QUILL Transcribe 1.0 Release Notes

Version 1.0.0 (build 3), released November 1, 2026.

QUILL Transcribe is new. It types what you say into the program you are using, with the same speech models and dictation you know from QUILL and QUILL Lite.

The short version

Choose your model, press your key in the field you want, and speak. Your words go into that field and nowhere else. Your speech stays on your computer unless you choose an internet model yourself.

New in build 3

Help > What’s New (Ctrl+Shift+F1) opens QUILL Transcribe’s changelog in your browser, on the same key as in every standalone QuillVille app.

Pinned to the taskbar, QUILL Transcribe opens as itself. Its shortcuts now carry its own taskbar identity, as Quill Radio’s have since 3.0.2, so pinning the running QUILL Transcribe pins its shortcut. An older pin that asks “Several QuillVille apps are installed” now opens the app you pick through its own launcher, with all of its settings. To replace an old pin, unpin it and pin QUILL Transcribe again after you update. Reported on QUILLVille for QUILL Lite; every QuillVille app has the fix.

New in this release

Everything in QUILL Transcribe is new, but if you tried a test build, these are the things that arrived since. The full list, grouped by area, is in the changelog (CHANGELOG in the docs folder); each part of the user guide named here explains one in full.

Starting and stopping. The start and stop key is Alt+Shift+Backslash from the start, chosen in File > Start and Stop Key… (Ctrl+Alt+Shift+D), and every key box has Record Key. Hold the key to talk, or say the wake phrase “Quill dictate” (both off until you turn them on). Three more keys you can choose ask for Dictation Status, resume, or discard held words from the program you are in. A list of programs dictation never types into is in More Dictation Settings (Alt+N).

What you hear. Your screen reader no longer says “c” or “period” after each phrase. A soft tick plays while a slow phrase is worked on. The words heard so far appear on a braille display, and Dictation Status (Alt+F9) says what dictation is doing in one sentence. Emoji are typed whole.

Commands and correcting. Say “QUILL” before a command so it is never typed (“QUILL scratch that”), or press F11 and say it. Spell an address across pauses, say a Markdown link, and insert snippets, abbreviations and clips by voice. “Correct that”, “replace Tuesday with Wednesday” and “delete last sentence” work in other programs on the words QUILL Transcribe typed itself. Dictation profiles (Ctrl+Shift+F), a dictation style, and “QUILL review that” (Shift+F4) are there when you want them. See “Commands while dictating into other programs”.

Your words and settings. Your dictation settings can be shared with QUILL, QUILL Lite and Quill Inkwell (Alt+Q in More Dictation Settings). My Words and Phrases (Alt+Shift+F10) imports and exports. Recent Phrases (Shift+F11) keeps what could not be typed until you type or discard it, and a dictation history is kept only if you turn it on.

Models. Recommended for this computer says which model suits your computer and why, every model has a tier, and twelve speech providers can be used with your own account. See “Models”.

Advice tested on real speech. The recommendation now rests on tests with recordings of real people, not a synthetic voice. For English on a computer with 8 GB of memory and four cores it names NVIDIA Parakeet Unified 0.6B, which got 3.2 percent of words wrong on audiobooks no model was trained on, against 6.5 for the built-in Moonshine. For Spanish on 8 GB it names NVIDIA Parakeet TDT 0.6B v3: 4.4 percent wrong, against 30.5 for the built-in Whisper. On 4 GB the built-in models are still the advice. Moonshine base no longer drops phrases longer than about eight and a half seconds, and no longer loses a word where a long phrase is cut: the pieces overlap by half a second. See “How we chose: tested on real speech”.

A better Spanish model, said once. 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 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 across every QuillVille program, and a computer with 4 GB hears nothing. See “Dictating in Spanish”.

Switch to Recommended Model, and free the old one. The Model Library (Ctrl+L) has a new button, Switch to Recommended Model (Alt+H). It makes the recommended model your model and changes nothing else. If the model is not on this computer yet, nothing is downloaded: you are told, and taken to its row so you can download it first. When you move away from a downloaded model, in Choose Model or the Model Library, a question offers to delete the old one. For example: “Now using NVIDIA Parakeet Unified 0.6B. Delete Whisper medium to free 946 MB?” No is the default, and the question says that QUILL, QUILL Lite and QUILL Transcribe share its files. It is not asked while dictation is running. Remove now says how much space was freed. See “The recommendation” and “The Model Library”.

Removed: five English-only models. Distil-Whisper small.en, Whisper small.en, Distil-Whisper medium.en, Whisper medium.en and Distil-Whisper large-v3 are no longer offered. Whisper base.en and every Whisper model without “.en” in its name stay. On 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, it becomes Parakeet Unified if you have already downloaded it, and otherwise the built-in model. Nothing is downloaded, and you are told once, the next time dictation starts. The 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, open the Model Library (Ctrl+L), download it, and press Use for Dictation (Alt+U). See “Models removed in October 2026”.

Practice. Practice Dictation (Ctrl+Shift+P) has a field of its own and a recognition test, and Help > Tutorials… (Ctrl+Alt+F1) has five lessons. See “Practice and lessons”.

Recordings and sounds. Transcribe a Recording can translate Spanish into English (Alt+E), and Review Transcript can type the transcript into a field (Alt+F). Options > App Sounds Play On… (Ctrl+Shift+S) chooses where QUILL Transcribe’s sounds play, and Alt+Shift+M turns them off. In the window, Alt+W opens the Window menu and Discard Held Words is Alt+C.

What is in 1.0

Dictation into any program

Put the cursor where you want to write, press your start and stop key, and speak. Each phrase is typed when you pause. Press the key again, or say “stop dictation”, to stop.

The start and stop key is Alt+Shift+Backslash unless you change it. To change it, choose File > Start and Stop Key… (Ctrl+Alt+Shift+D): type the key, or press Record Key… and press it. If you empty the box, the tray menu starts and stops dictation instead: choose Start Dictation, switch to the field you want, and dictation starts there.

Quiet with your screen reader

Your screen reader echoes what is typed, and because a phrase arrives faster than it can speak, you would hear only the last character: “period” after every sentence. When your phrase feedback is a tone or nothing, QUILL Transcribe presses Ctrl on its own straight after the words, which every screen reader takes as “stop talking”. When you have chosen to hear phrases read back, you hear everything, punctuation included.

Paused, never misplaced

If you move to another field or program while dictating, QUILL Transcribe stops typing and tells you. What you say next is held, not typed anywhere. Go back and choose Resume to have it typed where you started, choose Resume in This Field to write where you are now, or discard it.

The model you chose

The built-in Moonshine is chosen for you and works with nothing downloaded and no internet; the built-in Whisper is there too, and both know Spanish (Moonshine through Whisper’s multilingual model). Choose Model (Ctrl+M) shows every model with what matters: whether it runs on your computer or sends your speech over the internet, its tier (Lightweight, Balanced or High accuracy), which languages it knows, and whether it is downloaded. Recommended for this computer names the model that suits your computer, and why, from our tests with real recordings. Ten larger models, from 141 MB to 1.8 GB, download from the Model Library (Ctrl+L) only when you ask.

The one you choose is the one that runs. If it cannot run, you are told, and nothing else is quietly used instead. Use the Built-in Model This Time (Ctrl+B) is there when you want to keep going anyway.

Your own account, if you want it

OpenAI, with your own key, and twelve speech companies, from Microsoft Azure to IBM Watson, can do the listening instead. Set them up in Dictation Settings (Speech Providers…); keys go to Windows Credential Manager. Nothing is sent until you say yes, and No is the default. Choose Model does not list the providers yet: choose them in Dictation Settings.

Set up in one page

The first time it opens, QUILL Transcribe asks for a microphone, with a quick test, a model, a key if you want one, and whether to start with Windows. Skip changes nothing.

Recordings too

Transcribe a Recording turns an audio or video file into text in the background. Review the transcript, then copy it, save it, or open it in QUILL.

With Quill Inkwell

Only one program dictates at a time. While QUILL Transcribe runs, Inkwell’s Dictate Anywhere key is passed to it. The first time it runs, QUILL Transcribe brings over the settings and key you chose for Dictate Anywhere in Inkwell.

Dictate Here in 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 box you are in, starting it in the tray if it is not running.

Good to know

Questions and problems: Get Help from Support in the Help menu, or support@community-access.org.