QUILL Transcribe Changelog

All notable changes to QUILL Transcribe are recorded here, newest first. The release notes (release-notes-1.0.md) tell the same story as a narrative; this file is the itemized list.

1.0.0 - 2026-11-01

Testing rebuild: 1.0.0 (build 3). The packages have not been published yet. QUILL Transcribe 1.0.0 is the first public release, planned for 1 November 2026; there was no earlier public version, so everything below is new in 1.0.0, including what the first test builds (builds 1 and 2) already did. Build 3 adds Help > What’s New and keeps everything else. The changes are grouped by area; most say where in the user guide to read more.

Dictating into other programs

• Start and stop dictation with a key you choose, from the tray menu, or from the window. No key is registered until you choose one.
• Words are typed into the field that had the focus when dictation started, and only that field.
• When the focus moves, dictation pauses and holds what you say. Resume writes the held words back where you started; Resume in This Field moves dictation to the field you are in, after the same checks as a new start; Discard Held Words throws them away.
• Starting from the tray or the window waits until you are back in another program, and gives up after 20 seconds.
• Password fields, programs running as administrator, and QUILL’s own windows are refused with a sentence that says why.
• A phrase Windows did not fully take is reported as possibly incomplete, never as written.
• The same commands, punctuation, spelling, capitals and spacing modes and word list as dictation in QUILL. Commands that need to read the other program’s text are refused politely.
• The start and stop key is Alt+Shift+Backslash from the start. Change it, or empty it so the tray menu starts dictation instead, with File > Start and Stop Key… (Ctrl+Alt+Shift+D; Alt+F, then K). Every key box has Record Key: press the combination you want and it is written out for you; typing the key still works. When Windows says your key is taken, QUILL Transcribe now sends you to the File menu, where the row is, rather than the Dictation menu. User guide: “Keys”.
• Three more keys you can choose, each empty until you choose it in the File menu: the Dictation Status Key (Ctrl+Alt+Shift+A), the Resume Key (Ctrl+Alt+Shift+U) and the Discard Held Words Key (Ctrl+Alt+Shift+Y), so you can ask, resume or discard from the program you are working in.
• Hold the key to talk. With Hold the dictation key to talk checked in More Dictation Settings (Alt+H, off until you turn it on), hold your start and stop key while you speak; letting go stops dictation and keeps your last phrase. What you say while the key is down is typed when you let go, so nothing reaches the other program while Alt or Ctrl is held. A quick press still turns dictation on and off. It is the same setting as QUILL’s and QUILL Lite’s. User guide: “Holding the key to talk”.
• The wake phrase. With Listen for the wake phrase while dictation is off checked in Dictation Settings (off until you turn it on), say “Quill dictate” in the field you want and dictation starts there. While it waits, the State field and the tray say Listening for your wake phrase, and nothing it hears is written, kept or sent until it hears the phrase. It does not listen while QUILL Transcribe, QUILL or QUILL Lite, or a program on your never-type list, is in front. User guide: “Starting with your voice: the wake phrase”.
• Programs dictation never types into. A list you keep, in More Dictation Settings (Alt+N), that starts with the common password managers and the Windows sign-in prompts. Dictation and Insert Again refuse a listed program by name and type nothing; held words stay held. User guide: “Where it will not type”.
• Resume in This Field, from the window or the tray: choosing it again cancels the wait, and says your words are still held.
• Programs it has been tested with. A new section of the user guide lists Notepad, WordPad, Word, Outlook, Edge, Chrome, Firefox and Visual Studio Code, each marked supported, best effort, unsupported or untested, with the version and the day it was checked. So far every row says untested: a program that was not tried is never listed as unsupported.

What you hear, and what is typed

• Quiet with your screen reader. Words are typed as the keys of your keyboard layout, so a screen reader no longer says “c” before or after each phrase. With phrase feedback set to a tone or to nothing, QUILL Transcribe presses Ctrl straight after the words, which every screen reader takes as “stop talking”, so you no longer hear “period” after every phrase. With the read-back on, you still hear everything. User guide: “What you hear”.
• Saying the same phrase twice writes it twice. It was taken for the screen reader’s echo and dropped.
• A tick while a phrase is worked on. With a large model there can be seconds between your pause and the text. When a phrase takes more than a second, a soft tick plays once a second until it arrives; the first time in a session that one takes four seconds, you hear “Still working on that phrase.” On by default here; turn it off in More Dictation Settings, Tick while a phrase is being worked on. With What you hear after each phrase set to words only there is no tick, only the sentence; set to nothing, there is neither.
• A calmer spoken preview. With Show it, and say new words quietly chosen for the words heard so far, each word is said once, when the model has settled on it.
• Dictation on your braille display. While you work in the other program, the words heard so far (“Hearing: …”) and the changes that are shown but not said (Command mode, Waiting for a command, Listening for your wake phrase) appear on a braille display, without being spoken.
• Dictation Status (Alt+F9 in the Dictation menu, as in QUILL and QUILL Lite): one sentence with what dictation is doing, the program it is writing into, the model, and how many phrases are held. “Speech goes to OpenAI.” (or the speech provider’s name) ends the state line while an online engine has your speech.
• Emoji arrive whole. An emoji, or any character beyond the first 65,536 in Unicode such as a rarer Chinese character, is typed as one character. A plain edit box such as the Run box (Windows+R) used to receive half of it and show a box or nothing. For example, if a word in My Words is replaced with a thumbs-up emoji, saying that word in the Run box now types the thumbs-up. Limit: a program whose font has no picture for an emoji still shows a box, though your screen reader reads it by name.

Commands, spelling and correcting

• The command word: “QUILL scratch that”. A command starts with QUILL, and whatever follows QUILL is never typed into the other program. Said on its own, “scratch that” is typed as words (strict mode, on in QUILL Transcribe; “stop dictation” always works). If what follows QUILL is not a command you hear “Not a command”, and “QUILL type that” types it. “QUILL command mode” takes commands without the word until “dictation mode” or eight seconds with none. Choices in More Dictation Settings: the command word (Alt+C), strict mode (Alt+E) and the seconds (Alt+D). User guide: “The command word”.
• Command Key and Speak a Command. An optional system-wide key, empty until you choose it in setup or in File > Command Key… (Ctrl+Alt+Shift+K): press it in the field and the next thing you say is a command; twice, command mode. Dictation > Speak a Command (F11, as in QUILL and QUILL Lite) does the same from the window. Two new sounds play when QUILL hears the command word and when command mode ends.
• Spell an address with pauses. While spelling, each phrase joins the letters you spelled just before it, with no space: “QUILL start spelling”, “juliet oscar”, “at sign echo x-ray”, “dot charlie oscar”, “stop spelling” types jo@ex.co. Say “space” for a space. Limit: a whole word such as “smith” is not a letter and is left out; spell it.
• A Markdown link by voice. “open bracket the guide close bracket open paren page close paren” types [the guide](page), with no space between the bracket and the parenthesis.
• Snippets, abbreviations and clips by voice, in any program. “insert snippet” and its name, “insert abbreviation” and its name, and “paste clip” and a number are typed as one phrase, and “QUILL scratch that” takes all of it back. A snippet with blanks asks for each blank by name and each choice by its options, then types the whole snippet once; say skip, scratch that to go back, or cancel. “QUILL read that” reads back all of what dictation typed.
• What can I say is a spoken answer while dictating, so dictation is not paused, and the Dictation Commands window lists only what works in other programs, with your own words and phrases.
• Correcting in another program. “Correct that” and “choose one” to “choose three”, “capitalize that”, “all caps that”, “no caps that”, “replace Tuesday with Wednesday”, “change that to” and new words, “delete last sentence” and “delete previous paragraph” work on the words QUILL Transcribe typed there itself: it takes them back with Backspace and types the corrected words. It never reads the other program. A correction is refused, with one sentence and nothing sent, if the focus has moved, if you have typed, clicked or moved the mouse since, or, for the commands that retype words, in Word and Outlook, which change words as they are typed. User guide: “Correcting in another program”.
• A second opinion for “correct that”. With Moonshine, Whisper or a downloaded model, the other built-in engine listens to your last phrase again, and you can choose its answer. The sound stays in memory only.
• Dictation profiles: Standard, Email, Document, Markdown, Chat, Coding and Verbatim, each changing the pause, filler words, punctuation and how headings, bold and italic are written. Choose one in Dictation > Dictation Profile (Ctrl+Shift+F), the tray menu, More Dictation Settings, or by voice (“QUILL email mode”). Lock Profile (Ctrl+Shift+L) stops voice switching. Standard, the default, changes nothing. User guide: “Profiles: the kind of writing you are doing”.
• Bold and italic by voice: “QUILL bold that” and “QUILL italic that”. In the Markdown and Coding profiles they retype the phrase you just dictated with asterisks round it; in other profiles they say why they cannot format another program.
• Fix it when I correct myself (Alt+X in Dictation Settings, off until you turn it on): “meet on Thursday, no, Friday” types “meet on Friday”. It happens before anything is typed, on this computer, with no AI.
• Dictation style (Alt+Y in Dictation Settings): Verbatim, Clean or, when AI is set up in QUILL, Polished. One choice that sets filler removal, fixing your corrections and, for Polished, review first and polishing.

Review and polish

• Review before it is typed. Say “QUILL review that” and the last phrase is backspaced away and held in the Review Dictation window (Shift+F4 in the Dictation menu, as in QUILL and QUILL Lite). Read it, change it, keep talking into it, and with AI set up in QUILL, ask the AI to change it (“QUILL make it shorter”). “QUILL insert it” or Ctrl+Enter goes back to the field you were in and types it there once. You are asked before anything is sent to an AI. Send dictation to review first (Alt+V in More Dictation Settings) sends everything you say there. User guide: “Review before it is typed”.
• Polish what you said, when you ask. Check Allow polishing my dictation with AI when I ask (Alt+G in More Dictation Settings, off by default). Then “QUILL polish that” opens the last phrase in Review Dictation and asks the AI once to fix punctuation, capitals and small slips, keeping your words; Polish (Ctrl+P) in the window does the same. It never runs by itself.

Your words, settings and history

• Your dictation settings, shared with the family. More Dictation Settings has Share dictation settings with every QuillVille program (Alt+Q). When it is checked, what you save here is used by QUILL, QUILL Lite and Quill Inkwell, and what you save in them is used here within a second. Keys, the tray, the working tick, strict mode, what you hear after each phrase and every permission to send your speech stay QUILL Transcribe’s own. Checked for a new install; unchecked if you had already changed your settings. User guide: “Sharing dictation settings with the family”.
• The word list’s new home. My Words and Phrases lives in %LOCALAPPDATA%\QuillVille\Dictation, copied there from QUILL’s data folder the first time it is needed. The old file is left as a backup.
• Import and Export in My Words and Phrases (Alt+Shift+F10). Export saves your list to a Markdown file or a CSV file with say, write and kind columns. Import adds what is new from either, after showing how many it will add; nothing you have is replaced, and your list as it was is kept as dictation-before-import.md.
• A backslash in your own phrases is kept. A phrase that writes a folder, such as “my folder” writing C:\new\temp, used to read back as C:, a new line and ew, then a tab. It is now saved with each backslash doubled and reads back exactly as you typed it. To fix a phrase that already broke, open My Words and Phrases, choose it, Edit, and type the folder again. Limit: when you edit dictation.md by hand, write \\ for a backslash before n or t.
• The editors’ dictation keys work here. In the QUILL Transcribe window, Ctrl+F11 starts and stops, Alt+Shift+F6 opens Dictation Settings and Shift+F5 transcribes a recording; Ctrl+D, Ctrl+, and Ctrl+O still work. New here on the editors’ keys: My Words and Phrases (Alt+Shift+F10) and Switch Dictation Language (Ctrl+Shift+F11), both in the Dictation menu. User guide: “The same keys as QUILL and QUILL Lite”.
• Recent Phrases (Shift+F11): the last 20 phrases of this run, newest first, with what a correction replaced marked “(before correction)”. Insert Again types one into the program you go back to, with the same checks as dictation; Copy puts it on the clipboard.
• Nothing you said is lost. Words held while dictation is paused, and a phrase Windows stopped typing part-way, are kept on this computer, sealed for your Windows account, until you type or discard them, so a crash, a sleep or exiting loses nothing. Recent Phrases lists them after this run’s phrases, with a review you did not finish, and has Discard (Alt+D).
• Dictation history, off unless you want it. Keep a dictation history (Alt+Y in More Dictation Settings) keeps the text of what you dictate for one, seven or thirty days, never the sound. Clear History (Alt+H) in Recent Phrases deletes it after asking; turning it off deletes it too.

Models, microphone and speech providers

• Choose Model lists every model, marked as on this computer or over the internet, with its languages, size, source, licence and whether it is downloaded.
• The chosen model is the one that runs. A missing model, or one that will not load, stops the start and says so; Use the Built-in Model This Time uses Moonshine for one session, on purpose.
• OpenAI appears only with a saved OpenAI key, and asks before any speech is sent.
• The Model Library and Dictation Settings are the same windows QUILL uses.
• Choose Microphone, and a four-second microphone test in setup that stays on this computer.
• Which model suits this computer, and why. Choose Model (Ctrl+M) and Set Up show a read-only Recommended for this computer box: the model that suits your computer and dictation language, whether it keeps up, about how soon your words appear after you pause, and the download size. The Model Library (Ctrl+L) adds What matters most and Show Recommended Model. It is only advice; your model never changes by itself. Every model is labelled Lightweight, Balanced or High accuracy.
• Speech providers, your own account. Dictation Settings has Speech Providers… (Alt+V): twelve speech companies, from Azure and Google to Deepgram and Groq, each with only the fields it needs, a Test Connection that sends no speech, and keys in Windows Credential Manager. Choose one in the Speech engine list there (Choose Model does not list the providers yet); you are told what is sent and No is the default. If it cannot be reached, your model stays your choice. User guide: “Speech providers: your own account”.
• A proper guide to choosing a model. The user guide’s “Models” chapter explains every choice: the built-in Moonshine (the default, nothing to download) and Whisper, Windows speech recognition, the ten downloadable models with their sizes, languages, tiers and memory, the recommendation, OpenAI and the speech providers, with a worked example, “Dictating in Spanish” (commands stay in English) and “Limits”.
• Advice from real speech. The recommendation, the tiers and the figures now come from tests with recordings of real people (22 October 2026), not a synthetic voice, and the advice says so. For English on a computer with 8 GB and four cores it names NVIDIA Parakeet Unified 0.6B: 3.2 percent of words wrong on audiobooks no model was trained on, against 6.5 for the built-in Moonshine. With 8 GB and two cores it names Parakeet TDT 0.6B v3. In Spanish on 8 GB it names Parakeet TDT v3: 4.4 percent wrong, against 30.5 for the built-in Whisper. On 4 GB, or with little memory free, it stays with the built-in Moonshine (English) and Whisper (Spanish). Shared with QUILL and QUILL Lite. User guide: “How we chose: tested on real speech”.
• More memory on an 8 GB computer, and free memory counts. A computer with 8 GB may now give dictation 1.6 GB beside the screen reader (it was 1 GB), and never more than is free when you ask, less half a gigabyte. User guide: “The recommendation”.
• Tiers follow real speed. Parakeet TDT v3, Parakeet Unified and Whisper base.en are now Lightweight; Nemotron is Balanced. User guide: “What the tiers mean”.
• Removed: five English-only models. Distil-Whisper small.en, Whisper small.en, Distil-Whisper medium.en, Whisper medium.en and Distil-Whisper large-v3. On real recordings Parakeet Unified got 3.2 percent of words wrong on the held-out audiobooks, against 3.6 to 6.3 for these, and ran faster, in the same or a smaller memory tier. Whisper base.en and the multilingual Whisper models stay. If one was your model, it becomes Parakeet Unified if that is downloaded, otherwise the built-in model; nothing is downloaded, and you are told once, the next time dictation starts. Their files are not deleted: they stay in %LOCALAPPDATA%\QuillVille\Dictation\models until you delete the folder. To use Parakeet Unified, download it from the Model Library (Ctrl+L). User guide: “Models removed in October 2026”.
• Moonshine base hears long phrases. It wrote nothing for any phrase longer than about eight and a half seconds; it now hears them in pieces and joins them. The pieces overlap by half a second (23 October 2026), so a word that falls where two pieces meet is no longer lost. User guide: “The models you can download”.
• A better Spanish model, said once (23 October 2026). The first time you dictate in Spanish on a computer that can run Parakeet TDT v3 (in practice 8 GB of memory and four cores), while your Spanish would go to the built-in Whisper, you hear one sentence: “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.” (or “It is already downloaded; choose it in the model list.”). Nothing is downloaded or switched, a 4 GB computer hears nothing, and it is said once across every QuillVille program. At most one such sentence is said at each start: a downloaded model that gave way to the built-in one, then a removed model, then this. Shared with QUILL and QUILL Lite. User guide: “Dictating in Spanish”.
• Switch to Recommended Model (23 October 2026). The Model Library (Ctrl+L) has a new button beside Show Recommended Model: Switch to Recommended Model (Alt+H). It makes the recommended model your model and changes nothing else. A model not on this computer yet is never downloaded: you hear “… is not on this computer yet. Download it, then switch.”, and you are taken to its row. While a download is running, it asks you to wait or cancel first. Opened from Dictation Settings, the change is kept when you press OK there, and Dictation Settings says “Speech engine:” and the model’s name, then “Press OK to keep it.” User guide: “The recommendation”.
• Free the old model’s space (23 October 2026). After you choose another model in Choose Model or the Model Library, moving away from a downloaded model, a question titled Delete Old Speech Model offers to delete it, with No as the default: “Now using … Delete … to free …?” It says that QUILL, QUILL Lite and QUILL Transcribe share its files. Yes deletes them and says how much space was freed. It is not asked while dictation is running, or for a built-in engine. Remove now puts the space freed in Progress, and focus goes back to the list. Shared with QUILL and QUILL Lite. User guide: “The Model Library”.

Practice and lessons

• Practice Dictation (Ctrl+Shift+P in the Dictation menu): a window with a field of QUILL Transcribe’s own that dictation types into straight away, so you can learn and test without writing into another program. Start Dictation Here (Alt+H, or Ctrl+F11 in the window) starts and stops it. User guide: “Practice Dictation”.
• The recognition test. In Practice Dictation, Read This Sentence (Alt+T) clears the field and scores the first thing you say against the sentence shown: “All 9 words right”, or how many percent were right and which words were heard differently, with three suggestions below 80 percent. Score the Field (Alt+S) scores everything in the field.
• Tutorials (Ctrl+Alt+F1 in the Help menu): five short lessons (a first dictation, pausing and resuming, commands with the command word, snippets by voice, and review), in the same Tutorials window as QUILL, with a tutorial book beside the user guide.

Recordings

• Transcribe a Recording, with a read-only Review Transcript window to copy, save or open the result in QUILL.
• Transcribe a Recording translates into English. For a recording in Spanish, check Translate into English (Alt+E) and the transcript comes out in English, with the built-in Whisper or a multilingual Whisper you downloaded; with any other model the box is unavailable. The same in QUILL and QUILL Lite. User guide: “Transcribing a recording”.
• Type a transcript into a field. Review Transcript has Type into a Field (Alt+F): the transcript, or the part you selected, is typed into the program you go back to, with the same checks as dictation.

Setup, the window, the tray and sounds

• Set Up QUILL Transcribe: microphone, model, key, and starting with Windows, all on one page; Skip changes nothing.
• A calm window: state, model and microphone, what happens next.
• A tray menu that says what dictation is doing and which model it uses, and does everything the window does.
• Where QUILL Transcribe’s own sounds play, and Alt+Shift+M to turn them off. Options > App Sounds Play On… (Ctrl+Shift+S) offers System default, None (no sounds), or any sound card or headset by name. After a change one sound plays on the new card. If that card is unplugged the sounds play on the system default until it is back, and QUILL Transcribe says so once. Alt+Shift+M turns the sounds off and back on, and Options > App Sounds shows a check mark while they are on. Your screen reader is never moved. Asked for on QUILLVille by Rick Lewis. User guide: “Where QUILL Transcribe’s own sounds play”.
• Alt+W opens the Window menu again. The Discard Held Words button also claimed Alt+W, so Alt+W threw away held words instead of opening the Window menu. The button’s letter is now C (Alt+C). Dictation > Discard Held Words (Ctrl+Shift+D) is unchanged.

QUILL Transcribe and the QuillVille family

• Released on 1 November 2026. From then QUILL Transcribe is part of every public QuillVille build: it is on the QuillVille menu of QUILL, Quill Radio, Quill Weather, Quill Converter and Quill Inkwell (Alt+Q, then T; Alt+U, then T in Quill Converter), on the tray menus of Quill Radio, Quill Weather and Quill Inkwell, and offered for download from them when it is not installed. Nothing is downloaded until you say yes. QUILL Transcribe’s own QuillVille menu opens QUILL, Quill Radio, Quill Weather, Quill Converter and Quill Inkwell on Ctrl+Alt+Shift+F7 to F11. User guide: “The other QuillVille apps”.
• Dictate Here from Quill Cast and Quill Radio. Their Edit > Dictate Here (Ctrl+F11) asks QUILL Transcribe to dictate into the box you are in, starting it in the tray if it is not running, and shows in every copy. For example, in Quill Radio press Ctrl+F to open Search Stations, then Ctrl+F11, and say the station’s name. Ctrl+F11 there starts dictation but does not stop it; your own start and stop key does. User guide: “Dictate Here from Quill Cast and Quill Radio”.
• One microphone across the family. QUILL, QUILL Lite, QUILL Transcribe and Quill Inkwell share one microphone lock, held only while one of them is dictating, QUILL’s Locked Dictation (Ctrl+F9) included. Your start and stop key while an editor dictates names it and the key that stops it: “QUILL Lite is dictating. Press Ctrl+F11 in QUILL Lite to stop it, then try again.” Ctrl+F11 in an editor while QUILL Transcribe listens opens no second microphone and names your key: “QUILL Transcribe is listening. Press Alt+Shift+Backslash to stop it, then try again.” Waiting in the tray holds nothing, so the editors dictate as usual. User guide: “One microphone: QUILL, QUILL Lite and QUILL Transcribe”.
• QUILL Lite’s window is refused like QUILL’s. QUILL Transcribe does not type into QUILL Lite; it says “This is a QUILL window: press Ctrl+F11 to dictate here.”
• While QUILL Transcribe runs, Quill Inkwell’s Dictate Anywhere key is passed to it, so only one program dictates at a time. The first run brings over Inkwell’s Dictate Anywhere settings and key, without changing Inkwell.
• QUILL and QUILL Lite’s Dictate in Other Programs hands your editor’s dictation settings to QUILL Transcribe while it runs.

Help, updates and release channels

• Help > What’s New (Ctrl+Shift+F1), new in build 3, opens this changelog in your browser, on the same key as in every standalone QuillVille app. If the changelog is missing from a build, it says so rather than doing nothing. User guide: “What’s New”.
• Release channels: Stable, Beta and Dev (2026-10-24). Help > Release Channel… (Alt+H, L) chooses which versions your copy is offered, as in QUILL Lite, Quill Radio and QUILL Cast. Joining Beta saves a copy of your settings first, and going back to Stable checks it can read them. Check for Updates follows the channel. User guide: Updates and release channels.
• F1 help on every window and control.

Installing and Safe Mode

• Runs on Python 3.14, and older runtimes are cleaned up. The shared QuillVille runtime this build installs is Python 3.14. The installer then removes older runtimes (Python 3.13) that no installed QuillVille app uses any more; one that an app or a shortcut still points to is left alone. Portable copies carry their own Python and change nothing outside their folder.
• Safe Mode listens to nothing, registers no keys and downloads nothing.