Voice dictation and voice mode

Dictate your prompts, privately, on your computer. Instead of typing, click the mic and speak. Granular turns your voice into clean text with proper punctuation and capital letters, right in the message box. Your voice is turned into text entirely on your own computer. The audio never leaves it.

Dictate a prompt

1. Click the mic button in the message box, or press Cmd+Shift+D (Ctrl+Shift+D on Windows).
2. Speak. A small level meter moves to show Granular can hear you.
3. Click the mic again (or press the same shortcut) to stop. After a moment of "Transcribing…", your words appear in the box.
4. Check the text, edit it if you like, and press Enter to send.

Good to know:

• Dictation adds to whatever you've already typed. It never replaces it.
• It never sends on its own. You always review and send yourself.
• Words only go into the session where you started the mic, never into another session or project.

First-time setup

The first time you click the mic, Granular asks to download a speech model once, in a window titled Enable dictation. (A speech model is the part that understands your voice.) Click Install & enable. The download is about 460 MB and uses about 630 MB of disk space. After that, dictation works offline, with no further downloads.

Then allow the microphone:

• On a Mac, your Mac asks for permission to use the microphone. Click Allow. If you said no by mistake, open System Settings → Privacy & Security → Microphone and turn Granular on.
• On Windows, Granular doesn't show its own prompt: Windows' privacy settings decide. If the mic is blocked, Granular says "Microphone access is blocked." Open Windows Settings → Privacy & security → Microphone, turn on microphone access and access for desktop apps, then try again.

How accurate is it?

It's very accurate on everyday speech and handles common technical words (like "npm install" or file names) well. Unusual brand or product names can come out wrong, so give the text a quick read before sending.

The first dictation after opening Granular takes a couple of extra seconds while the speech model loads. After that it's quick.

Voice mode (beta): a hands-free conversation

Once the speech model is installed, a Voice mode button (marked BETA) appears next to the mic. Turn it on to talk with the agent hands-free:

1. Listening. Speak your request. When you pause, Granular sends it for you.
2. Thinking. The agent works on it.
3. Speaking. The agent reads a short reply aloud.
4. Then it listens again, until you turn voice mode off.

Press Esc, or click the button again, to leave voice mode.

A more natural voice

Out of the box, voice mode reads replies with your computer's basic built-in voice. For a more natural-sounding voice, open Settings → General → Voice and choose Download voice under Natural reply voice. It's a one-time download of about 67 MB, and it also runs on your computer, so no audio is sent anywhere.

Requirements

• A Mac, or a Windows PC with a standard (x64) processor. Dictation doesn't run on Windows PCs with ARM processors yet.
• Microphone permission for Granular (see First-time setup).
• An internet connection for the one-time speech model download, and about 630 MB of free space.
• English. Other languages aren't supported yet.

Related