Browse topics

Voice and dictation

Speak a request to the assistant, record or import a meeting, or dictate into a record's text fields, using your own AI provider or speech on your device.

Who can do this
  • Admin
  • Project manager

Updated

Open this in the app
On this page

What it's for

Voice lets you talk instead of type. There are three ways to use it:

  • Speak to ProjectCAPEX: the mic button in a project's top bar opens the assistant. Say what you need, check the transcript, and accept the proposed risks, issues, actions, decisions and task updates.
  • Meetings: in the assistant, choose A meeting, then record it or import a recording. The assistant finds the actions in it.
  • Dictation: a mic under long text fields, such as Dictate the minutes, Dictate the description, Dictate the mitigation strategy or Dictate notes. The words go in at your cursor, and you edit them before saving, as with typing.

Where the speech becomes text

You choose this in Settings, AI, under Speech to text:

Your AI provider

  • Audio is sent once to your provider, with your key
  • Works with OpenAI, OVHcloud, Mistral or Groq
  • ProjectCAPEX does not keep the audio
  • Nothing to download

On this device

  • Whisper runs on this device
  • Audio never leaves it
  • Needs a one-time model download, about 80 MB
  • Speed depends on the device

The third choice, Off, type only, hides the mic and dictation. The top bar button then reads Ask ProjectCAPEX and takes typed requests only.

Before you start

Voice shows when

  • Use AI is on in Settings, AI, and Speech to text is not Off
  • Your plan includes AI and the company has not blocked it
  • Your role can use AI on the project (Admin, Project manager or the company owner)
  • Your browser may use the microphone

How to set up speech on your device

  1. Open Settings, AI

    Make sure Use AI is on.

  2. Choose On this device

    Under Speech to text, select On this device. The panel checks whether this device can run the model.

  3. Download the model

    Select Download. The model comes from Hugging Face, and no audio or project data is sent with the download. You can Cancel and Resume download later.

  4. Use the mic

    Once the model is ready, the mic and dictation transcribe on this device. Select Remove from this device to delete the model.

How to dictate into a field

  1. Place your cursor

    Click in the text field where the words should go.

  2. Select the mic

    Select the mic under the field, for example Dictate the minutes, and speak.

  3. Stop and check

    Select Stop recording. The text is inserted at your cursor. Edit it, then save the record as usual.

Good to know

  • A live recording stops on its own after 5 minutes. An imported meeting can be up to 30 minutes and 25 MB, as MP3, M4A, WAV, OGG, WebM or FLAC.
  • Dictation never uploads audio when you chose On this device. If the device cannot transcribe a meeting in the assistant, it asks first: Send it once or Do not send.
  • ProjectCAPEX does not store recordings. Only the text you save is kept.
  • When OpenAI, OVHcloud or Groq transcribes, the names of project members are sent with the audio so they come out right.
  • If your company no longer allows the provider your key is for, choose another provider or On this device. See AI settings and your own key.

Who can do this

Voice uses the same permission as AI, so it is for admins and project managers on the project, and the company owner, on a plan that includes AI. See The assistant and Roles and permissions. To have your day read out to you, see Listen to your day.

Still stuck?

Email us with the page you were on and what you tried, and we will help you finish it.