microVoice is an AI dictation app that lives in your Mac's menu bar. Press a hotkey and talk — speech is transcribed on-device, then AI removes fillers, fixes misrecognitions and polishes the text before inserting it at your cursor. Your audio never leaves the Mac by default.
Full 2-day trial · macOS 15+ · Apple silicon recommended
Meeting notes
⌘Hold right ⌘ and just speak
9:00esc
Demo: real cleanup examples, looping
01PRIVATE BY DESIGN
Audio stays on this Mac. By default.
Transcription runs on-device. AI cleanup sends text only — never stored on our servers. Want even better recognition? Turn on High-accuracy recognition (cloud). Whether anything leaves your Mac is your call.
Voice
→
Recognition on this Mac
Recognition
Text
Insert at cursor
✓ Text only× No audio→
AI cleanup (never stored)
Audio stays on-device by defaultCleanup is text-only, never storedHistory stays local too
Typical cloud dictation
Streams your audio to a server for recognition and cleanup — whether to send is not your choice.
microVoice
On-device recognition by default. Only text goes to the cloud (never stored). High-accuracy recognition — cloud audio processing — is strictly opt-in.
“Don’t send” as the default, “send for accuracy” as an option. Both are in your hands.
02SMART CLEANUP
Delete. Fix. Polish.
AI cleanup does three jobs in one pass: deletes fillers and false starts, fixes misrecognitions from context (their → they’re), and polishes speech into writing. A two-stage AI pipeline separates cleanup from review — meaning, numbers and names never change.
With a fidelity guard: if the AI’s output drops a number or proper noun, it is discarded and the local cleanup result is used instead. Offline, rule-based cleanup keeps working — you never lose your words.
As recognized → AI cleanup (two-stage, reviewed)
Um, well, about next week’s like, meeting —
Fillers deletedErrors fixedPolished in contextMeaning preserved
2 AI
Two-stage AI: cleanup + review
~ 2 s
To polished text
0
Audio or text stored on servers
えーとあのーなんかまあこう、ですねあっ、違うえっと言い直しどもり
03ANY LANGUAGE
It knows spoken language. Any language.
Fillers, false starts, stutters, homophone errors — spoken-language quirks exist in every language. microVoice’s AI cleanup is natively multilingual: it polishes your speech in the language you spoke, into that language’s written form.
04INSTANT
Listening the instant you press.
Recording starts the instant you press the hotkey, so the first word is never lost. Live preview while you speak. Release, and AI-polished text lands at your cursor in seconds.
⌘Pressrecording at 0 ms
Speaklive preview
ReleaseAI cleans up
Insertedin one paste
05YOUR CALL
You decide how much to polish.
From “keep every word” to “full rewrite” — polish strength is a three-step slider. The trade-off between fidelity and readability stays in your hands, not the app’s.
Input (as recognised)
Um, is the deck for the, ah no, for Thursday’s meeting ready
What gets inserteddeletions only
Um, is the deck for the, ah no, for Thursday’s meeting ready
Removes fillers and false starts, fixes punctuation. Your wording stays untouched.fidelity ●●●●●readability ●●●○○
What gets inserted+ error fixes
Um, is the deck for the, ah no, for Thursday’s meeting ready?
Also fixes clear misrecognitions (homophones) from context. Sentence shape stays as spoken.fidelity ●●●●○readability ●●●●○
What gets inserted+ written style
Um, Is the deck for the, ah no, for Thursday’s meeting ready?
Default. Folds afterthoughts into the sentence and shapes speech into natural writing.fidelity ●●●○○readability ●●●●●
What gets inserted+ restructuring
Um, Is the deck for Thursday’s meeting ready?
Collapses hedges and thinking-aloud, restructures into tight prose. Meaning and tone preserved.fidelity ●●○○○readability ●●●●●
Rewrite mode changes wording (by the strength you choose). At every strength, meaning, numbers, names and tone are preserved — the AI never invents content.
06THE DETAILS
Small things that add up.
One hotkey, anywhereRight ⌘ by default (or fn / right ⌥). Records while held, inserts the moment you release, esc to cancel.
HUD voice barWaveform, live preview and time left — without ever stealing focus from where you’re typing.
DictionaryFeed technical terms and proper nouns to the AI as vocabulary hints. Also normalizes casing: github → GitHub.
Replacement rulesDeterministic fixes for often-misheard phrases: “Piar Times” → PRTimes.
Context-aware cleanupReads the app, window and your recent dictations into the same screen to fix misrecognitions and match the context (opt-in).
Nothing gets lostThe clipboard restores itself after pasting, and your text survives even if insertion fails.
Local-only historyRaw and cleaned text stay on this Mac. Keep for 1 / 7 / 30 days or forever.
Personalization (learning)Gradually learns your vocabulary, recurring misrecognitions and style, so cleanup keeps getting smarter (opt-in, resettable anytime).
AI cleanupCloud AI (text only, never stored). Rule-based cleanup offline
Session lengthUp to 9 minutes per session
LanguagesRecognition and cleanup are multilingual (Japanese, English and more). UI in 14 languages
Does it work without a connection?
Yes. Offline it runs on-device recognition plus rule-based cleanup, so you never lose your words. Online, cloud AI polishes the text (text only, never stored), and with High-accuracy recognition (opt-in) audio is also recognized in the cloud.
What about English dictation?
Yes. Both recognition and AI cleanup work in many languages, including English and Japanese — cleanup polishes your speech in the language you spoke. The UI is available in 14 languages.
Do I need an account?
AI cleanup and high-accuracy recognition require a free sign-in. Without one — or offline — the app keeps working with on-device recognition plus rule-based cleanup.
Is anything I say kept anywhere?
History (raw and cleaned text) is stored only on this Mac. Text sent for cloud cleanup is not stored on our servers (except the optional learning profile, deletable anytime). Retention: 1 / 7 / 30 days or forever.
What is High-accuracy recognition (cloud)?
An opt-in mode that recognizes your speech with a state-of-the-art cloud model — markedly better at names, technical terms, and self-corrections. Up to 300 dictations per day (then it falls back to on-device automatically). Leave it off and audio never leaves your Mac.
Simple pricing
One license for your Macs. Try everything free for 2 days.
Your license key appears on your microAI account (Account → Licenses) after purchase and follows your account. Checkout is handled by a Merchant of Record, taxes included. The number of Macs a license can activate is limited. High-accuracy recognition (cloud) covers up to 300 dictations per day; beyond that it automatically falls back to on-device recognition, so dictation never stops.
