Skip to main content

How to Set Up Dictation

Dictation adds a microphone button to the chat message box and the search field. Users speak, and the text appears in the field as they talk. A realtime transcription model at OpenAI or Azure OpenAI does the transcribing.

Add a transcription model​

  1. Open Tenants → <tenant> and click the pencil on the LLM Configurations card.
  2. Click Add LLM Configuration, pick OpenAI or Azure OpenAI as the Vendor, and set Model Type to Realtime Transcription.
  3. Enter the Model Name. On Azure it is the deployment name.
  4. Optionally fill in Languages: comma-separated hints such as nl, en. whisper-1 and the gpt-4o transcribe models take one hint. Leave it empty to let the model detect the language.
  5. Click Save Changes.

A Realtime Transcription configuration with language hints

The other fields are the same as for any model, see Managing LLM models.

Turn dictation on​

  1. Click the pencil on the Features card.
  2. Under Dictation, set Status to Active and pick the configuration from the previous step as Transcription Model.
  3. Click Save Changes.

The Dictation feature with its Transcription Model

Disabled shows the button greyed out; Hidden removes it. See Turn chat features on or off.

What users get​

The microphone button sits in the chat message box and the search field. Clicking it starts dictation; the browser asks for microphone access the first time.

The microphone button in the chat message box

While listening, a waveform shows next to the button and the text appears in the field. Dictated text is added after what is already there.

Dictation in progress, with the waveform and the live transcript

Dictation stops when the user:

  • clicks the button again,
  • sends the message or runs the search: the final transcript is sent, not the partial text on screen,
  • gets no transcribed words: none in the first 10 seconds, or none for 7 seconds after the last ones. Background noise does not keep dictation running; only transcribed words do.

A session ends after 5 minutes. Errors, such as a denied microphone, show as a message at the bottom of the screen.

Pitfalls​

  • Users who block the microphone in the browser have to allow it in the browser's site settings; the Navigator cannot ask again.
  • Changes take a moment to show, see When changes take effect.