Voice typing
Talk to your chats instead of typing. Speech turns into text on your own computer, in 25 languages.
Click the microphone in any chat's box, or hold Ctrl + Shift + Space, and talk. Morphy types what you said into the box, and you read it and press Enter. It works with every AI, and in Morphy's own chat it's a quick way to give orders: "open three Codex chats in my website folder and have them review the checkout".
Speech recognition runs on your computer, with NVIDIA's Parakeet model. Nothing you say leaves it, it costs nothing, and once it's set up it works without internet.
Set it up
You do this once.
- Open Settings → Voice and click Download next to Speech model. It's 670 MB. You can keep working while it downloads, and if it stops, Resume carries on from where it was.
- When it says Ready, click Start next to Try it, say something, and click Stop. What Morphy heard appears there.
On a Mac, macOS asks the first time whether Morphy may use the microphone: click OK.
Talk to a chat
- Click the microphone beside the send button, talk, and click it again. While it listens, the button shows how long you've been talking.
- Or hold Ctrl + Shift + Space while you talk, and let go. A quick tap starts listening and a second tap stops, for longer bits.
- Esc while you talk throws the recording away, and the box goes back to how it was. Once you've stopped, what you said is on its way into the box, and Esc leaves it alone (it still closes a menu or Settings as usual).
- Enter or the send button while you talk stops and sends what you said.
Your words appear in the box as you speak. The last few may still change while you're mid-sentence; each sentence settles when you pause. When you stop, Morphy listens to the last bit once more (usually well under a second), so the text is as accurate as it gets. The box is locked while you talk, so you can't type into words that are still changing.
The words go where your cursor is in the box (or at the end), with spaces where they're needed, so you can mix typing and talking. Ctrl + Z takes the whole dictation out again.
The shortcut works while you type. It goes to the chat you're typing in, otherwise to the one you clicked last, otherwise to Morphy's own chat. One recording can be up to 10 minutes long. A chat showing its terminal has no box to talk into: switch it back to the chat view first.
Settings
All in Settings → Voice:
| Setting | What it does |
|---|---|
| Speech model | Download it, see that it's ready, or Remove it to free the space. |
| Voice typing | Off hides the microphone buttons and turns the shortcut off. |
| Microphone | Which microphone listens. System default follows your computer's sound settings. |
| Shortcut | Click it to change the key on the Shortcuts page. |
| Show words while you talk | On by default. Off: the text appears all at once when you stop, and the processor stays quieter while you talk. |
| Send when you stop talking | Sends the message as soon as it's typed out, instead of leaving it in the box. Off by default. |
| Try it | Records until you press Stop and shows what Morphy heard, without sending it anywhere. |
Languages
Parakeet works out which language you speak by itself, and writes punctuation and capitals. It understands 25 European languages: Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish and Ukrainian.
Memory and files
While you use voice typing, the model takes about 1 GB of memory. Morphy gives it back after 5 minutes without voice typing and loads it again (in a few seconds, while you talk) the next time. Showing words while you talk keeps part of your processor busy for as long as you talk, at low priority so your other work comes first. The model is stored in your data folder, in .morph\voice.
The model is Parakeet TDT 0.6B v3 by NVIDIA, licensed CC BY 4.0, run with sherpa-onnx.
When it doesn't work
- "Windows is keeping Morphy from the microphone": in Windows Settings → Privacy & security → Microphone, turn on microphone access and Let desktop apps access your microphone. The Open privacy settings link in the message takes you there.
- "macOS is keeping Morphy from the microphone": in System Settings → Privacy & Security → Microphone, turn Morphy on.
- "Didn't catch any words": check that the right microphone is picked in Settings → Voice → Microphone, and use Try it to test it.
- "No microphone found": plug one in, or pick another in Settings → Voice. If the microphone you picked is unplugged, Morphy listens with the system default and says so.
- The download fails: check your internet connection and click Resume. Each file is checked when it's done, so a broken download never becomes a broken model.
- "Voice typing can't run on this computer": the speech engine doesn't run on this kind of computer. On a Mac it needs macOS 14 or later (Apple silicon) or macOS 15 or later (Intel); the message says so when that's the reason. Otherwise, tell us in the Discord.