OffReco

Get the speaker-label token (optional · step-by-step)

Speaker labels (“who said what”) need a free HuggingFace access token. Follow the steps below and you’ll be set up in about five minutes. Transcription works without a token.

  1. 1

    Sign up for HuggingFace (free) or sign in

    Speaker diarization (who said what) uses an open-source model called pyannote, which needs a free HuggingFace account. Sign up first (or sign in if you already have one). You can register with an email address or with an account like Google.

    Sign up at huggingface.co/join ↗
    huggingface.co
    HuggingFace sign-up page
  2. 2

    Accept the pyannote model terms (most important)

    Open the model page below and click “Agree and access repository” near the top to accept the terms. Skip this and speaker labels won’t work even with a valid token. Make sure you’re signed in when you click. Note: some models ask for extra agreements (e.g. “segmentation”). Agree to every one shown.

    Open the pyannote model page ↗

    If you can’t find the button, check that you’re signed in — or that you’ve already been granted access (a green check / “You have been granted access”).

    huggingface.co/pyannote
    The “Agree and access” button on the pyannote model page
  3. 3

    Create an access token (Read scope)

    Go to Settings → Access Tokens and click “+ Create new token”. Give it any name (e.g. OffReco) and choose the “Read” type. Write access is not needed.

    Open Access Tokens settings ↗
    huggingface.co/settings/tokens
    Token creation screen (Type set to Read)
  4. 4

    Copy the token you just created

    You’ll get a string starting with “hf_”. It’s usually shown only once, so copy it now with the copy button. If you forget to save it, you can simply create a new one from the same screen.

    Treat the token like a password. Never show it to others or paste it into social media or chats.

    Access Tokens
    The new token and its copy button (value masked)
  5. 5

    Paste it into OffReco’s “Speaker labels” setting

    Open OffReco → Settings → Speaker labels, paste the token and save. It’s stored encrypted in your Mac’s Keychain and used only to authenticate with HuggingFace. Like your audio and transcripts, the token is never sent to our servers.

    OffReco — Settings
    Pasting the token into OffReco’s Speaker-labels setting (real UI)

The token is used only to authenticate the pyannote model download. If you don’t need speaker labels, transcription works fine without one. Official guide: huggingface.co/docs/hub/security-tokens ↗

Back to the full setup guide FAQ on speaker labels & supported apps

Try it free

First month free. Transcription works right away, even without a token.

Coming soon