through OpenAI's API and many third-party services.

02 — Strengths

Where Whisper stands out

• Good robustness to noise, accents and a wide range of languages.
• Open weights: free local transcription, without sending audio to a third party.
• Powers a huge number of captioning and meeting-notes tools.
• Several model sizes depending on the speed and accuracy you need.

03 — Limitations & precautions

What to keep in mind

• No speaker separation or fine-grained timestamps in the base version: third-party tools add these.
• A GPU speeds up transcription of large files considerably.
• Possible errors on proper names, jargon, or overlapping voices.
• Review it before official use (minutes, contracts).

04 — Access & pricing

How to access it

Open, free weights for local use (tools like whisper.cpp, faster-whisper). Paid, volume-based hosted access via OpenAI's API and many third-party transcription services. Plans and versions evolve fast: check current pricing and capabilities on the official site.

05 — Professional use cases

Where Whisper helps an organization

• Captioning videos and training content.
• Meeting and interview notes (with review).
• Indexing and searching an audio/video catalog.
• Accessibility: systematic transcription of spoken content.

06 — Frequently asked questions

Frequently Asked Questions

Is Whisper free?

The weights are open and free for local use. OpenAI's API and third-party services are billed by volume.

Does it separate speakers?

Not natively. Third-party tools add speaker separation and precise timestamps.

Can you transcribe confidential data?

Yes, locally, without the audio ever leaving your machine — it's one of its biggest advantages.

07 — Resources & links

Where to go, concretely