QUILL Transcribe User Guide

QUILL Transcribe types what you say into the program you are using. Write an email in Outlook, fill in a form in Edge, Chrome or Firefox, add a note in Notepad or a paragraph in Word, all by speaking, without leaving the program.

It uses the same speech models and the same dictation as QUILL and QUILL Lite. If you already dictate in QUILL, the words, punctuation and commands you know work here too.

Version 1.0.0 (build 3), released on 1 November 2026: the first public release.

What it promises

• Your words go where you started. Dictation writes into the field that had the focus when you started it, and only that field. If you move somewhere else, dictation pauses and keeps your words until you say where they go.
• It never reads the other program. QUILL Transcribe types into the field. It does not read what is already there, what the page says, or what is in your email.
• Your speech stays on your computer. The built-in models run on this computer. Your speech only goes over the internet if you choose an internet model yourself, and you are asked first.
• The model you choose is the model it uses. It never switches models on its own.
• Nothing listens until you start. The microphone opens when you start dictation and closes when you stop.

Getting started

The first time

The first time QUILL Transcribe opens, it shows Set Up QUILL Transcribe. It has one page:

1. About QUILL Transcribe - a short description to read with the arrow keys.
2. Microphone - the Windows default, or a microphone you name. Press Test Microphone to record four seconds and hear how loud it was and what it heard. The test stays on your computer and nothing is kept.
3. Speech model (Alt+D) - the models already on this computer, each with its tier (see “Choosing a model”). Moonshine is chosen to begin with, and is a good place to start: it is built in, quick, works with nothing downloaded, and adds punctuation for you. Below it, Recommended for this computer (Alt+P) is one line naming the model that suits your computer and why. It is only advice: the model in the box above is the one you get.
4. Start and stop key - the key that starts and stops dictation from any program. It is Alt+Shift+Backslash (Alt+Shift+\) unless you change it. Type another, for example Ctrl+Alt+PageDown, or press Record Key (Alt+C) and then press the combination you want: it is written into the box for you. Leave the box empty for no key. A key another QuillVille app already uses is refused with the name of the app that uses it. If a key does something else when you press it in Record Key, another program already uses it: type it, or choose another.
5. Command key - optional, and empty unless you choose one. Press it in the field you are dictating into and the next thing you say is a command, as if you had said QUILL first (see “The command word”). Type a key, or press Record Command Key (Alt+N) and press it. It must not be the start and stop key.
6. Start QUILL Transcribe when I sign in to Windows - off unless you check it. When it is on, QUILL Transcribe opens in the tray. It still does not listen until you start dictation.

Press Finish to keep your choices or Skip to change nothing. You can open setup again at any time from the Help menu (Ctrl+Alt+S).

The tray

QUILL Transcribe lives in the notification area, the system tray. To reach it from the keyboard, press Windows+B, find QUILL Transcribe with the arrow keys, and press the Applications key or Shift+F10 for its menu.

The menu starts with three lines you can read but not press: what dictation is doing, which model it uses, and which dictation profile is in use. Then:

• Start Dictation or Stop Dictation
• Resume, Resume in This Field and Discard Held Words, when dictation is paused
• Choose Model, Choose Microphone, Dictation Settings, Dictation Profile and Model Library
• Transcribe a Recording
• User Guide and Check for Updates
• Show QUILL Transcribe and Exit QUILL Transcribe

The tray icon’s name also says what dictation is doing, for example “QUILL Transcribe - Listening”. With a profile other than Standard it says that too: “QUILL Transcribe - Listening - Email profile”.

If Windows has no notification area for QUILL Transcribe, its window stays open instead, and closing it puts it on the taskbar. QUILL Transcribe says so when it starts.

The window

Press Enter on Show QUILL Transcribe in the tray menu, or your show and hide key if you chose one. The window shows:

• State - Ready, Listening, Hearing you, Paused, or Waiting for you to switch to a field.
• Start Dictation - the button changes to Stop Dictation while dictation is on.
• Resume (Alt+R), Resume in This Field (Alt+T) and Discard Held Words (Alt+C) - for when dictation is paused. Discard Held Words was Alt+W until 2026-10-17, but Alt+W is the Window menu, and when a button and a menu share a letter Windows presses the button: Alt+W threw your held words away instead of opening the Window menu. Now Alt+W always opens Window, and Alt+C is the button. For example, if dictation paused because you left the field and you no longer want those words, press Alt+C; Dictation > Discard Held Words (Ctrl+Shift+D) does the same from the menu.
• Model and microphone - the model, microphone, dictation profile (and whether it is locked) and keys in use.
• What happens next - what starting dictation will do, or why it paused or could not start.
• Choose Model, Choose Microphone and Dictation Settings.

Closing the window keeps QUILL Transcribe running in the tray. To quit, choose Exit in the File menu (Ctrl+Q) or Exit QUILL Transcribe in the tray menu.

Practice and lessons

New to dictation, or to a new microphone or model? Practise in QUILL Transcribe’s own window first. Nothing you say there goes into another program.

Practice Dictation

Open Practice Dictation from the Dictation menu (Ctrl+Shift+P). The window has, in Tab order:

• Sentence to read aloud - a sentence for the recognition test.
• Practice field - QUILL Transcribe’s own field. The focus starts here.
• Result - the last recognition test’s score.
• Start Dictation Here (Alt+H) - starts dictation in the practice field, or stops it. Ctrl+F11 does the same while you are in this window.
• Read This Sentence (Alt+T) - the recognition test, below.
• Score the Field (Alt+S) - scores everything in the practice field against the sentence.
• Next Sentence (Alt+N) - shows another of eight sentences.
• Close (Escape) - stops dictation in the field and closes the window.

Everywhere else, starting dictation from QUILL Transcribe’s own window waits for you to switch to another program, because the window itself is never dictated into. The practice field is the one exception: Start Dictation Here writes into it straight away. The same checks as any start still apply (Safe Mode, the microphone, your model), and if dictation is already on in another program you are asked to stop it first.

Everything works in the practice field as it does elsewhere: punctuation by voice, “QUILL scratch that”, “QUILL caps on”, snippets, Review Dictation, and your profile. Formatting profiles (Document, Email) treat the field as plain text, as they would Notepad. When dictation stops, by any route, it goes back to working in other programs as usual.

Example. Press Ctrl+Shift+P, then Alt+H. You hear “Dictation on”. Say “this is my first dictation period” and pause. You hear your words read back, and the practice field holds “This is my first dictation.”

The recognition test

Press Read This Sentence (Alt+T). The practice field is cleared, dictation starts in it if it was off, and the status bar shows the sentence. Read the sentence aloud at your usual pace, then pause. The first thing written is compared with the sentence word for word, and you hear the result once; it also stays in the Result field. For example:

• “All 9 words right. Recognition is working well.”
• “78 percent right: 2 differences in 9 words. Not heard as said: jumps, lazy. Try speaking a little slower, moving closer to the microphone, or a larger model in Choose Model.”

Capitals and punctuation do not count; the test is about the words. Below 80 percent you are given the three suggestions above. If you said the sentence in two breaths, only the first part is scored; press Score the Field (Alt+S) to score everything in the field instead. The score uses the same word-error count as QUILL’s dictation benchmark, so it means the same thing.

Limits: the sentences are in English, and the test checks your microphone, model and voice together. It cannot tell which of the three to change.

Tutorials

Tutorials in the Help menu (Ctrl+Alt+F1) opens five short lessons in the same Tutorials window as QUILL and QUILL Lite:

1. Your first dictation, in the Practice window.
2. Pause and resume in another program.
3. Commands with the command word.
4. Snippets and clips by voice.
5. Review before it is typed.

Each step says what to do, why, and what you should hear. Where a step names a command, Try it runs it for you. The same lessons are in the tutorial book, tutorials.html beside this guide.

Dictating

With your key

1. Go to the field you want to write in: the body of an email, a box on a web page, a document.
2. Press your start and stop key. You hear the dictation tone and “Dictation on”.
3. Speak. When you pause, what you said is typed where the cursor is.
4. Press the key again, or say “stop dictation”, to stop. Anything you were still saying is written first.

Holding the key to talk

If you would rather hold a key while you speak, as with a walkie-talkie, open Dictation Settings, choose More Dictation Settings, and check Hold the dictation key to talk (Alt+H). It is off until you turn it on, and it is the same setting QUILL and QUILL Lite have for Ctrl+F11. Then:

• Hold your start and stop key and talk. When you let go, dictation stops, keeping your last phrase.
• What you say while the key is down is typed the moment you let go, not before. Your start and stop key has Alt or Ctrl in it, and letters typed while one of those is held would reach the other program as shortcuts.
• A quick press still turns dictation on, and the next press turns it off.

Keys with Ctrl in them make most screen readers stop talking while they are held, so a key with Alt, such as the default Alt+Shift+Backslash, works best for holding.

Starting with your voice: the wake phrase

You can start dictation by saying the wake phrase, “Quill dictate” unless you change it, instead of pressing a key. Open Dictation Settings and check Listen for the wake phrase while dictation is off (Alt+H in that window); the Wake phrase box (Alt+W) is where you change the words. It is off until you turn it on, and when you turn it on you hear “Listening for your wake phrase whenever another program is in front.”

While it is on:

• The microphone stays open whenever another program is in front, and the State field and the tray say Listening for your wake phrase. Nothing it hears is written, kept or sent anywhere until it hears the wake phrase at the start of what you say.
• Say “Quill dictate” in the field you want, and dictation starts there. Anything you say after the wake phrase in the same breath is typed: “Quill dictate dear Sam comma” writes “Dear Sam,”.
• When you stop, it goes back to listening for the wake phrase, and says so.
• It does not listen while QUILL Transcribe’s own window, QUILL or QUILL Lite is in front (they have their own dictation), or while a program on your never-type list is in front, such as a password manager.

From the tray or the window

When you start from the tray or the window, QUILL Transcribe has the focus, and it never types into itself. So it waits: QUILL Transcribe says “Switch to the field you want”, and dictation starts as soon as you are back in another program. Alt+Tab, or Escape out of the tray menu, takes you there. If no other program comes forward within 20 seconds, it gives up and says so. Choosing the same command again while it waits cancels it.

What you hear

You hear a tone when dictation starts and stops, and after each phrase. Your screen reader already reads what is typed, so QUILL Transcribe does not read the words back unless you ask it to in Dictation Settings.

With a large model there can be a few seconds between your pause and the text. So that the wait is not silent, QUILL Transcribe plays a soft tick once a second while a phrase is still being worked on. There is no tick for a phrase that arrives within a second. The first time in a session that a phrase takes four seconds, you also hear “Still working on that phrase.” The ticks stop as soon as the text arrives, or if recognition fails. While a tick plays, the microphone is not listened to for a moment, as with every QUILL sound, so the tick is never typed as speech. To turn the tick off, open Dictation Settings, choose More Dictation Settings, and clear Tick while a phrase is being worked on. It is on by default here and off in QUILL and QUILL Lite.

What you hear after each phrase in Dictation Settings decides the tick too. Set to nothing, there is no tick and no sentence. Set to words only, there is no tick; you hear “Still working on that phrase.” when a phrase has taken four seconds, each time that happens. If your sound pack has no tick sound, you hear the sentence at four seconds instead of the ticks, each time.

It speaks when something happens that your screen reader cannot tell you: that dictation paused, that it could not start and why, that a phrase may be incomplete.

On a braille display, while you are working in the other program, you see the words heard so far (“Hearing: dear Sam”) with a model that shows them while you speak, and the changes dictation shows without saying them: Command mode, Waiting for a command, and Listening for your wake phrase. These go to the display only and are not spoken. To hear the words so far as well, choose Show it, and say new words quietly for While you speak, the words heard so far in More Dictation Settings. You then hear each word once, when the model has settled on it, never a word it is still changing its mind about, and never the same word twice.

Dictation Status (Alt+F9 in the window, as in QUILL and QUILL Lite) says in one sentence what dictation is doing, the program it is writing into, the model, and how many phrases are held, for example “Dictation is paused in OUTLOOK, using Moonshine, with 2 phrases held.” To ask from another program, choose a Dictation Status Key (see Keys).

When the focus moves: paused, not lost

If you move to another field, another program or another window while dictating, QUILL Transcribe stops typing. It does not follow you. You hear “Dictation paused” and the reason.

While it is paused, it keeps listening and holds what you say. Nothing is typed anywhere. Then you choose:

• Resume (Ctrl+R in the window) - go back to the field you started in first. The held words are typed there and dictation carries on.
• Resume in This Field (Ctrl+Shift+R) - make the field you are in now the one dictation writes into. It gets the same safety checks as a fresh start. If you choose it from the tray or the window, it waits for you to switch to the field, as starting does. Choose it again to cancel the wait; your words stay held.
• Discard Held Words (Ctrl+Shift+D) - throw away what was held. Dictation stays paused until you resume or stop.

While paused, “scratch that” takes back the last held phrase, and “stop dictation” stops. If you stop or exit while words are held, you are told how many were not written, and exiting asks before throwing them away.

Up to 50 phrases are held. Held words are also kept on this computer, sealed for your Windows account, so a crash, a sleep or exiting loses nothing. Resume and Discard Held Words delete them. If you stop or exit while words are held, they stay for next time: stopping says “They are kept in Recent Phrases”, and the next start says how many are waiting. A phrase Windows stopped typing part-way through is kept the same way.

To resume or discard without coming back to QUILL Transcribe’s window, choose a Resume Key and a Discard Held Words Key in the File menu (see Keys). Pressed in the field you started in, the Resume Key types the held words there and carries on; the Discard Held Words Key throws them away.

Where it will not type

QUILL Transcribe refuses, with a sentence saying why, when:

• the field is a password field;
• the program is running as administrator and QUILL Transcribe is not, because Windows does not allow it;
• the focus is in QUILL Transcribe itself, or in a QUILL or QUILL Lite window. You hear “This is a QUILL window: press Ctrl+F11 to dictate here.” Both editors dictate into their own documents with Ctrl+F11;
• the program is on your list of programs dictation never types into. You hear, for example, “Dictation does not type into KeePass. Move to another program.” The list starts with the common password managers and the Windows sign-in, lock and administrator prompts. Change it in More Dictation Settings, with Programs Dictation Never Types Into… (Alt+N). Programs are matched by file name, like keepass.exe, never by a window title, and words already held stay held;
• the chosen model or microphone is not available.

While you use OpenAI or a speech provider, the state line ends “Speech goes to” and the company’s name, for example “Speech goes to OpenAI.”, because those are the only choices that send your voice off this computer (see Models).

If Windows does not take all of a phrase, QUILL Transcribe stops and says part of it may be missing. Check the field before you carry on. It never says a phrase went in when it is not sure.

Programs it has been tested with

QUILL Transcribe types into almost any program with a text field, but we only promise what we have tested. The table below comes from a test that types the same short dictation, phrase by phrase, into each program and then reads the program’s text back to check every phrase arrived exactly. It runs on a real Windows computer, not on our build servers, so a program that was not installed there is listed as untested, which does not mean it fails.

• Supported: at least 95 out of every 100 phrases arrived exactly.
• Best effort: at least half did. Dictation works, but check what was typed.
• Unsupported: fewer than half did.
• Untested: not tried yet on that version.

Programs update often, so each row says when it was checked and when it is due to be checked again. If a program you use behaves differently from its row, tell us which version you have.

Each tested program was sent the same 12 phrases. A phrase counts as exact when the program’s text, read back afterwards, grew by exactly that phrase. A program’s own formatting, such as Word’s curly quotes, and its own line ending are not counted as changes.

Commands while dictating into other programs

The command word

In QUILL Transcribe a command starts with the command word, QUILL: “QUILL scratch that”, “QUILL caps on”, “QUILL start spelling”. Said on its own, “scratch that” is typed as words, because a command typed into an email by mistake is much harder to take back than one in a document. Only the stop phrase, “stop dictation”, works without the word, and “stop spelling” while you are spelling. This is called strict mode, and it is on in QUILL Transcribe; QUILL and QUILL Lite leave it off.

• QUILL counts only as the first word after a pause. “I told Quill about it” is typed as words.
• Whatever follows “QUILL” is never typed. If it is not a command, you hear the error sound and “Not a command. Say it again, or say QUILL type that to write it.” Saying “QUILL type that” types those words after all, once.
• To type the word itself, say “literal QUILL”.
• Say “QUILL” on its own and pause, and the next thing you say is a command. A short rising tone tells you it was heard.
• “QUILL command mode” takes several commands in a row without the word. You hear the tone and “Command mode”, and the State field and the tray icon’s name say Command mode. Say “dictation mode” to go back. With no command for eight seconds it ends by itself with a falling tone, and what you say next is typed.
• The Command Key, if you chose one, does the same from the field you are in: press it, then say the command. Press it twice for command mode, and once more to go back. If dictation is off, it starts dictation in that field first. In the QUILL Transcribe window, Speak a Command (F11, the key QUILL and QUILL Lite use) does the same.

The command word, strict mode, and how long command mode waits are in More Dictation Settings (see “Dictation Settings”).

What works in another program

These work in any program, after the command word:

• “stop dictation” (with or without the word)
• “scratch that” - removes the last phrase, but only while you are still in the same field
• spelling: “start spelling” and “stop spelling”. While spelling, each phrase joins the letters you spelled just before it with no space, so you can spell an address a few characters at a time: “juliet oscar”, a pause, “at sign echo x-ray”, a pause, “dot charlie oscar” types jo@ex.co. Say “space” for a space; a whole word such as “smith” is not a letter and is left out
• capitals and spacing: “caps on”, “all caps on”, “no space on” and their “off” forms
• “switch to Spanish” and “switch to English”
• punctuation and layout: “comma”, “period”, “new line”, “new paragraph” and the rest. A Markdown link works too: “open bracket the guide close bracket open paren page close paren” types [the guide](page), with no space between the bracket and the parenthesis
• emoji and rarer characters: a word in My Words that is replaced with an emoji is typed as one character, even in a plain box such as the Run box (Windows+R). A program whose font has no picture for it may show a box, but your screen reader reads it by name
• your snippets, abbreviations and clips: “insert snippet” and the snippet’s name, “insert abbreviation” and the abbreviation, “paste clip” and a number from one to twelve. They come from the same lists QUILL uses, and are typed as one phrase, so “scratch that” takes all of it back
• “what can I say” for a short spoken answer
• correcting the words dictation itself typed: “correct that” and “choose one” to “choose three”, “capitalize that”, “all caps that”, “no caps that”, “replace” some words “with” new ones, “change that to” and new words, “delete last sentence” and “delete previous paragraph”. See “Correcting in another program” below

A snippet with blanks asks for each one before anything is typed. Say “QUILL insert snippet bug report” and you hear the snippet’s name, how many blanks it has, and the first blank’s name, like “Summary?”. Say the answer; spoken punctuation works in it. For a choice you hear the options, “Choose NVDA, JAWS or Narrator.”: say one, or “choose two”. Say “skip” to leave a blank empty, “scratch that” to answer the one before again, or “cancel” to type nothing. The command word is allowed before an answer but not needed. When the last blank is answered, the whole snippet is typed as one phrase, so “QUILL scratch that” takes all of it back. A date or time in a snippet comes from this computer’s clock.

One limit: where a snippet or an abbreviation would put the cursor in the middle, the cursor stays at the end, because QUILL Transcribe cannot move it inside another program’s text without reading it. For the same reason, “QUILL read that” reads back what dictation typed, all of it, but “read last sentence” and the other commands that read a document work only in QUILL.

Commands that need to read the other program’s text, such as “select that”, “go to”, “correct” and a word, or “copy all”, only work in QUILL’s own documents. In another program you hear “That works in QUILL’s own documents” and nothing changes.

Correcting in another program

QUILL Transcribe never reads the other program’s text. What it can do is remember the words it typed there itself, since the focus last moved, and change those: it takes them back with Backspace and types the corrected words, starting from the first letter that changes.

• “QUILL correct that”, then “QUILL choose two”: with Moonshine, Whisper or a downloaded model, the other built-in engine listens to your last phrase again, and you hear both answers, for example “1: the whether is nice. 2: the weather is nice. Say choose and the number.” With Windows speech recognition you hear its own other guesses. The phrase is retyped with the one you choose.
• “QUILL replace Tuesday with Wednesday” looks only in the phrase you dictated last. If you said “see you on Tuesday at noon”, Backspace takes back “Tuesday at noon” and “Wednesday at noon” is typed. If the words are not in that phrase, you hear “Not found.” and nothing changes.
• “QUILL change that to see you Friday” puts the new words in place of the last phrase, keeping its capital and closing full stop.
• “QUILL all caps that”, “QUILL capitalize that” and “QUILL no caps that” change the capitals of the last phrase.
• “QUILL delete last sentence” and “QUILL delete previous paragraph” take back the end of what dictation typed, never anything before it.

With strict mode off, you can say these without QUILL.

To keep your own typing safe, a correction is refused, with one sentence, and nothing is sent, when:

• the focus has moved since dictation typed the words: “The focus has moved since dictation typed those words, so they were left alone.”
• you have typed, clicked or moved the mouse since then, because the cursor may no longer be where dictation left it: “You have typed or clicked since dictation typed those words, so they were left alone. Change them with the keyboard, or say them again.” Windows tells QUILL Transcribe only when a key or the mouse was last used, never which key. Reading keys count too, so if you arrow back to check a word, change it with the keyboard instead.
• the program is Word or Outlook, for the commands that retype words. Those programs change words as you type them (AutoCorrect), so the count of letters to take back could be wrong: “Word can change words as they are typed, so dictation does not retype them there. Say scratch that and say it again.” “Scratch that” and the two delete commands still work there, because they only take words away.

Whatever a correction replaced goes into Recent Phrases (Shift+F11), marked “(before correction)”, so you can type it again with Insert Again.

What can I say

Saying “what can I say” while you dictate does not open a window, because a window would take the focus and pause dictation. Instead you hear: “In another program, you can say QUILL scratch that, new line, QUILL start spelling, QUILL insert snippet and its name, or stop dictation. For the full list, open Dictation Commands from the Dictation menu, Ctrl+Shift+C.” With strict mode off, the commands are said without QUILL.

Dictation Commands (Ctrl+Shift+C, in the Dictation menu) lists exactly what works in other programs, and nothing that does not, followed by your own words and phrases from My Words and Phrases.

Review before it is typed

Sometimes you want to hear and check what you said before it goes into the email. Say “QUILL review that”: the last phrase is backspaced away again (or, if dictation was paused, the words it was holding are taken) and opened in the Review Dictation window. Nothing more is typed into the other program until you insert it.

• Read the words with the arrow keys. They are in an ordinary text box, so your screen reader reads them by line, word and character, and a braille display shows them. Type to change them, or keep talking: while the window is open, what you say is written into it, and the program you were in is left alone. This is an expected pause, so you do not hear “Dictation paused”.
• Ask the AI to change them: say “QUILL make it shorter”, “QUILL more formal” or “QUILL fix the grammar only”, or type the instruction in Ask the AI to change it (Alt+K) and press Enter. You hear “Working”, then what changed, in two sentences at most, such as “Changed next Thursday to Friday. Removed um, so.” QUILL Transcribe works that out by comparing the words itself, never from the AI’s own account. Talk about it (Alt+A) makes everything you say an instruction, without the command word.
• Insert (Ctrl+Enter, or say “QUILL insert it”) goes back to the field you were dictating into and types the words there once, with the same checks as every phrase: the same field, no password field, nothing on the list of programs it never types into. If the field cannot be typed into, nothing is typed, you are told why, and the words are in Recent Phrases (Shift+F11) for Insert Again. “QUILL scratch that” takes the inserted words back out.

The window’s other keys: Ctrl+R reads all the words (“QUILL read it back”), Ctrl+D says again what changed (“QUILL what changed”), Ctrl+Z undoes the AI’s last change (“QUILL undo that change”), Ctrl+Shift+Z goes back to exactly what you said (“QUILL back to what I said”), and Escape stops a request on its way or discards the words (“QUILL discard it”). From the QUILL Transcribe window, Review Dictation (Shift+F4, in the Dictation menu, the key QUILL and QUILL Lite use) opens it too, empty if there is nothing to take.

The AI. QUILL Transcribe has no AI menu of its own. It uses what you set up in QUILL: your ChatGPT subscription (Use My ChatGPT Subscription) or your own key (Use My Own AI Key). Without either, or in Safe Mode, AI is off: “QUILL review that” says “AI is off. Your words were written as you said them.” and leaves the phrase typed, and the window still opens from Shift+F4 to edit and insert by hand. The first time, QUILL Transcribe asks before anything is sent and names who receives it; No is the default and sends nothing. Only the words in the window and your instruction are sent, with My Dictation Instructions if you wrote any; nothing is ever read from the other program. Each change you ask for is one request.

To send everything you say to the review window first, turn on Send dictation to review first (Alt+V) in More Dictation Settings.

Polishing what you said, only when you ask

Polish tidies punctuation, capital letters and small slips without rewording anything, before the words are typed into the other program. It is one request you ask for each time, it is off until you turn it on, and it never runs by itself.

1. Open Dictation Settings (Alt+Shift+F6), press More Dictation Settings, and check Allow polishing my dictation with AI when I ask (Alt+G). Press OK in both windows.
2. Dictate into your email as usual. When you want the last phrase polished, say “QUILL polish that” (or “QUILL polish it”). The phrase is backspaced away and opened in Review Dictation, then the polish request is made there. If Review Dictation is already open, the words in it are polished.
3. In the Review Dictation window you can also press Polish (Ctrl+P). The button is there only while polishing is allowed.
4. You hear “Polishing”, then what changed. Read the words, then Insert (Ctrl+Enter, or “QUILL insert it”) types them into the field once, or Undo Change (Ctrl+Z) takes the polish back.

The AI is asked to fix punctuation, capital letters, spelling and obvious grammar slips only, and to keep your words, meaning, order and tone. It uses the same AI and the same first-time question as the rest of the review: nothing is sent until you say yes, and only the words in the window go. With AI off the Polish button is greyed out, and with polishing not allowed “QUILL polish that” says “Polish my dictation is off” and sends nothing.

Correcting yourself as you talk

With Fix it when I correct myself, like Thursday, no, Friday (Alt+X) checked in Dictation Settings, QUILL Transcribe types only what you meant when you correct yourself in the middle of a sentence:

The marking words are “no”, “I mean”, “no I mean”, “sorry” and “scratch that word”, and a correction is made only between two days, two months, two numbers or two names from the Vocabulary list in My Words and Phrases. Everything else is typed as you said it: “no, I don’t think so” and “Is it Thursday? No, Friday.” stay as they are. The fix happens on this computer before anything is typed, with no AI. It is off until you turn it on.

Dictation style (Alt+Y), in the same window, sets these choices together: Verbatim (every word as you say it), Clean (no filler words, and your corrections fixed) or, only when AI is set up in QUILL, Polished (Clean, plus Send dictation to review first and Allow polishing my dictation with AI). Choosing Polished sends nothing by itself. Change a check box yourself and the style shows Your own mix of the settings below.

Remembering a correction. When you fix the same word the same way twice with “QUILL correct that” and then “QUILL choose two” (see “Correcting in another program”), QUILL Transcribe offers to add it to My Words and Phrases, so it comes out right next time. The question is spoken, never shown in a window that would take the focus from your field: “Add a correction: Ann becomes Anne? Say QUILL yes to add it, or QUILL no.” Answer within 30 seconds. Yes adds it to the Corrections list, which you can change in My Words and Phrases (Alt+Shift+F10); No, silence or anything else adds nothing. Each pair is offered once each time QUILL Transcribe runs, and only a word or a short phrase of up to four words.

Profiles: the kind of writing you are doing

A dictation profile changes dictation for one kind of writing, without you changing several settings each time. There are seven:

Standard is where everybody starts, and is your Dictation Settings exactly as they are.

Choosing a profile

• Dictation Profile in the Dictation menu (Ctrl+Shift+F), or in the tray menu, lists the seven with a sentence about the one you are on, and has the Lock the profile check box (Alt+L). OK uses it from your next phrase.
• By voice, after the command word: “QUILL email mode”, “QUILL markdown mode”, “QUILL standard mode”, and so on (“email profile” works too). You hear “Email profile.” and the choice is kept.
• More Dictation Settings in Dictation Settings has the same choice (Alt+A) and lock (Alt+B).

Lock Profile (Dictation menu, Ctrl+Shift+L) stops a spoken profile name from switching profiles: you hear “Lock Profile is on, so the profile stays Email.” You can still choose one from the menu. QUILL Transcribe never switches profiles on its own, for example by which program you are in.

The profile in use shows in the window’s Model and microphone field, the tray’s third line, the tray icon’s name and Dictation Status.

Headings, bold and italic in other programs

QUILL Transcribe never reads the other program’s text, so:

• In Markdown and Coding, “heading two” at the start of a phrase types “## “ before your words, and “QUILL bold that” retypes the phrase you just dictated with two asterisks either side, as one step that “QUILL scratch that” takes back. “QUILL italic that” uses one asterisk.
• In Document and Email, other programs cannot be formatted by keystrokes QUILL Transcribe can trust, so “heading two” starts a new line, writes your words and says “Headings work in QUILL documents or the Markdown profile.” Bold and italic say the same about themselves.
• In Chat and Verbatim, “heading two” starts a new line with your words and nothing else, and bold and italic say “This profile writes plain text, so nothing was formatted.”

Example. In an email, choose the Email profile, then say “um thanks for the notes” and pause. “Thanks for the notes.” is typed: the filler is gone and the full stop added.

Profiles and the Dictation style. While a profile other than Standard is on, its own choice about filler words wins over the Dictation style in Dictation Settings: Verbatim keeps fillers even if the style is Clean, and Email, Chat and Coding remove them even if the style is Verbatim. A profile never changes Fix it when I correct myself or polishing. Standard leaves your style exactly as it is.

The settings are stored as windows_dictation_profile and windows_dictation_profile_locked, under the same names as QUILL’s and QUILL Lite’s, in QUILL Transcribe’s own copy of the dictation settings.

Models

A speech model is the part that turns what you say into words. QUILL Transcribe works the moment it is installed: the built-in Moonshine is chosen for you, and it needs nothing downloaded and no internet. Everything in this chapter is optional. Read it when you want fewer mistakes, faster text, Spanish, or a speech company you already pay for.

What you can choose from

• Moonshine is the quickest. In our tests with real recordings it took about a twentieth of a second of computing for each second you speak, used at most 584 MB of memory, and got 6.5 percent of words wrong on audiobooks no model was trained on. Your words arrive about a fifth of a second after you pause. It knows only English: with the dictation language set to Spanish, QUILL Transcribe uses the built-in multilingual Whisper instead, without you changing anything.
• Whisper is a little slower, used at most 521 MB, and got 8.3 percent of English words wrong. It knows English and Spanish, but in Spanish it got about one word in four wrong (30.5 and 27.0 percent), so on a computer with 8 GB a downloaded model does much better in Spanish.
• Windows speech recognition is Windows’ own recogniser. It needs nothing extra and works in any speech language installed in Windows, but it does not punctuate by itself, and it often mishears a voice it has not been trained on. Its row says “the languages Windows has” and has no tier.
• Windows voice typing (Windows+H) is in the Speech engine list in Dictation Settings, because QUILL and QUILL Lite can hand over to it. It is Windows’ own feature, so QUILL Transcribe does not run it: with it chosen, dictation does not start, and you are told to press Windows+H instead.

Moonshine and Whisper come in the installer and the portable zip, with the part that runs them, so nothing is fetched the first time you dictate.

Choosing a model

Open Choose Model from the Model menu (Ctrl+M). The focus starts in the Models list (Alt+M). Each row says, in this order:

• whether it runs on this computer or sends your speech over the internet;
• its tier: Lightweight, Balanced or High accuracy (Windows speech recognition and OpenAI have none, because they are not models on this computer);
• the languages it knows;
• “not downloaded”, if it is not on this computer yet.

For example: “Moonshine (on this computer; Lightweight; English and Spanish)”, or “Whisper small (on this computer; High accuracy; English and Spanish; not downloaded)”. The built-in Moonshine row says English and Spanish because, in Spanish, the multilingual Whisper does its listening.

About this model (Alt+A) says more about the row you are on: whether your speech leaves the computer, whether it is ready, what its tier means, its languages and size, who publishes it, its licence, and one line on what it is good for. Press Enter, or Use This Model, to choose it. It stays your model until you choose another. A model that is not downloaded is listed so you know it exists; choosing it says it is not downloaded yet and points you to Model Library… (Alt+L), which is also a button in this window.

You can also choose in the Speech engine list in Dictation Settings (Alt+Shift+F6). That list shows the built-in engines, the models you have downloaded (not the ones you have not), OpenAI when a key is saved, and the speech providers you have set up. Both windows change the same setting.

What the tiers mean

The tier says how much work a model is for your computer, in the same words everywhere in QuillVille:

• Lightweight: keeps up comfortably even on a modest computer. The built-in Moonshine and Whisper, Moonshine base, Whisper base.en, and both Parakeet models.
• Balanced: more work than the built-in engines; keeps up on a computer with four or more cores. Nemotron and Whisper base.
• High accuracy: the most work; needs a capable computer, or patience. Whisper small and larger. Despite the name, none of these that we tested on real recordings made fewer mistakes than Parakeet Unified.

The tiers follow how fast each model ran on real speech. Until October 2026 they followed tests with a synthetic voice, which put both Parakeet models in Balanced and Nemotron in High accuracy.

“Keeps up” means your words arrive soon after you pause. A model that does not keep up still writes everything, only later: you hear the soft working sound once a second while you wait (see What you hear).

The models you can download

All of these run on the processor, so no graphics card is needed, and one would not make them faster. All of them keep your speech on this computer. “Fits a computer with” is the smallest amount of memory the model runs in beside a screen reader.

“Words wrong” is the percentage of words each model got wrong on public-domain audiobooks no model was trained on (see “How we chose: tested on real speech”); lower is better, and a dash means it was not tested there.

A few differences worth knowing:

• Parakeet Unified made the fewest English mistakes of every model that keeps up on one processor thread, and is what QUILL Transcribe recommends for English on a computer with 8 GB of memory and four cores.
• Parakeet TDT v3 is lighter still, and the most accurate model for Spanish: 4.4 and 7.3 percent of words wrong on the two Spanish sets.
• Nemotron shows your words while you are still talking: on a braille display you see “Hearing:” and the words so far. The others write each phrase when you pause.
• Parakeet never invents words out of silence. Whisper models sometimes write a phantom “thank you” when you said nothing.
• Moonshine base now hears long phrases. Before October 2026 it wrote nothing at all for a phrase longer than about eight and a half seconds; it now hears long phrases in pieces and joins them. The pieces overlap by half a second, so a word that falls where two pieces meet is not lost.
• Whisper medium, large-v3-turbo and large-v3 are very slow on most computers, and too slow for us to test on real speech.

Models removed in October 2026

Five English-only models were removed on 22 October 2026: Distil-Whisper small.en, Whisper small.en, Distil-Whisper medium.en, Whisper medium.en and Distil-Whisper large-v3. Whisper base.en and every Whisper model without “.en” in its name stay. In our tests with real recordings, NVIDIA Parakeet Unified 0.6B got 3.2 percent of words wrong on the held-out audiobooks, against 3.6 to 6.3 for these five. It also ran faster on one processor thread, in the same or a smaller memory tier.

If one of them was your model:

• When you update, your model becomes Parakeet Unified if you have already downloaded it, and otherwise the built-in model. Nothing is downloaded.
• The next time dictation starts, you hear one sentence saying which model was removed and what dictation uses now. You hear it once across every QuillVille program, not once in each.
• Its downloaded files are not deleted. They stay in %LOCALAPPDATA%\QuillVille\Dictation\models (in a portable copy, data\dictation\models inside the portable folder), in a folder named whisper-small.en-int8, whisper-distil-small.en-int8, whisper-medium.en-int8, whisper-distil-medium.en-int8 or whisper-distil-large-v3-int8. Delete that folder to free the space.
• To use Parakeet Unified instead, press Ctrl+L for the Model Library, move to NVIDIA Parakeet Unified 0.6B, press Download… (Alt+D), and then Use for Dictation (Alt+U). If it is already downloaded, choose it in Choose Model (Ctrl+M).

Which one should I choose?

Start from what matters most to you.

• Privacy, and working offline. Every built-in and downloaded model keeps your speech on this computer and works with no internet. Only OpenAI and the speech providers send it away.
• Speed. Moonshine. Nothing else answers as quickly, on any computer.
• Fewer mistakes. On a computer with 8 GB of memory and four cores, NVIDIA Parakeet Unified 0.6B: 3.2 percent of words wrong, against Moonshine’s 6.5. With two cores, Parakeet TDT v3 is lighter and nearly as good.
• Memory. On a computer with 4 GB, or with little memory free, stay with the built-in Moonshine (English) or Whisper (Spanish), or the 4 GB rows above, because your screen reader needs its share too.
• Spanish. On a computer with 8 GB, Parakeet TDT v3. On 4 GB, the built-in Whisper. Nemotron and the Whisper models without “.en” in their names know Spanish too.
• No downloads, in any language Windows has. Windows speech recognition, if you do not mind saying the punctuation aloud.

If you are not sure, keep Moonshine and read the recommendation.

The recommendation

Recommended for this computer (Alt+C in Choose Model) is a read-only box. It names the model that suits your computer and your dictation language, and says why: whether it keeps up, about how soon your words appear after you pause, its tier, and how big the download is, or that it is already installed. Set Up QUILL Transcribe has the same line (Alt+P).

The Model Library has four controls for it:

• What matters most (Alt+W): A balance of speed and accuracy (where it starts), The quickest response, or The fewest mistakes.
• Recommendation (Alt+C): the advice for that choice.
• Show Recommended Model (Alt+S): moves to that model in the list, so you can read its details and download it.
• Switch to Recommended Model (Alt+H): makes that model your model and changes nothing else. The Model Library closes. It never downloads anything: if the model is not on this computer yet, you hear “… is not on this computer yet. Download it, then switch.”, and you are taken to its row so you can press Download… (Alt+D). While a model is downloading, it asks you to wait for the download or press Cancel Download first. If you opened the Model Library from Dictation Settings, press OK there to keep the change; Dictation Settings says “Speech engine:” and the model’s name, then “Press OK to keep it.”

It is only advice. QUILL Transcribe never changes your model by itself, and OpenAI and the speech providers are never recommended. The figures come from our tests with real recordings, on one computer, and the advice says so; yours may differ.

What it recommends depends on memory as well as speed. A model must fit beside your screen reader: up to 600 MB on a computer with 4 GB, and up to 1.6 GB on a computer with 8 GB (it used to be 1 GB). It must also fit in what is free when you ask, less half a gigabyte kept back, so on a busy computer it may name a lighter model than usual. As a guide, this is what it usually suggests:

On a slower computer, or one with little memory free, it names a lighter model, down to the built-in Moonshine or Whisper.

How we chose: tested on real speech

The advice and the tiers come from a test on 22 October 2026 with recordings of real people, never a synthetic voice. The fairest set is public-domain audiobooks published on LibriVox after every model was trained, so no model can have learned them. Each model heard each recording one phrase at a time, on one processor thread, as it does when you dictate. “Speed per thread” is seconds of computing per second of speech; dictation’s budget is 0.25, a quarter of a core, leaving the rest for your screen reader.

English, percent of words wrong (lower is better):

The five English-only models removed after this test got 3.6 to 6.3 percent of words wrong on the held-out audiobooks (see “Models removed in October 2026”).

Spanish, percent of words wrong (lower is better):

The test ran on one fast laptop, so the apps judge speed against the built-in Moonshine’s speed on your own computer, measured by a two-second check. What it did not settle: how a real 4 GB computer behaves with NVDA or JAWS running, which still needs a person to try. The full record, with every model and set, is the benchmark report in QUILL’s documentation, docs/qa/dictation-real-speech-benchmark-2026-10-22.md.

A worked example

Sam has a laptop with 8 GB of memory and four cores, and dictates emails in English. Moonshine is quick, but Sam would like fewer mistakes.

1. Sam presses Ctrl+L for the Model Library.
2. In What matters most (Alt+W), Sam chooses The fewest mistakes. Recommendation (Alt+C) now names NVIDIA Parakeet Unified 0.6B, says it keeps up on this computer, and that it is a 663 MB download.
3. Sam presses Show Recommended Model (Alt+S), then Download… (Alt+D). A question names the size, the licence and where it will be saved. Sam answers Yes, and carries on working while it downloads.
4. When it has finished, Sam presses Use for Dictation (Alt+U), or Switch to Recommended Model (Alt+H). From the next phrase Parakeet Unified is the model, and the window’s Model and microphone field says so.
5. A colleague’s name still comes out wrong, so Sam adds it to My Words and Phrases (Alt+Shift+F10).

If Parakeet Unified turns out to be too slow while other programs are busy, Sam presses Ctrl+M, moves to Moonshine, and presses Enter. A question offers to delete Parakeet Unified to free its space; Sam presses Enter for No, and Parakeet Unified stays downloaded for another day.

The Model Library

Model Library (Ctrl+L, in the Model menu) is the same Speech Models window QUILL and QUILL Lite open from Dictation Settings with Better Accuracy: Speech Models… (Alt+B); that button is in QUILL Transcribe’s Dictation Settings too. It lists every model you can download. For the one you are on:

• Details (Alt+T) says what it is good for, its tier, its size, the accuracy its publisher reports, its licence and where it is saved. The publisher’s figure comes from their own tests; the recommendation uses ours, on real recordings (see “How we chose: tested on real speech”).
• Download… (Alt+D) asks first, naming the size, the licence and where it goes. Nothing is fetched until you answer Yes. On a metered connection it asks about that too. It checks the free space first, shows Progress (Alt+P), and carries on from where it stopped if the connection drops.
• Cancel Download (Alt+L) stops it.
• Use for Dictation (Alt+U) makes it your model.
• Switch to Recommended Model (Alt+H) makes the recommended model your model, when it is on this computer (see “The recommendation”).
• Remove (Alt+R) deletes it from this computer. Progress then says how much space was freed, and focus goes back to the list.

When you change model, you can free the old one’s space. After you choose another model in Choose Model or the Model Library, and the one you were using is a downloaded model still on this computer, a question titled Delete Old Speech Model asks, for example: “Now using NVIDIA Parakeet Unified 0.6B. Delete Whisper medium to free 946 MB?” It adds that QUILL, QUILL Lite and QUILL Transcribe share its files, so any of them still set to it will need it downloaded again. No is the default, so pressing Enter keeps the files. Yes deletes them, and you hear the model’s name, “deleted”, and the space freed. If it cannot be deleted, you are told that something may still be using it (“Remove it in Speech Models later”: Speech Models is the Model Library’s window). The question is not asked while dictation is running (remove the old model later in the Model Library), or when you move away from a built-in engine.

Every file comes from Hugging Face, is pinned to one exact version, and is checked before it is used. A download that did not finish is never marked ready. Models are saved in %LOCALAPPDATA%\QuillVille\Dictation\models (a portable copy keeps them in its own data\dictation\models folder). That folder is shared with QUILL, QUILL Lite and Quill Inkwell, so you only download a model once. In Safe Mode nothing downloads.

When your model cannot run

If your chosen model is missing or will not load, QUILL Transcribe does not quietly use another one. It says which model could not run and what to do, for example “Whisper small is not on this computer yet. Download it in the Model Library, choose another model, or use the built-in model this time.”

To dictate anyway, choose Use the Built-in Model This Time in the Dictation menu (Ctrl+B). It uses Moonshine for one session and says so. Your chosen model stays your choice, and the next session tries it again.

OpenAI

If you have an OpenAI account, QUILL Transcribe can send your speech to OpenAI to be recognised. It is very accurate and punctuates, and it needs the internet.

1. Save your OpenAI key. Open Dictation Settings (Alt+Shift+F6), press More Dictation Settings…, and press Add or Change OpenAI Key… (Alt+K). A key you already saved in QUILL’s Use My Own AI Key is used here too. The key is kept in Windows Credential Manager.
2. OpenAI now appears in Choose Model, as “OpenAI (sends your speech over the internet; many languages)”, and in the Speech engine list. It is never chosen for you.
3. Choose it. The first time, QUILL Transcribe tells you exactly what is sent and where. No is the default, and with No nothing is sent.

Which OpenAI model listens is chosen in More Dictation Settings: one writes your words as you speak, the other writes each phrase when you pause. OpenAI bills your own account. QUILL’s free AI never carries your speech. Safe Mode refuses OpenAI. While OpenAI has your speech, the window’s state line ends “Speech goes to OpenAI.”

Speech providers: your own account

QUILL Transcribe can also use a speech company you already have an account with. There are twelve: Microsoft Azure AI Speech, Google Cloud Speech-to-Text (Chirp 3), Amazon Transcribe, Deepgram, AssemblyAI, Speechmatics, ElevenLabs Scribe, Gladia, Rev.ai, Groq, Mistral Voxtral and IBM Watson Speech to Text.

To set one up:

1. Open Dictation Settings (Alt+Shift+F6) and press Speech Providers… (Alt+V). The focus is in the Providers list.
2. Choose the company. About this provider says what it is and what it needs. Fill in its fields, such as the key, the region or the project: press F1 on a field to hear where to find it in your account.
3. Press Test Connection (Alt+T). It asks the company whether the account works, and never sends speech. Last result says what happened.
4. Press Save (Alt+S), then Close. Remove (Alt+M) deletes a provider’s details.
5. Back in Dictation Settings, choose the company in the Speech engine list, where it is listed as, for example, “Deepgram (your own account; sends your speech to Deepgram)”. You are told exactly what is sent and where. No is the default, and with No nothing is sent. Press OK.

What you should know:

• Choose Model does not list the providers yet. Choose them in Dictation Settings.
• Keys go straight to Windows Credential Manager, never into a settings file or a log. The region and other plain details are kept with your dictation settings when you press OK.
• What is sent: your speech while you dictate, and the words in My Words and Phrases as hints, where the company accepts them. Nothing is sent while you are silent.
• Words while you speak: Azure, Amazon, Deepgram, AssemblyAI, Speechmatics, ElevenLabs, Gladia, Rev.ai and IBM Watson show your words as you talk. Google, Groq and Mistral write each phrase when you pause.
• When something goes wrong: if the network drops, that phrase is lost, you hear one sentence, and dictation keeps listening. A refused key or an empty account stops dictation. QUILL Transcribe never moves to another model by itself; Use the Built-in Model This Time (Ctrl+B) is there if you need to keep going.
• Cost: the company bills you. QUILL Transcribe shows no prices.
• Providers are for dictation. Transcribe a Recording does not use them.
• They are new, and have not yet been tried against every company’s real accounts. Safe Mode refuses them. While one has your speech, the state line ends “Speech goes to” and the company.

Dictating in Spanish

Dictation understands English, or Spanish (new; commands stay in English for now). Change it in Dictation Settings, with Switch Dictation Language (Ctrl+Shift+F11), or by saying “switch to Spanish” and “cambiar a inglés”.

• The built-in Moonshine and Whisper both use the built-in multilingual Whisper for Spanish, which gets about one word in four wrong. On a computer with 8 GB, NVIDIA Parakeet TDT 0.6B v3 is recommended instead: 4.4 percent of words wrong. Nemotron and the Whisper models without “.en” know Spanish too.
• A better Spanish model, said once. If your computer can run Parakeet TDT v3 (in practice, 8 GB of memory and four cores) and your Spanish would otherwise go to the built-in Whisper, the first time you dictate in Spanish you hear one sentence saying so, for example: “For Spanish, NVIDIA Parakeet TDT 0.6B v3 gets about 6 percent of words wrong, against 29 for the built-in model. It is a 670 MB download in the model list.” Nothing is downloaded or changed. You hear it once, in whichever QuillVille program you dictate in first. A computer with 4 GB hears nothing.
• A downloaded model that knows only English does not start in Spanish: you are told, and can choose another model or language.
• Windows speech recognition needs Spanish installed in Windows.
• Commands stay in English. Say “QUILL scratch that”, not a Spanish command. The wake phrase and stop phrase have Spanish forms: “Quill dicta” and “deja de dictar”.
• Spanish punctuation words such as “coma” and “punto” work while automatic punctuation is off, or with Windows speech recognition.

Limits

• Choose Model does not list the speech providers yet, nor Windows voice typing, which QUILL Transcribe does not run.
• The speed and accuracy figures come from tests with real recordings on one computer, so yours may differ. Whisper medium, large-v3-turbo and large-v3 were too slow to test on real speech, and a real 4 GB computer with a screen reader running has not yet been tried.
• A graphics card does not speed up any of these models.
• The recognition test’s sentences are in English.

Settings

Microphone

Choose Microphone (Ctrl+Shift+M) lists your microphones. The Windows default follows whatever Windows is set to. A named microphone stays chosen even if the default changes.

Too quiet, and it may not hear you at all. If the microphone’s level is set very low, or you speak very softly or far from it, dictation can miss a whole sentence, especially just after something louder. In our tests with very quiet recordings it missed some entirely. Move closer, or raise the level in Windows’ sound settings.

Dictation Settings

Dictation Settings (Alt+Shift+F6, the key QUILL and QUILL Lite use; Ctrl+, still works) is the same window QUILL and QUILL Lite use: language, punctuation, filler words, the dictation style and fixing it when you correct yourself (see Correcting yourself as you talk), how long a pause ends a phrase, what you hear after each phrase, and your dictation word list. Share dictation settings with every QuillVille program (Alt+Q, in More Dictation Settings) decides whether they are shared: when it is checked, what you save here is used by QUILL, QUILL Lite and Quill Inkwell too, and what you save in any of them is used here within a second. See Sharing dictation settings with the family, below. The wake phrase and holding the key to talk work here too, both off until you turn them on (see Holding the key to talk, and Starting with your voice).

More Dictation Settings in that window has Tick while a phrase is being worked on, on by default in QUILL Transcribe (see What you hear). Its Commands group has the Command word (Alt+C; QUILL unless you change it, and empty for none, so only the Command Key gives commands), Commands only after the command word (strict mode) (Alt+E; on in QUILL Transcribe), and Seconds command mode waits for a command (Alt+D; 8, from 5 to 30). A common word, or one of your own words in My Words and Phrases, cannot be the command word, and you are told why.

Sharing dictation settings with the family

When Share dictation settings with every QuillVille program is checked, QUILL, QUILL Lite, QUILL Transcribe and Quill Inkwell use one set of dictation settings.

• What is shared: the model and engine, the language, the microphone, automatic punctuation, the pause, filler words, the wake phrase, the command word, review first, the never-type list and the other dictation choices.
• What stays QUILL Transcribe’s own: its keys and tray, the working tick and strict mode (both on here), what you hear after each phrase, and every permission to send your speech to a company.
• When: saving Dictation Settings with OK, or changing the language, shares at once. A change saved in QUILL or QUILL Lite is used here within a second, without a word from QUILL Transcribe: it is your own change.
• To begin with: checked for a new install; unchecked if you had already changed your dictation settings, so nothing you chose is replaced. Check it when you want to join in.
• Nothing is lost: the first time QUILL Transcribe takes settings from the family, its own are saved as before-sharing-transcribe.json in %LOCALAPPDATA%\QuillVille\Dictation. If two programs save at once, the last one saved wins. If the shared copy cannot be saved, you hear “Saved here, but the other QuillVille programs will not see this change”, and why.

For example: in QUILL, choose a different model in Dictation Settings and press OK. The next phrase you dictate into Outlook with QUILL Transcribe uses that model.

Your words, the language, and your last phrases

• Recent Phrases (Shift+F11, in the Dictation menu) lists the last 20 phrases you dictated, newest first, in the same window as QUILL and QUILL Lite. Insert Again types the phrase into the program you go back to: QUILL Transcribe says “Switch to the field you want. The phrase is typed there.”, and types it, with the same checks as dictation, as soon as you are back in another program. Copy puts it on the clipboard and says “Copied” and the phrase. This run’s phrases are kept in memory only. After them come the words kept from before: “Not written” (held while paused, or typing that stopped), “Review not finished”, and, with Keep a dictation history on, “Earlier” phrases. Insert Again types one and then deletes it from the computer; Discard (Alt+D) deletes it unwritten; Clear History (Alt+H) deletes the history after asking, with No as the answer unless you choose Yes. When a correction changed a phrase, the words it replaced are in the list too, marked “(before correction)”; Insert Again and Copy leave the label out.
• Keep a dictation history (Alt+Y in More Dictation Settings) is Off unless you choose One day, Seven days or Thirty days. Only the text is kept, never the sound, sealed for your Windows account. Turning it off deletes it.
• My Words and Phrases (Alt+Shift+F10, in the Dictation menu) is the same window as in QUILL and QUILL Lite, over the same word list as QUILL, so a word you teach one is known to the other. The list lives in %LOCALAPPDATA%\QuillVille\Dictation; the first time it is needed after the update it is copied there from QUILL’s data folder, and the old file is left as it was, as a backup. QUILL Lite shares it once you check Use the shared word list in its own My Words and Phrases window. Export… saves the list to a Markdown or CSV file, and Import… adds the new entries from one after showing you what it will add; nothing you have is replaced. You hear the counts, such as “Imported 12 words, 3 already there.”
• Switch Dictation Language (Ctrl+Shift+F11, in the Dictation menu) changes between English and Spanish and keeps the choice for next time. You hear “Español.” or “English.” If dictation is on, the next phrase is heard in the new language. Saying “switch to Spanish” or “switch to English” does the same.

The same keys as QUILL and QUILL Lite

Where QUILL Transcribe has the same dictation command as the editors, it has the same key: Ctrl+F11 starts and stops, Alt+Shift+F6 opens Dictation Settings, Alt+Shift+F10 opens My Words and Phrases, Ctrl+Shift+F11 switches the language, Shift+F11 opens Recent Phrases, F11 is Speak a Command, Shift+F4 is Review Dictation, Alt+F9 is Dictation Status, and Shift+F5 transcribes a recording. The keys from version 1.0 (Ctrl+D, Ctrl+, and Ctrl+O) still work. These work in the QUILL Transcribe window; from other programs, use your start and stop key.

Three of the editors’ dictation commands are not in QUILL Transcribe, because they work on a document and QUILL Transcribe has none: Live Transcript, Dictation Context and Tidy Dictated Text.

If you choose Dictate in Other Programs in QUILL or QUILL Lite while QUILL Transcribe is running, your editor’s dictation settings are handed to QUILL Transcribe.

Keys

There are six keys you can choose, all in the File menu. Only the start and stop key has one to begin with; the others are not set, and nothing is taken from other programs, until you choose them:

• Start and Stop Key (Ctrl+Alt+Shift+D) - chooses the key that starts and stops dictation from any program: Alt+Shift+Backslash unless you change it. Type the key, or press Record Key and press it.
• Command Key (Ctrl+Alt+Shift+K) - chooses the key that makes the next thing you say a command, from any program (see “The command word”).
• Show and Hide Key (Ctrl+Alt+Shift+H) - shows and hides the QUILL Transcribe window from any program.
• Dictation Status Key (Ctrl+Alt+Shift+A) - says Dictation Status’s one sentence from any program.
• Resume Key (Ctrl+Alt+Shift+U) - when dictation is paused, types the held words into the field you started in and carries on.
• Discard Held Words Key (Ctrl+Alt+Shift+Y) - when dictation is paused, throws the held words away.

Type the key, or leave the box empty for none. If the family already uses the key, you are told which app uses it. If another program has already taken it, Windows says so, but Windows does not say which program, so neither can QUILL Transcribe.

Options

• Start QUILL Transcribe with Windows (Ctrl+Alt+W) - opens it in the tray when you sign in.
• Start minimized to the tray (Ctrl+Alt+M).
• Close button keeps QUILL Transcribe running (Ctrl+Alt+C) - on by default.

Transcribing a recording

Transcribe a Recording (Shift+F5 in the window, as in QUILL and QUILL Lite; Ctrl+O still works) turns an audio or video file into text with the same models, using the same window as QUILL. Choose the file and the model; it is transcribed in the background while you carry on.

For a recording in Spanish, check Translate into English (Alt+E) and the transcript is written in English instead. It needs a model that translates: the built-in Whisper, or a multilingual Whisper you downloaded (not the ones marked English). With any other model, or for an English recording, the box is unavailable.

When it finishes, you hear how long the recording was and how many words came out. Open Notifications (Ctrl+Shift+N) and press Enter on the transcript to review it. The Review Transcript window shows it read-only, with:

• Type into a Field - types the transcript, or only the part you selected, into another program. QUILL Transcribe says “Switch to the field you want. The phrase is typed there.”, and types it once you are there, with the same checks as dictation;
• Copy All - puts the whole transcript on the clipboard;
• Save As - saves it as a text file where you choose;
• Open in QUILL - opens it in QUILL to edit.

Nothing is typed, copied or saved unless you choose it.

Quill Inkwell and QUILL Transcribe

Quill Inkwell also has Dictate Anywhere. Only one of the two dictates at a time:

• While QUILL Transcribe is running, it owns dictation. Pressing Inkwell’s Dictate Anywhere key passes the press to QUILL Transcribe, so you get QUILL Transcribe’s dictation, with its pause when the focus moves.
• When QUILL Transcribe is not running, Inkwell dictates as it always has.
• If Inkwell is in the middle of dictating when you start QUILL Transcribe, QUILL Transcribe waits until Inkwell stops.

The first time QUILL Transcribe runs, it brings over the Dictate Anywhere settings and key you chose in Inkwell. Inkwell’s own settings are not changed.

Dictate Here from Quill Cast and Quill Radio

Quill Cast and Quill Radio have no speech engine of their own. Their Edit > Dictate Here (Ctrl+F11) asks QUILL Transcribe to dictate into the field you are in, such as Cast’s Add Podcast box or Radio’s Search Stations box.

• If QUILL Transcribe is running, it starts dictating there, just as if you had pressed your start and stop key, and you hear its usual “Dictation on”.
• If it is not running, it starts in the tray and begins dictating when it is ready; Cast or Radio says “Starting QUILL Transcribe. Dictation starts here when it is ready.” The first time, when QUILL Transcribe opens its window to set up, it does not start dictating; finish setting up, go back to the field and press Ctrl+F11 again.
• If Quill Inkwell is dictating, you hear that Inkwell is dictating, and nothing starts.

Your words, commands and never-type list all work as usual, because it is QUILL Transcribe doing the dictating. Stop with your start and stop key, or say “stop dictation”.

Dictate Here is on the Edit menu of every copy of Quill Cast and Quill Radio from QUILL Transcribe’s release on 1 November 2026. For example, in Quill Radio press Ctrl+F to search for stations, press Ctrl+F11 in the station name box, say “jazz”, pause, and press Enter.

The other QuillVille apps

QUILL Transcribe is one of the QuillVille family of free apps. The QuillVille menu (Alt+Q) in its window opens the others. Each row has its own letter, so Alt+Q and then one letter opens an app with nothing held down:

• Open QUILL: Alt+Q, then Q (or Ctrl+Alt+Shift+F7)
• Open Quill Radio: Alt+Q, then R (or Ctrl+Alt+Shift+F8)
• Open Quill Weather: Alt+Q, then W (or Ctrl+Alt+Shift+F9)
• Open Quill Converter: Alt+Q, then V (or Ctrl+Alt+Shift+F10)
• Open Quill Inkwell: Alt+Q, then I (or Ctrl+Alt+Shift+F11)

If the app is already running it comes to the front. If it is not on your computer yet, QUILL Transcribe offers to download it, and nothing is downloaded until you say yes.

The other apps list QUILL Transcribe in the same way, last on their QuillVille menu: Alt+Q, then T in QUILL, Quill Radio, Quill Weather and Quill Inkwell, and Alt+U, then T in Quill Converter. Its key there is Ctrl+Alt+Shift+F11 (Ctrl+Alt+Shift+F10 in Quill Radio, whose menu does not list Inkwell). The tray menus of Quill Radio, Quill Weather and Quill Inkwell have Open QUILL Transcribe too.

One microphone: QUILL, QUILL Lite and QUILL Transcribe

Only one program listens at a time, so the same words are never written twice. The microphone belongs to whichever of QUILL, QUILL Lite, QUILL Transcribe and Quill Inkwell started dictating first, for as long as it is dictating. QUILL Transcribe waiting in the tray does not hold it, unless the wake phrase is on: then it holds the microphone while it listens for the wake phrase, and lets go whenever QUILL or QUILL Lite comes to the front.

• If QUILL or QUILL Lite is dictating when you press your start and stop key, QUILL Transcribe does not start, and says who is and the key that stops it, for example “QUILL Lite is dictating. Press Ctrl+F11 in QUILL Lite to stop it, then try again.” QUILL’s Locked Dictation (Ctrl+F9) counts too.
• If QUILL Transcribe is listening when you press Ctrl+F11 in QUILL or QUILL Lite, the editor does not start a second microphone. It names your start and stop key: “QUILL Transcribe is listening. Press Alt+Shift+Backslash to stop it, then try again.” With no key chosen, it says “Stop it from its tray icon” instead.

The microphone is let go when dictation stops, when it stops because something went wrong, when the window closes, and when the program exits for any reason.

Where QUILL Transcribe’s own sounds play

QUILL Transcribe makes short sounds of its own, such as the ones that mark dictation starting and stopping. You choose where they play, and you can turn them off.

• Alt+Shift+M turns QUILL Transcribe’s sounds off. Press it again and they come back, on the same sound card as before. You hear a short sound as they go off, and another where they now play when they come back on, and both are spoken too. Options > App Sounds has a check mark that shows whether they are on.
• Options > App Sounds Play On… (Ctrl+Shift+S) lets you pick where they play:
  • System default. This is where it starts. The sounds go where Windows sends sound.
  • None (no sounds). QUILL Transcribe makes no sounds at all. Everything it tells you is still spoken.
  • Any sound card or headset in the list. The sounds get a device of their own.

After you choose, one sound plays on the new device, so you know straight away where it went.

If the device you chose is unplugged or asleep, you still hear every sound, on the system default, and QUILL Transcribe tells you once. When the device comes back, the sounds go back to it by themselves.

This changes QUILL Transcribe only. Your other QUILL apps keep their own choice, unless you have checked “Share these choices with my other Quill apps”, which copies a change here to the others that share. Your screen reader is never moved.

What’s New

What’s New (Ctrl+Shift+F1, on the Help menu) opens QUILL Transcribe’s changelog in your browser: everything that changed, grouped by area. It is on the same key in every standalone QuillVille app. If a build has no changelog, QUILL Transcribe says so instead of doing nothing.

Updates and release channels

Check for Updates (Ctrl+Alt+U, on the Help menu and the tray menu) looks for a newer QUILL Transcribe and offers to download and install it. Nothing is installed unless you say so.

QUILL Transcribe comes in three flavours, and you choose which one your copy follows. They are called release channels.

• Stable is the version we recommend. It has been checked with JAWS and NVDA, and it is what most people use. Every copy starts here.
• Beta gets new features a few weeks early. It is mostly finished, but some things may not work right yet.
• Dev is the work in progress. Things will break. Only choose it if a developer asked you to, or you enjoy testing.

1. Open the Help menu, then choose Release Channel… (press Alt+H, then L).
2. Arrow through Stable, Beta and Dev. The box below tells you what each one means and what choosing it would do. Nothing changes yet.
3. Press Switch to move, or Close (Escape) to leave things as they are.
4. For Beta or Dev, a short warning comes first: what could go wrong, how your settings are protected, and how to come back. Read it, check the I understand box, and choose Move to Beta (or Move to Dev). Stay on Stable, or Escape, changes nothing.

Before it moves, QUILL Transcribe saves a copy of your settings and dictation choices. If it cannot, it stays where it is and tells you why. Your recordings and transcripts are not copied, and updates do not change them. Then it checks for updates on the new channel straight away. After that, Check for Updates offers the newest version on your channel.

Beta and Dev versions aren’t signed, so when you install one, Windows SmartScreen may warn that it comes from an unknown publisher; that’s expected, and choosing More info, then Run anyway, installs it.

When other QuillVille apps are installed too. They run on one shared engine on your computer. When an app moves to Beta or Dev, it gets its own copy of that engine, about 335 MB, so the apps you leave on Stable are never touched. It goes away by itself when your last app comes back to Stable.

If an update doesn’t start. After an update, QUILL Transcribe checks that the new version really opens. If it doesn’t open within two minutes, the update undoes itself and puts back the version you had, and the next time QUILL Transcribe starts it tells you so.

Coming back to Stable. Open Release Channel again and choose Stable. If you have a newer Beta or Dev version, QUILL Transcribe first checks that Stable can read your settings; if it can’t yet, it offers to wait for Stable or to put back the copy saved when you joined, and saves a copy of how things are now first, so nothing is thrown away.

Safe Mode

In Safe Mode, QUILL Transcribe does not listen, registers no keys, downloads no models and sends nothing over the internet. The window and tray menu still open so you can look at your settings.

Privacy

• Nothing is recorded before you start dictation.
• Recordings of your dictation are never kept. Words that were not written (held while paused, or typing that stopped part-way) are kept on this computer, sealed for your Windows account, until you write or discard them in Recent Phrases.
• The text of what you dictate is kept only if you turn on Keep a dictation history, for one, seven or thirty days; Clear History deletes it, and turning it off deletes it too. It is off unless you choose it.
• Words waiting in Review Dictation are kept in a file in QUILL’s data folder while the window is open, so a crash does not lose them, and the file is deleted when you insert or discard them. Close keeps them for next time.
• QUILL Transcribe never writes what you said, or what is in other programs, to its logs.
• There is no usage tracking of any kind.
• Uninstalling QUILL Transcribe leaves your downloaded models, your dictation word list and your settings in place, because QUILL and the other apps use them too.

Licences and notices

QUILL Transcribe is free and released under the MIT License. It also includes software, data and sounds from other projects, each under its own licence.

To read them, open the Help menu, choose About QUILL Transcribe, and then choose the Licences and Notices button. The list opens as a plain text file, with one heading for each component: what it is, its licence, who holds the copyright and what the licence asks of us. The full licence texts are at the end.

The same file, THIRD-PARTY-NOTICES.txt, and QUILL Transcribe’s own licence, LICENSE.txt, are also in the folder where QUILL Transcribe is installed or unpacked, so you can open them without starting the app.

Keyboard shortcuts

In the QUILL Transcribe window:

In the Practice Dictation window, Ctrl+F11 starts or stops dictation in the practice field, and Escape closes it.

From any program, only these: your start and stop key (Alt+Shift+Backslash unless you changed it), and your Command Key, show and hide key, Dictation Status Key, Resume Key and Discard Held Words Key if you chose them.

F1 anywhere in QUILL Transcribe says what the window is for and what the control you are on does.

Getting help

Choose Get Help from Support in the Help menu (Ctrl+Alt+F2), or write to support@community-access.org.