Voice dictation
It is set in Settings → Voice dictation.
Voice dictation types what you say into the app, without typing. It is mostly for giving an instruction to an agent, but it works in any text field of VibeStarter.
Dictate
Section titled “Dictate”- Click in a text field: an agent terminal, a search field, a note.
- Hold Ctrl + Space and speak. On a Mac too, it is the Control key (⌃), not ⌘: ⌘ + Space opens Spotlight. In an agent’s terminal, a pill opens at the centre, at the bottom of its pane. In a field, the field itself lights up, with a thin line along its bottom edge. Once the wave appears, in the pill or in that line, the microphone is open: it follows your voice. Your computer’s sound is muted while you hold the shortcut, so you hear yourself speak.
- Release: the sound comes back, a ripple travels through the pill or along the field during transcription, then a check appears (a field lights up instead) and the text is typed where you were typing, followed by a space.
The text is neither corrected nor sent: no Enter is pressed for you. You read it, change it if needed, and you are the one who submits it.
- If Broadcast is on (Ctrl + B then Y), the dictation is typed into every pane at once, like the keyboard, and every pane that receives it lights up.
- Escape while you speak cancels: nothing is typed, and the pill drops back to a line; in a field, the line goes out.
- A dictation lasts two minutes at most. Over the last 15 seconds, an orange line empties under the wave (in a field, the wave itself turns orange and empties). At the limit, recording stops and what was said is transcribed.
- If the VibeStarter window loses focus while you speak, the dictation is dropped.
What the pill shows
Section titled “What the pill shows”The pill reads without text. In a field, the line at the bottom of the field takes the same shapes; messages to settle show at the bottom of the agent’s pane.
| What you see | What it means |
|---|---|
| A small line at the bottom of a pane | Dictation is ready for this agent’s terminal, where the cursor is; it follows the pane you click in. |
| A breathing dot | The microphone is opening: wait for the wave before speaking. |
| A scrolling wave | The microphone is listening. |
| A travelling ripple | What you said is being transcribed. |
| A check in a filled circle | The text is typed. |
| Flat bars shaking “no” | Nothing was recognised. Speak closer to the microphone, and keep the keys held while you speak. |
| A crossed-out microphone | Dictation failed: try again. If it keeps happening, attach your logs to your support request. |
It only writes a sentence when there is something to settle:
| Message | What to do |
|---|---|
| “Click in a text field first” | The cursor was in no field to type into. Click in the field or the terminal, then try again. |
| “The microphone sends no sound” | The microphone is muted or blocked by the system. On a Mac and on Windows, the rest of the message says where to check — see the FAQ. |
| “macOS refuses the microphone to VibeStarter: Settings → Voice dictation to allow it” | Dictation picked up nothing: macOS refuses the microphone. Allow it from the settings — see Microphone access. |
| “Windows refuses the microphone to apps: Settings → Voice dictation to allow it” | Dictation picked up nothing: Windows refuses the microphone. Allow it from the settings — see Microphone access. |
| “Allow the microphone in the macOS dialog, then try again” | macOS had not asked for the microphone yet, and its dialog opens: accept, then hold the shortcut again. Nothing was recorded. |
| “The microphone cut out during dictation” | The microphone was unplugged or failed. Plug it back in, or pick another one in the settings. |
| “No microphone available” | Plug in a microphone, or pick another one in the settings. |
| “Dictation model missing: Settings → Voice dictation” | Download it in the settings — see below. |
| “The dictation model is on its way: 42%” | The model is downloading. Wait for it to finish, then try again. |
| “Dictation is off: Settings → Voice dictation” | Dictation is turned off: tick Enable dictation. This message only shows once. |
| “Dictation does not run on this processor” | Your processor is too old for the model — see Limits. |
The settings
Section titled “The settings”Dictation has its own entry in Settings → Voice dictation, with two cards.
The Shortcut and microphone card holds:
| Setting | What it does |
|---|---|
| Enable dictation | Turns the shortcut off or back on, without removing the model. |
| Shortcut to hold | Click it, then hold the new combination. |
| Microphone | “System default microphone”, or a specific microphone. |
| Test | Hold it: the microphone opens and bars show what it picks up. Nothing is typed. |
| Mute sound while dictating | On by default: your computer’s sound is muted while you hold the shortcut. Untick it to keep the sound on. |
| Microphone access | On a Mac and on Windows: the system’s answer, and the button to change it — see Microphone access. |
The Recognition model card states “It runs on your computer: your voice never leaves it.” It offers Download (486 MB), follows the download, then shows “Installed” with the model’s size and a Remove button.
With dictation off, the shortcut, microphone, Test and Mute sound while dictating are greyed out; Test is also greyed out during a dictation. Remove frees the space and the model is no longer downloaded again, until a new Download. Reset restores the original shortcut, Ctrl + Space.
The shortcut starts with Ctrl, or on a Mac with ⌃ or ⌘, which dictation tells apart. Shift and Alt (⌥ on a Mac) may be added, plus one key. On Windows, AltGr is never taken for Ctrl + Alt: typing @ or # on an AZERTY keyboard does not start dictation. Refused:
- the C, V and Z keys (copy, paste, undo in the terminal), even with Shift or Alt;
- the X and A keys alone (cut, select all);
- the B key with Ctrl, with or without Shift: Ctrl + B starts the agents’ shortcuts. ⌘ + B stays free;
- the command palette’s shortcut.
A refused shortcut is flagged in the card: pick another one. Letters show as on your keyboard (AZERTY: Ctrl + Q).
During the capture, Escape, Tab or a click elsewhere cancel it.
The language has no setting: dictation recognises the one you speak by itself, among 25 European languages including English and French. You can mix English words into a French sentence, such as “fais un push sur la branche dev”. It writes what you say, without translating it.
The microphone you picked stays remembered even when unplugged: the list then shows “Unplugged microphone”, and dictation finds it again when you plug it back in.
Microphone access
Section titled “Microphone access”On a Mac and on Windows, the system can refuse the microphone to VibeStarter. The Microphone access line shows its answer:
| System’s answer | What the line shows |
|---|---|
| Never asked (Mac) | Allow the microphone and “macOS asks for your consent only once.” The button opens the macOS dialog. |
| Allowed | “Allowed”. |
| Refused on a Mac | Open System Settings and “macOS refuses the microphone to VibeStarter. Turn VibeStarter on in Privacy & Security → Microphone.” |
| Refused on Windows | Open Windows Settings and “Windows refuses the microphone to apps. In Privacy & security → Microphone, turn on “Microphone access” and “Let desktop apps access your microphone”.” |
The button shown after a refusal opens the system’s Microphone page directly. The line updates when you come back to the app. macOS never asks again after a refusal: this button is the only way back.
The first launch offers dictation
Section titled “The first launch offers dictation”The setup wizard has a Dictation step, “Talk to your agents”. On an installation that is already set up, it shows alone at the first launch after the update. What it offers depends on your computer:
| Your computer | What the step offers |
|---|---|
| 16 GB of memory or more, and at least 8 logical cores | “Your computer is ready for dictation”: Install and continue installs it in the background, or Do not install dictation. |
| In between, including an 8 GB Apple Silicon Mac | “Dictation can run on your computer”: Install dictation or Later. |
| Less than 8 GB of memory, 8 GB on a PC, or fewer than 4 logical cores | “We do not recommend dictation on this computer”: Continue without dictation, or Install it anyway. |
| A processor that is too old (see Limits) | “Dictation cannot run on this computer”: Continue, with nothing installed. |
The model downloads once the app is open, while you work. Whatever your answer, you can install or remove it later from the settings.
The microphone. On a Mac that has not asked yet, the step warns: “macOS will then ask for microphone access.” Choosing to install dictation then brings up the macOS dialog. Choosing not to install it asks for nothing.
If the microphone is refused, on a Mac after that dialog or on Windows when its privacy settings already block it, the step shows “The microphone is refused to VibeStarter”. It offers a button that opens the system settings, and Continue, I will do it later. If you allow the microphone in the settings and come back to the app, the wizard moves on by itself.
On a Mac, if the step was passed without answering macOS, the first dictation records nothing: the pill says “Allow the microphone in the macOS dialog, then try again” and the macOS dialog appears. The next dictation works.
What dictation keeps, and what it sends
Section titled “What dictation keeps, and what it sends”- The microphone is only open while you hold the shortcut, or the Test button in the settings.
- Only the transcribed text is typed into the chosen field. If it is an agent terminal, it goes to that agent’s AI provider when you send it, like typed text.
- The recognition model is downloaded once from VibeStarter’s servers. It is kept in your storage folder, under
dictation— see Where your data is on disk.
What it costs your computer
Section titled “What it costs your computer”- Disk: 486 MB for the model.
- Memory: about 650 MB while you dictate. Recognition runs in a separate process, stopped after five minutes without dictation: the memory is then given back.
- Nothing in the agent environment: dictation runs beside it and does not slow it down.
The first dictation after a pause takes a little longer: the model reloads.
Limits
Section titled “Limits”- 25 European languages, recognised by themselves: 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.
- A recent enough processor. On a PC it must support the AVX2 instructions: Intel Core processors since 2013 and AMD ones since 2015 do, some Celeron and Pentium chips do not. Every Apple Silicon Mac does. Otherwise the wizard step and the settings say so, and nothing is downloaded.
- No button to dictate: dictation only goes through the held shortcut. Test listens to the microphone, it types nothing.
- Dictation only types into the VibeStarter app, not into other software.
- No voice commands: dictation types, it does not act in the app.
Frequently asked questions
Section titled “Frequently asked questions”Nothing happens when I press the shortcut. If dictation is off, the pill said so once: “Dictation is off: Settings → Voice dictation”. Tick the box Enable dictation, and check that the cursor is in a text field. Another application may also be catching the combination: pick another one.
My computer’s sound goes off when I dictate. That is on purpose: it comes back as soon as you release the shortcut, or press Escape. A sound you had already muted stays muted. To keep the sound on while dictating, untick Mute sound while dictating in Settings → Voice dictation.
The shortcut does nothing on a Mac, or with a Chinese or Japanese keyboard. The system then takes Ctrl + Space to switch keyboard language: on Windows with a Chinese or Japanese keyboard (IME), on a Mac when several input sources are on. Pick another shortcut in Shortcut to hold.
I refused the microphone by mistake. macOS never asks again. In Settings → Voice dictation, the button of the Microphone access line opens the system’s Microphone page: see Microphone access.
The wave does not move, or the pill says “The microphone sends no sound”. The microphone in use hears nothing. Pick the right one in Microphone, then hold Test while speaking: the bars should move. Otherwise, look at the Microphone access line: see Microphone access.
The text is wrong or odd. Speak close to the microphone, and prefer a headset to a laptop’s built-in microphone. A rare technical word may be written the way it sounds: fix it before sending.
The model download stopped. “The download was interrupted. Start it again: it resumes where it stopped.” Click Download again. If the message is about space, free at least 1 GB on the drive of the storage folder.