QUILL Studio 2.0.0 - Release Notes

Built, not yet released.

QUILL Studio 2.0.0 follows Audio Studio 1.0.0

QUILL Studio 2.0.0 follows Audio Studio 1.0.0. The previously unpublished 2.2.0 build is renumbered to 2.0.0 with the QUILL Studio rename. Studio continues to use the shared QuillVille Runtime; each app has its own release version.

This build has not been published. QUILL Studio is not yet offered in the cross-app menu and there is no 2.0.0 download page. What follows is what is in the build, waiting on the release.

A new name: QUILL Studio

QUILL Audio Studio is now QUILL Studio. The window, the tray, the wizard, the installer and the shortcuts use the new name. Nothing else moves: an installed copy upgrades in place, your books, voices, settings and shared data stay where they are, and the old command-line name still works.

Screenshots, global hotkeys, and a Screenshot Inspector (AI optional)

A new Capture menu adds screenshot capture tools directly in Studio. These features are available without enabling AI; AI inspection is an optional extra for describing or discussing what is in an image.

Highlights:

Chapters for any recording

Studio > Find Chapters in a Recording… (Ctrl+Shift+R) picks any MP3, M4A or M4B – a lecture or a meeting as much as a book – and opens the Chapter Workbench on it with Propose chapters from silences already started. Proposals are titled “Proposed chapter 1, 2…”, so a guess is never mistaken for a real title. If the recording has chapters of its own you are asked whether to keep them and add only new boundaries (the default) or replace them; nothing is replaced silently. Press the button again (it reads Stop proposing chapters) to stop a scan.

Sliders that go the way you expect, and Slider Keys to choose

Up and Page Up raise the player’s Position and Volume sliders and Down and Page Down lower them, in every Quill app. Home and End go with them: Home is the top of a volume slider, 100 percent, and End the bottom. If you would rather have the keys a standard Windows slider uses, open Studio > Preferences… (Ctrl+,) and choose the Slider Keys… button: Up is more (the default) or As Windows sliders (Down, Page Down and End raise it). It is one choice for every Quill app at once. Right is always more and Left less, and each key press tells your screen reader the new value. Ctrl+Up and Ctrl+Down, the volume keys that work from anywhere in the apps that have them, are not changed.

The Studio’s own sounds, where you want them

The App Sounds Play On… button in Studio > Preferences… (Ctrl+,) sends the Studio’s own sounds to the system default, to any sound card or headset by name, or nowhere. Alt+Shift+M turns them off and back on where they were, and Studio > App Sounds shows a check mark while they are on. Your screen reader is never moved.

What’s New, and a guide that matches the menus

Help > What’s New (Ctrl+Shift+F1) opens the Studio’s changelog in your browser, on the same key as every other Quill app. The user guide has a new section on the speech-recognition engines (Whisper.cpp, the QUILL speech models such as Parakeet and Nemotron, and Faster Whisper) and how to download them, and its keyboard reference now matches the menus: Alt+V is the View menu and Alt+I the Voices menu.

Install Python once, and every Quill app starts instantly

The single biggest change is how the Studio is delivered. Every Quill app - QUILL, Quill Radio, Quill Weather, and QUILL Audio Studio - now shares one Python engine, the QuillVille Runtime, installed once per user and reused by all of them. Install it a single time with any Quill app, and every app you add afterward launches immediately instead of downloading its own copy of everything. The runtime is reference-counted: it is removed only when the last app that needs it is uninstalled, so removing one app never leaves another unable to start.

The Studio comes as two downloads. The portable zip (about 675 MB) runs from a USB stick with no installation and no internet, carrying a genuine, unmodified copy of Python plus the offline speech and text-to-speech engines. The installer installs the shared runtime, if it is not already present, plus the app. The installer now also offers a choice of setup: Full, Compact (the program without the bundled documentation), or Custom.

Earlier test builds also came as a small “Lite” installer and a “Companion” zip. Those are gone. Take the installer in place of the Lite installer, or the portable zip in place of the Companion zip.

The Studio’s launcher is now a genuine, tiny native program, and the bundled Python is the official unmodified build. Earlier versions used a renamed and modified copy of Python’s own launcher, a pattern some antivirus tools flagged as a false positive. That pattern is gone, so the Quill apps are far less likely to be flagged. Releases are still not code-signed, so SmartScreen may caution on first run.

Update in one click

When an update is available, choose Download, then Install and restart now. The Studio applies the update itself - extracting the new portable files over your folder, or running the installer quietly - and relaunches, keeping every setting, voice, and book position. No closing the app, unzipping, and swapping folders by hand.

Help > Check for Updates… now looks at the shared Quill project releases and picks out the Studio’s own download for the flavor you run. Each app updates on its own schedule from that one place, so an update to Quill Radio is never mistaken for an update to the Studio.

A status bar you can actually read

The Studio’s status bar used to be a strip of text a screen reader could not review. It is now arrow-navigable, exactly like the one in QUILL and Quill Radio: press F6 to move focus into it, Left and Right to move between cells, Enter to act on the one you are on, and Escape or a second F6 to leave. The cells are Activity, Progress, Sleep timer, Your books, and Time, so “what is going on right now” is always one key away.

The Progress cell carries a live narration or build run - the percentage, how many files are done, and the current step - and the Studio writes the same text into its system-tray tooltip. An overnight narration therefore stays reviewable with the window tucked into the tray, instead of going quiet the moment you minimize it. View > Show Status Bar turns the bar on or off and remembers your choice.

Converting audio, without leaving the Studio

Voices > Convert Audio… changes audio files between formats - MP3, M4A/M4B, Opus, Ogg, FLAC, WAV, AAC and more - and pulls the audio track out of video files (MP4, MKV, MOV, WebM and the rest) along the way. It converts through the ffmpeg the Studio already bundles, so there is nothing new to install, nothing is uploaded, and it works in Safe Mode.

Build a queue of files and whole folders, pick a destination, and choose one of ten presets: Just convert (no processing), MP3 at 320, 192 or 128 kbps, Podcast, Audiobook M4B, Voice memo, Web voice (Opus), Archival (FLAC), and Hearing-aid mono. The format list only ever offers what your ffmpeg can genuinely encode, so a long batch cannot die half way for a missing encoder. Conversion runs in the background with several files at once, a real Cancel, and a spoken summary at the end that names the files that failed rather than quietly reporting success. Progress appears in the status bar’s Progress cell and the tray tooltip, the same as a narration run.

An Advanced panel is there when you want it, and neutral when you do not: loudness normalization to audiobook (ACX) or podcast level, gain, a high-pass filter, silence trimming, pitch-preserving tempo, a compressor, a night-mode leveler, fade in and fade out, and a safety limiter.

Voices > Convert from URL… takes a link - YouTube and the many other sites the downloader supports - fetches the audio, and drops it straight into the converter. The downloader is never bundled: it installs on demand after a one-time consent and rights notice, and the whole feature is unavailable in Safe Mode.

Announcements reach your braille display

A braille user reported that messages the app speaks never appeared in braille. The diagnosis was blunter than the report: nothing had ever been sent to a braille display at all - not by this app, not by any of them. A whole output channel was missing.

Braille is now written at the same point as speech, so every message the Studio makes is covered at once. Three rules govern it:

Behind that, speech, braille, sound, and the status line are now delivered by one shared announcement service - the same one QUILL uses - so a channel added anywhere appears in every app. Announcements are recorded as they go, so a message that passed while you were reading something else is not simply gone.

New shared accessibility settings control the braille style (the full spoken string, or a compact position-first form), how long an identical braille message is held back, whether an error sticks on the display instead of flashing past, whether the companion apps play sound cues, whether a cue stands in for speech while Quiet Mode is on, and which severities interrupt speech. The Studio’s own Preferences dialog stays the short list it has always been; these live in the settings store the Studio shares with QUILL, so setting them in QUILL applies here too.

A transcription engine that cannot make things up

Whisper — the engine most offline transcription runs on, the Studio’s included — has one famous flaw: fed silence, it sometimes invents text. A pause in a recording can come back as a phantom “thank you” or a subtitle credit that nobody said, because under the hood Whisper is a language model guessing at what silence “should” say.

The Studio now offers Parakeet 3, an NVIDIA model that works the other way round: it only writes down words it actually heard, so silence produces nothing — which matters for exactly the material a studio transcribes, long recordings full of pauses. It covers 25 languages and detects which one is being spoken, runs on any CPU with no graphics card and no heavy libraries, and downloads (about 650 MB, checksum-verified from QUILL’s own mirror) from the same model manager as every other engine. Every engine and model in that manager now also states its capabilities in plain language — “detects the spoken language; never invents text from silence” — before you download, so choosing is no longer guesswork.

Alongside it, a silence pre-pass now trims the quiet lead-in and tail from captured audio before any engine hears it, and your custom vocabulary (names, product words, jargon) is fuzzy-matched into finished transcripts so a name the engine mangles still comes back spelled your way.

Built from QUILL’s own source

The Studio used to be assembled from a copy of QUILL’s audio code kept in a separate repository and re-synced by hand. It is now built directly from QUILL itself. Nothing changes in what you see, but the standalone Studio and QUILL’s built-in Audio Studio can no longer drift apart, and a fix made upstream is in the next Studio build with nothing to copy across.

Two smaller results of the same work: Piper voices now download from the project’s own signed mirror rather than a third-party model host, so a voice download no longer depends on an outside service being reachable from your network; and the shared components each app needs (ffmpeg, for the Studio) are counted, so uninstalling one Quill app can never take away something another app still needs. The build also leaves out the large libraries the Studio never loads, making both the installer and the portable zip meaningfully lighter.

An icon of its own

Until now QUILL Audio Studio shipped a byte-identical copy of Quill Radio’s blue broadcast-wave icon – not a similar drawing, the same file. So did Quill Inkwell and Quill Weather. On a desktop with more than one Quill app installed, four different products wore one face in the taskbar, in Alt+Tab, in the Start menu and in the notification area, which for a tray-resident app is where it lives its whole life. Nobody chose that; each new app was built from the last one’s template, and an icon is easy not to notice.

Every app in the family now has its own. They are still recognisably a set – one rounded tile shape, one gold accent – but each has its own colour and its own picture. The Studio’s is a three-bar waveform on a dark slate tile: editing audio rather than broadcasting it, and the only glyph in the family built from repetition.

The drawing was made for the size that actually matters. At 16 by 16 pixels – the notification area, and the small icons in a file list – a first attempt with five bars merged into a solid slab. There are three, with real gaps between them, because gaps are what make a waveform read as bars rather than as a block.

Fixed

Alt+B opens the Book Tools menu again. The main window’s Build From Recordings button also claimed Alt+B, so Alt+B started a new book. The button is now Build From Recordings… on Alt+R, and Alt+B opens Book Tools.

Edit a Book opens again. It stopped with an error before its window appeared. It opens, and Ctrl+Shift+T (Sleep Timer) and Ctrl+Shift+Q (Play Queue) now work inside it.

Resume Last Book on Launch is Ctrl+Alt+F10. It was Ctrl+Alt+Shift+R, Quill Radio’s system-wide show and hide key, so with Radio running the menu key never fired.

Kokoro narration no longer stops mid-run on a portable copy. A long, dense passage could overrun the neural model’s input window; the Studio then fell back to a heavyweight code path a portable copy does not carry, and announced “Kokoro voices need one more component” - for a component you had already installed, 29 chapters into a run. The Studio now re-splits that one passage at safe boundaries and carries on, so a portable narration completes. The per-passage size limit is capped below the model’s window everywhere it is used, translated editions included, and when a passage genuinely cannot be synthesized the message names the real cause and where the report is, instead of pointing at a download you do not need.

Known limitations