How can we help?

Click into any text field, press ⌃⌥Space, speak, then press ⌃⌥Space again. The text is typed where your cursor is. Or hold Fn (🌐) while you speak and let go to insert. To keep listening hands-free, double-tap Fn, or tap it once if “Press 🌐 key to” is set to “Do Nothing”, then tap it once more to finish. Press Escape to cancel. In Keyboard you can change the keys, hold Right Command, Right Option or Right Control instead of Fn, or dictate from a mouse button.

Most keyboards Apple did not make never send Fn to the Mac. In Keyboard, set the activation key to Right Command, Right Option or Right Control: hold it to talk and release to insert, or tap it once to lock hands-free and tap again to finish. A double tap from habit still locks. Shortcuts that use the key, such as ⌘C, still work and never start a dictation. You can also dictate from a mouse's middle or side button: hold it to talk, or press it once to lock.

Two, and both stay on this Mac. Microphone, so ZenVoice can hear you. Accessibility, so it can type the finished text into the app you're using. Without Accessibility, ZenVoice still works: the text is copied, and you paste it with ⌘V. The Permissions page shows both and opens the right place in System Settings.

ZenVoice recommends Parakeet Unified for English and Parakeet Ultra for the other European languages it supports. Voxtral Realtime writes 13 languages while you speak, and Whisper Large v3 Turbo covers 99. For a language no other model has, such as Tamil or Swahili, use Omnilingual ASR. Open Models to see each engine's speed and accuracy. ZenVoice times every dictation, so the speed shown is measured on this Mac. You can switch at any time, and your settings stay the same.

Press ⌃⌥V (Paste latest dictation) in any app. It's handy when the text went to the wrong window. Every dictation is also saved in History, if Keep a history of what I dictate is on, and you can copy any of them from there.

Yes. In Dictation → More ways to dictate, turn on Start ZenVoice from other apps (it starts off, since any app can open these links). It lists a link for each action: start and insert, start, stop, and cancel. Copy one into Raycast, Alfred, the Shortcuts app (Open URLs) or a Stream Deck button (Website). Pressing it does the same as the keys.

Yes. In General → Recording HUD, choose Notch (it hangs from the camera housing) or Floating panel (a glass bar you can put at any corner or edge). There you can also turn the live transcript on or off and set ZenVoice to launch at login. To keep the bar on screen even when you're not dictating, turn on Show ZenVoice at all times in Dictation.

Check that ZenVoice is running: its icon should be in the menu bar. Another app may use the same keys; pick a different shortcut in Dictation. If Fn opens the emoji picker or Apple's dictation instead, go to System Settings → Keyboard and set “Press 🌐 key to” to Do Nothing. Then check Permissions: both should show as allowed.

ZenVoice is probably listening to the wrong microphone. In Dictation → Input device, pick the one you're speaking into; the bars should move when you talk. Make sure Microphone is allowed on the Permissions page. Bluetooth headsets can take a second to switch to their microphone, so pause briefly before you speak.

Usually one of three reasons. Accessibility isn't allowed (see Permissions). The field is a password field, or an app has secure input on: macOS blocks all typing then, so ZenVoice copies instead; click another app and back, or reopen the app it names. Or Copy only is on in Insertions. Either way the text is on your clipboard: press ⌘V. If one app or website never accepts a paste, open Insertions → Per app, add it and choose Type: ZenVoice then enters the text key by key.

Nothing is lost if Keep recordings that didn't work is on (Settings). The recording goes to History → Recovery Inbox, where you can try it again or delete it. A dictation you cancel with Escape goes there too, so a slip of the finger is never final. Very short clips are dropped as accidental presses.

ZenVoice skips recordings with no speech, so noise alone shouldn't type anything. If words still appear, check Dictation → Input device: a noisy or far-away mic can sound like speech. For a name it keeps mishearing, add it to your Dictionary in Personalisation. To undo a word while you're talking, say “scratch that”.

The first dictation after launch loads the model, so it takes a moment longer. After that, most of the time goes to the engine and the formatting level. Try a faster engine in Models (Parakeet Ultra for English), or use Clean instead of Polished in Personalisation. Clean formats with rules only and adds almost no time.

ZenVoice pastes through the clipboard. With Restore clipboard on in Insertions, whatever you had copied is put back straight after the paste. Turn Copy only on if you'd rather paste every dictation yourself. To keep the clipboard out of it for one app, choose Type for that app under Insertions → Per app.

Yes. In Settings → Transcribe, install the command-line tool, then run zenvoice transcribe meeting.m4a. It uses your local models and licence with the app closed. Add --json for structured results, --stream for completed words as they arrive, --dictionary to use saved Dictionary corrections, or --output <folder> to save files. zenvoice models lists local engines, zenvoice doctor checks setup, and zenvoice --help lists every option.

Use zenvoice transcribe lecture.mov --resume <folder>, then repeat the command with the same recording and settings. Fully completed windows are reused. Checkpoints contain unencrypted transcript text in owner-only files; delete the folder when finished. A resumed run emits the full transcript again, so replace your previous output rather than appending to it.

Verbatim types exactly what the engine heard. Clean removes fillers (um, uh), repeated words and false starts, and writes numbers, prices, times and dates the way you'd type them. Polished also fixes punctuation, capitals and flow, so it reads like you wrote it, using an on-device model. Choose the level in Personalisation. It never changes what you meant.

At Clean and Polished, “three thousand dollars” becomes $3,000, “twelve dollars and ninety nine cents” becomes $12.99, “twenty five percent” becomes 25%, and “three thirty pm on March fifth” becomes 3:30 pm on March 5. Short numbers follow context: “one twenty chairs” is 120, “meet at one twenty” is 1:20, “a dollar twenty” is $1.20, and “flight UA one twenty” is UA 120.

Say “scratch that” and carry on: “send it by Friday, scratch that, Thursday” types “send it by Thursday.” “New line” and “new paragraph” break the text. Start with “verbatim” to have the rest typed exactly as heard. Personalisation → Spoken commands lists every command you can say.

Yes. Select the text, press your dictation shortcut and say what to do: “make it shorter”, “fix the grammar”, “make it more formal”, “translate to Spanish”. The selection is rewritten in place. If what you say isn't an instruction, it's typed over the selection as usual. Voice edits use Apple Intelligence and work in English for now.

On purpose. In terminals (Terminal, iTerm, Ghostty, Warp…) and code editors, ZenVoice types exactly what the engine heard, with no formatting, so commands aren't changed without you seeing it. In a coding agent's prompt box, such as Claude Code, Cursor's agent, T3 Code or the Claude app, you get a clean paragraph with code words, file names and flags kept exact.

Yes. In Personalisation → Tone, choose how Personal messages, Work messages, Email and Everything else should read, from Very casual to Professional. ZenVoice picks the tone from the app you're dictating into. Tone applies at the Clean and Polished levels.

Turn on Review before insert in Insertions. When you finish speaking, the text opens in a small card where you can edit it. Press Return or your shortcut to insert it, or Escape to discard it.

Text you insert by saying a trigger phrase, like “my address” or “sign off”. Add them on the Snippets page, and use the Try it box to see which snippet a phrase would insert.

Yes. Open Transcribe in Settings and drop in an MP3, M4A, WAV, AIFF, FLAC, MP4 or MOV file. Text appears as it goes, so you can keep working, and dictation always goes first. You can also drop a file on the menu bar icon, choose Transcribe File… in its menu, right-click a file in Finder and choose Transcribe with ZenVoice, or use Open With. If the Finder option is missing, turn it on in System Settings → Keyboard → Keyboard Shortcuts → Services. The text is saved as a .txt file in the folder you choose, and you can export Markdown, SRT or VTT. Recordings have their own model, language and formatting level. To use it in Terminal, install the command in Transcribe, then run zenvoice transcribe meeting.m4a. Everything happens on this Mac.

In Keyboard, add the languages you speak. Then ⌃⌥L (or the shortcut you choose) moves to the next one from anywhere, and the HUD shows which you are on. Pressed while you dictate, the switch applies to your next dictation.

Yes. In Personalisation → Dictionary, add names, products and jargon: ZenVoice listens for them and fixes near misses. In Text replacement, add what ZenVoice hears and what it should write instead. Under Suggestions, ZenVoice offers words you keep correcting, but nothing is added until you say yes. You can export and import your Dictionary, and it stays on this Mac.

Choose your spoken language in Models. Whisper models cover 99 languages, Parakeet Ultra and V3 cover 25 European languages, Omnilingual ASR more than 1,600, and each engine lists its own. Some engines can translate what you say into English as they type. First runs are set up for English; change the language whenever you like. Hindi in Devanagari isn't available yet; for Hindi speech, use Hinglish.

Pick Hinglish as the spoken language in Models. ZenVoice uses Hinglish Prime for it: download it once from Models if it isn't there yet. It types Hindi and mixed Hindi-English speech in the Latin alphabet, English words included, never Devanagari.

Turn on Last dictation under the notch in General. Rest the pointer on the notch while ZenVoice is idle and it drops down with what you last said, the app it went into and when, with Copy and Paste again. It is off by default because it shows your words on screen, and it is hidden from screen sharing and recordings.

No. Recording, transcription and formatting all happen on this Mac, and your recordings and text are stored here. ZenVoice goes online for model downloads, update checks, and license activation, deactivation and periodic verification. A licensed installation checks approximately weekly, sending its license key, this Mac's device public key and a signed challenge response—not your recordings or transcripts. Once a model is downloaded and access is activated, dictation works offline.

Settings → What ZenVoice remembers shows everything kept right now: your dictations, recordings waiting to be tried again, text replacements and downloaded models. Turn off Keep a history of what I dictate to stop saving text. Turn off Keep recordings that didn't work to delete failed audio straight away; kept ones are removed by themselves after a day. Audio from successful dictations is deleted afterwards, unless you turn on Keep new recordings in History → Audio.

In Settings, delete each item on its own, or use Erase everything. Hold the button to confirm. It deletes all your dictations, recordings, stats, saved words and the Transcribe list, and starts fresh. Transcript .txt files in your folder stay. Downloaded models stay; remove them in Models (Manage, on the same list).

Word counts, times, and which apps you dictated in, used for Insights and your streak. With Keep my usage stats longer on (Settings), they stay after your dictations are deleted. Your words and recordings are never part of them.

ZenVoice checks for new builds in the background when Check for updates automatically is on, or you can check any time in About. Every build is verified before anything on this Mac changes.