Guide

Learn Parroting in five minutes

From the first time you speak to the point where it really gets you. Here is every basic you will need.

Before you start

Install: Open the downloaded Parroting-macos-arm64.dmg, drag Parroting onto the Applications folder, then open it from Applications. Opening it from Applications is what lets it update itself later.
First launch: Parroting walks you through allowing Microphone and Accessibility in System Settings. The microphone records your voice; Accessibility lets Parroting paste text into the app you are using.
First launch after an update: The first time you open Parroting after each update, the speech model takes about 20 seconds to get ready. After that there is no wait.
Privacy: Parroting does not read the contents of password fields or apps such as online banking.
Something wrong?: In Settings → Advanced → About → Show all, click “Report a problem” and describe what happened. You can attach the log, which has what you said and personal data removed, and see exactly what will be sent first.

Speak

Hold: release to paste: Right ⌥＋Right ⌘Hold and speak. When you let go, the text is pasted right where your cursor is. Best for a sentence or two.
Tap: the overlay stays on screen and keeps listening: Continuous dictation is on by default (under Talk → More settings → Continuous dictation, the switch is “Paste each phrase when you pause”). A quick tap of the shortcut (released within 0.3 seconds) keeps the overlay on screen, listening, and what you say shows in the overlay as you say it. Pause for a moment (0.4 seconds by default) and that part is tidied and pasted, and it keeps listening; to paste right away, click “Done” on the overlay. You can switch apps along the way; text goes to whatever field you are typing in. Tap again to stop. Best for long text or thinking out loud.
Paste once when you finish: In Talk → More settings → Continuous dictation, on by default: the text shows in the overlay first and is tidied and pasted in one go after a short pause; with AI cleanup on, filler words and restarts are removed too. Turn it off and each phrase goes straight into the text field; how long a pause ends a phrase is set by “Pause length before pasting” (0.7 seconds by default). While it is on, how long to wait before pasting is set by “Paste after a pause of” (0.3–3 seconds).
Show captions while speaking: On by default (also a switch in Quick settings on the Home page): as you speak, the recognized words appear one by one above the overlay, both when holding the hotkey and in continuous dictation; what is not settled yet shows dimmed. Turn it off to see nothing while you speak.
Hold or tap?: Holding longer than 0.3 seconds is a hold; letting go quickly is a tap. You can mix both without switching modes.
Cancel: EscCancel at any time; nothing is pasted. In continuous dictation, text that was not pasted yet is copied to the clipboard, and segments that were already cut can be found again under “Recover recordings”.
New line after each paste: Turn this on in Quick settings on the Home page in Settings and each segment ends with a line break. It never presses send for you.
External keyboards: If your keyboard has no right-hand ⌥ or ⌘, pick another combination in Settings → Talk → Shortcuts.

AI cleanup

Drops filler words and false starts: “AI cleanup” is on by default (you can turn it off on the Home page in Settings, card “AI & rewrite panel”) and works once you add an AI key; the recording overlay then gets a colored border to show AI is tidying up. Then saying “Um, let’s meet at three, no wait, four” pastes “Let’s meet at 4.” Numbers, names and negatives are never changed.
Without an AI key: Parroting falls back to on-device rules that remove standalone fillers like “um” and “uh” and repeated words, so “I I’ll go tomorrow” becomes “I’ll go tomorrow.”
Short sentences skip the AI: “Short sentences skip the AI” under AI → More settings → Speed is on by default (up to 15 characters; 25 or 40 also available; an English word counts as two characters): a very short sentence is pasted as processed on your Mac without waiting for the AI, about 0.5–1 seconds sooner. The trade-off is that those sentences miss the AI’s small fixes, such as doubled words or self-corrections.

Rewrite

The “✦ Rewrite” button: After each paste, “✦ Rewrite” appears next to the overlay for about 4 seconds. Click it to open the rewrite panel. It also appears in AI chats, so you can translate or organize before you send. It does not appear when the text is shorter than 4 characters, while the rewrite panel is already open, or when there is no AI service. Press it before you have dictated anything and it opens an empty panel, ready for your next sentence.
Keep the rewrite panel open: Turn on “Keep the rewrite panel open” on the Home page in Settings (card “AI & rewrite panel”). Per-sentence formats process at each pause; whole-piece formats wait for Done. Replace changes the text but keeps the panel open. × or Esc turns this keep-open switch off, but the chosen format still works if Auto replace is on.
Without an AI key: Rewriting and AI cleanup need an API key on the AI page first. Until then those switches are greyed out; click one to see a hint, and “Set up” takes you to the key field. Without an AI service there is no rewrite panel, and tone hints still show on the overlay; if the AI service stops working, the panel closes and “Keep the rewrite panel open” turns off.
Pick a format: Built-in formats: 1 Tone (the default), 2 Organize, 3 Translate, 4 Meeting notes, 5 Email. A highlighted format applies to every sentence. Press it again for “Showing only”: the panel shows what you said without rewriting; Auto replace, Replace and Again are greyed out. The choice survives restarts.
Auto replace: Per sentence (Tone, Organize, Translate) rewrites and pastes at each pause; Whole (Meeting notes, Email and custom formats by default) collects captions while you speak and pastes nothing until you press Done, then pastes one complete piece. Only the rewrite is pasted. With the panel closed it still works: the recording pill shows a green “✦ Organize” tag beside “Continuous”; click it to change format or turn Auto replace off.
Adjust the tone: Tone uses the Warmth and Formality sliders.
What you said and the result: The top of the panel shows what you originally said, and you can hide it. The green box below is the rewritten text, which you can edit directly.
Resize and text size: Drag the panel’s edges or bottom-right corner to make it bigger or taller; its size and position are remembered. Press ⌘＋ / ⌘－ to change the text size.
Replace: Click “Replace” to swap out the text you just pasted. Not happy with it? Press ⌘Z to undo, or click “Again” for another version.
Pick a format with number keys: With the panel open, press 1–9 to pick the matching format; press the same number again for Showing only. While editing the result, use ⌘1–⌘9.
During continuous dictation: While the panel is open and you haven’t touched it, the original switches to the newest segment you said. Once you rewrite or edit, it stays put; if more text is pasted after that, “Replace” copies the result for you to paste yourself, so nothing gets replaced in the wrong place.
My formats: In AI → Rewrite formats → Add format, name it and describe how to write. Set When to process to Per sentence or Whole; your formats default to Whole. Built-in formats can be edited too.
Meeting notes and Email: English stays English in every format. Meeting notes use Topic, Key points and To-dos; email reads like “Subject: … / Hi Sarah, / … / Best regards,”. Details appear only if you said them.
Organize: For thinking out loud. It turns a rambling monologue into orderly text: the point first, one topic kept together, repeats and abandoned versions dropped, paragraphs or a list where it helps, and every fact you mentioned kept.
Translate: The direction is picked for you: Chinese, Japanese or Korean becomes English, and English becomes Chinese.
Keep the panel open: every sentence translated or organized first: Want every sentence translated or organized before it lands? See Keep the rewrite panel open for the steps and examples.

Tone reminders

English and Chinese: The on-device model checks each sentence for its language. English and Chinese can be mixed freely. It can flag harsh, passive-aggressive or overly casual wording before you send. No API key or upload is needed. Personal tone training works in English too.
The reminder card: If something you are about to send could hurt, a reminder card pops up and the rewrite panel opens with it. You can rewrite, or send it as it is.
Wrong call: If you think the reminder was not needed, click “Wrong call”. This feedback adds up and is used to train judgment that is yours.
No hints in this app: One click on the reminder card adds the current app to the no-reminder list. You can remove it later under Tone & context → More settings → Where to show hints.
Where to show hints: In Tone & context → More settings → Where to show hints, add the apps where you never want reminders. You can also switch reminders off by scene; chats with an AI are quiet by default.
A gentle nudge when you are too casual: When the situation calls for formality but you sound too casual, a quiet hint appears in the corner without interrupting you.
Hints in continuous dictation too: Tone hints work in continuous dictation, not only when you hold the hotkey. Chats with an AI are a muted scene by default, so there are no hints there. Chinese typesetting and script conversion settings appear only when the interface or speech language is Chinese.

Dictionary

Added only when you confirm: Words you correct are not added to the dictionary automatically. They first go into Review on the Personalization page, for example “Before: Catherine / After: Kathryn”, and are only added when you click confirm. Parroting tells the two apart by sound: words that sound alike count as misheard and can go into the dictionary; words that sound quite different count as reworded, get their own tab and never go into the dictionary or training. You can correct its call. Confirmed items fold away under “Confirmed (N)”.
Add your own: Go to Words & history → Dictionary → Add and enter the correct spelling and the ways it is commonly misheard.
Correct a few more times at first: Speech recognition uses the model you chose during setup, which occasionally mishears sound-alike words or English terms. Correct it a few more times at the start and your common words will soon come out right. You can also switch to a larger model in Settings.
Manage your dictionary: Search, sort and delete in bulk, and import or export CSV.

Speech models

Choosing a model: During setup, pick a speech recognition model to download. None is marked “Recommended” for English users. Recognition runs on your Mac and your voice is never uploaded. For what each model is good at, accuracy and licenses, see Speech recognition models.
Other models: On the Talk page, open “Speech model & language” and click “Manage models”: “Available to Download” lists other models such as Whisper and Qwen3-ASR. Larger models are usually more accurate and slower.
Import your own model: On the Talk page, open “Speech model & language” and click “Manage models”, then “Import model…” to the right of the “Downloaded Models” heading and pick a .gguf file. Parroting checks the format, copies the file in, then has your Mac’s built-in voice read a sentence as a test. Only if it recognizes the words is the model turned on and selected. You can move or delete the original file afterwards. To train a model with your own recordings, see Learns from you.
Test with your read-aloud recordings: At the top of “Manage models”, “Test with your read-aloud recordings” runs the current speech model on your read-aloud test set (one in five readings) on this Mac and shows the error rate. Test before and after importing a model you trained to see if it improved.
If the import fails: You’ll see why, for example “This isn’t a speech model Parroting can use.” The model you were using stays as it is.
Switch back or delete: Imported models are marked “Imported”, with any features they don’t support (such as live streaming) listed. While one is in use, “Back to the standard model” appears next to it. You can delete an imported model at any time.

Read aloud

A checkup of how well it hears you: The top card on the Personalization page in the sidebar. Read a few sentences from a script and see each misheard word marked, with your error rate at the end of the round. Click “Start a round (10 sentences, about 2 min)”.
How to read: Click “Start reading”, or hold your dictation shortcut. A 1.5 s pause ends the reading, or click “Done”.
Whose mistake was it?: When the marked words sound like the script, Parroting picked the wrong words and you read it right. When they sound quite different, you may have said other words: use “I said different words” to type what you said. If the script itself was wrong and Parroting heard it right, click “What Parroting heard is right”.
Where the sentences come from: Things you have said, your dictionary and a built-in bank. Sentences you have read do not come back.
Where the recordings stay: Read-aloud recordings stay on this Mac and are not cleared by the recording retention setting, until you use “Delete all read-aloud recordings”.
Use them for training: “Export training data…” on the card packs the recordings with their correct text, for training your own speech model on a computer with an NVIDIA GPU. Parroting itself never trains a speech model. Steps: Learns from you.

Personalize tone hints

Where to find it: Under “Personalize tone hints” on the Personalization page in the sidebar. It only tunes tone hints, never speech recognition, with one section for each language you give feedback in. For your own speech model, see Learns from you.
Start training: Once you have confirmed 20 pieces of tone feedback, you can click “Start training”. Training runs on your own Mac, and your data is not uploaded.
Check, then adopt: When training finishes you first see the results: which sentences are now judged differently. If you are happy with them, click “Adopt”. You can go back to the previous version at any time.

Want speech recognition to learn your voice too? See Learns from you →

AI service

Where to set it up: On the AI page, choose an AI service and enter your API key. The key is stored in the macOS keychain.
Without a key: AI cleanup falls back to on-device rules, and rewriting is unavailable. Tone reminders, the dictionary and on-device speech recognition all keep working.
Step-by-step guide: First time? See Get a free Gemini API key, with a picture for each of the four steps.
Getting a Gemini key: 1. Open the page: In your browser, open aistudio.google.com/api-keys and sign in with your Google account. The first time, you need to accept the terms of service. Going straight to this address is quickest; the entry point is easy to miss from the AI Studio home page.
2. Create a key: Click “Create API key” in the top-right corner, leave the name and project as they are, then click “Create key”. A string starting with AIza appears; click the copy icon next to it.
3. Paste it into Parroting: On the AI page, choose Gemini in the AI service card and paste the key you just copied into the API key field. If you forget it later, no problem: you can copy it again from the same page at any time.
4. Limits and cost (optional): The free tier has per-minute and per-day request limits, which are usually enough for everyday dictation and rewriting. If you use it heavily, add a credit card under Billing in the AI Studio sidebar to raise the limits. Google sets a monthly spending cap on newly linked accounts, so a large bill will not appear out of nowhere.

Overlay

Move it: Drag the overlay anywhere you like, and Parroting remembers where you put it. To put it back, click “Back to default position” in Tone & context → More settings → Recording overlay.

Ready to start?

Download for Mac ↓

v0.4.13 · 50 MB