Download StellAurator for Windows

Long jobs show what they are doing. Bringing in a manuscript, re-running who speaks, Find the beats, Bring your recordings, Remove this run's changes, Convert all, Regenerate with the new voice and Best fit auto assign now open the progress sheet: the real step ("Splitting the lines: chapter 12 of 54", "Reading scene 3 of 9"), a bar, the time taken, Cancel, and Keep working in the background. The three rotating lines the manuscript import used to show are gone; the sheet shows the parse's own steps. A cancel says what it kept.

The user agreement is version 1.6. Section 8(d) says beta where it said alpha; nothing else changed. The app asks you to accept it once, on the next start.

Everything in 0.51.0 is in this build too.

This installer is signed.

Conform by chapter. Conform to revised manuscript now opens a pairing screen instead of comparing the whole book at once: your chapters on one side, the file's on the other. Let it detect the order automatically or pair them yourself; press Match to see how well a pair matches ("97 % of the lines matched", with what changed, and whether chapters were added, removed or reordered before it); then Import changes. The changes go into a new edition, chapter by chapter, and the recorded book is never touched. Choose whether the recordings come along (a changed line's take is marked stale) or the edition is text only.

The old words stay with the line. In the new edition, every changed line shows a "Was:" note with the old words above it and wears a change mark; lines the file lost are shown where they were. The notes have their own menu (copy the old words, hide, mark as reviewed), and the changed lines are a scope: jump to the next or previous change, or record a part over just those lines.

The conform can no longer take the engine down. Comparing a revised manuscript with the book now runs beside the engine instead of inside it, so the app keeps answering while it works, and the comparison itself is bounded: a revised file whose chapter headings were not recognized used to arrive as one chapter against the whole book and ran the engine out of memory; the same case now takes about a second.

You can see the conform working. Conform to revised manuscript opens a sheet that says what it is doing ("Comparing chapter 12 of 54"), with a bar, the time it has taken, Cancel (nothing is written until the end, so canceling costs nothing) and Keep working in the background.

The app asks before a bad match. When the revised file reads as one chapter and the book has many, the sheet says so and asks whether to compare anyway, since every chapter but one would count as removed.

The app recovers its own engine. When the audio engine stops answering, the app ends it (only when it can prove the process is its own) and starts it again instead of giving up. If that does not work either, the box offers Try again, Show the log and Close. Every start, every missed probe and every restart is written with the time into engine.log in the data folder, so a problem leaves a trace.

Starting on a busy computer. The engine's launcher is no longer canceled by the app before the system has run it, which could leave the app with no engine at all on a loaded machine.

Everything in 0.50.0 is in this build too.

This installer is signed.

Finish tells you which lines fail. A chapter's verdict is no longer a sentence you have to go and act on by hand. Press it, or right-click it, and the failing lines are listed with the number that failed and the limit: each one jumps to the line, and Fix these lines… opens Restoration with exactly those lines selected and the ACX processing tool ready. Change the target… opens the export's loudness settings in place.

Export some chapters. The export dialog has a Chapters choice: the whole book, or the chapters you tick. They go into a folder of their own, the M4B's chapter marks start at 1, and the book's own files are never overwritten. Each target tile now shows its loudness numbers, and a right-click on a tile (or its three dots) changes the target.

The ACX warning names the lines. When a book mixes recorded and generated audio, the warning before export lists the generated lines by chapter, each a jump, with "Select them in Restoration". If you turned the warning off, a line under Export says so and turns it back on.

Restoration: everything on the page has its own menu. A baked stage (Hear up to here, Compare with before, Reopen its steps, Save as a preset, File as a new take, Changes, Undo this bake, which on an earlier stage sets that stage and the ones after it aside as a branch you can return to). A chain step (On, Move up and down, Duplicate, Reset to defaults, Save this step as a preset, Hear after this step, Show me, Remove). A toolbox tool (Add at the end, Add after the open step, Add to the default chain, Run on the selection now, Show me). A preset (Add to the chain, Use as the default, Rename, Delete). A line in the list (Show, Play, Select this chapter's or this speaker's lines, Go to the line, Flag for a pickup, Original or Processed, Remove all changes). The clip itself (Edit this clip, the clip editor, Original or Processed, Clip gain, View: compact, large or strip, Spectrogram, Suggested repairs). Every menu opens by right-click, by its three dots, and by the Menu key. The Large view's chevron now says what it does and the size is remembered; Spectral Repair opens the clip large by itself.

Everything in 0.49.0 is in this build too.

This installer is signed.

PDF and EPUB manuscripts open. New book and Add a chapter take a .pdf or an .epub as well as .docx, .txt and pasted text. A PDF's pages come in as the author wrote them: the lines of a paragraph joined, words broken at a line end mended, running headers and page numbers dropped, headings kept on their own line so the chapters are found. A PDF that is only pictures of pages (a scan) says so and asks for the manuscript from your writing program. An EPUB comes in chapter by chapter in its reading order, with its title and author.

Every visible thing has its own menu. Right-click a line, a take, the speaker's name, a chapter (in the list or its header), the book's title, Listen, the recording meter, a cue, a sound in the library, a line or a piece on the timeline, a line in Proof listen, a flagged line in Pickups, or the waveform in the editor, and a menu lists what that thing can do. The same menu opens from a three-dot button on the thing and from the keyboard (the Menu key, or Shift + F10). The common actions come first, the deeper ones under Advanced, and the last row opens Settings at the matching field. Settings keeps every setting too.

A zoom bar under the timeline. The bar is the chapter end to end; the lighter part is what you see. Drag its middle to scroll, drag either end to zoom from that side, double-click it to see the whole chapter. The playhead can be dragged. Ctrl + wheel still zooms.

Smaller things. Listen has a menu for speed, output, mono, lines ahead and follow playback; a new speed reaches the line already playing. The recording meter's menu changes the microphone, the count-in, when a take stops after quiet, monitoring and what happens after each take. Pickups' Record flagged now records. Double-click a take to open its waveform without playing it. The cue sheet's Move button is Edit this cue; Make with lists every music service, with a way to add a key for one that has none; the Score page's genre is set right there. Four messages that pointed at the wrong Settings tab point at the right field and open it.

Everything in 0.48.0 is in this build too.

This installer is signed.

Ducking. Music now dips while the voice speaks, in the exported files and while you listen. On the Score page, next to How much score, choose Continuous (the music goes down as a passage of speech begins, stays down through the small gaps between words, and comes back up in the pauses between lines) or Between the words (it dips under each word and comes back up even in short gaps), and how far it dips. Books start on Continuous at 6 dB. Sound effects are not ducked.

Your own plugins. Restoration has a Plugins button at the top right. It lists the VST 3 plugins on this computer, lets you add a plugin file or a folder of them, and puts one on the take you are working on as a step of its chain. The plugin's own window opens with the clip looping through it, so you hear what you set; Bypass compares. Like every other step, it can be switched off, undone, saved in a chain or applied to a whole chapter. Plugins run in a separate program, so a plugin that crashes cannot stop StellAurator or a recording.

The pickup list as a file. Pickups has Export, Import and Print at the top right. Export saves every flagged line as a spreadsheet file that opens in Excel or Google Sheets. Import reads that file back into the same book and flags those lines (a list from another book is refused, and a line already flagged is left alone). Print gives a page to check off, with a Done box for each line.

Also: the Bring your recordings screen opens a moment faster.

Everything in 0.47.0 is in this build too.

This installer is signed.

Readings are divided into lines on your computer. Read a chapter straight through and StellAurator now divides it into the manuscript's lines by the words you read, with a transcription model that comes with the app. You don't need a transcription service or a key, and nothing is sent anywhere. If you already connected a service, it is still used; to use the built-in one instead, choose On this computer (free) in Settings, API & AI, Transcription. When you import a recording made in another program, Transcribe on this computer is offered first.

Every take is checked against the script. A few seconds after you record a line, StellAurator listens to the take and compares it with the line. If a word was left out, added or changed, the take says so, for example: Reads differently: Left out "in there". Press Look to see the line and what was heard, and say why: on purpose (it sounds more natural, better delivery, timing), a mistake, or the transcription got it wrong. Names and numbers the transcriber can't know are not counted as differences. A change you keep on purpose is kept, and the script does not change unless you edit the line. You can turn the check off in Settings, Recording.

When a narrator records for you. The narrator sees the same check after each take. For a change made on purpose, they are asked to also record the line exactly as written and star the take they prefer. Their reason goes with the takes when they send them. You hear both and choose: Keep this reading, Keep the take read as written, or Ask for a pickup. Once you choose, that take is not flagged again.

Open beta. StellAurator is now in open beta, and the app says beta where it used to say alpha.

The installer is about 130 MB larger because it now includes the transcription model.

Everything in 0.46.3 is in this build too.

This installer is signed.

Create bed. The timeline on the Score page has a Create bed button at the top. Press it and it changes to Select area: drag across the timeline where the music should play, over the music lane or anywhere else. The start and end are set from your drag and shown in the bed form above, with the length. Fill in the mood and the rest as usual and press Create this bed. Press the button again, or Esc, to stop without making one.

Everything in 0.46.2 is in this build too.

This installer is signed.

No more stutter in the pauses. Since 0.46.0 the pauses between lines could stutter: they were filled by repeating a tiny piece of sound from the edge of a line, sometimes the end of a word, or by turning a very quiet noise print up too far. Now each chapter's room sound is made from the quiet moments inside its own lines, and a pause plays whole stretches of it joined smoothly, so nothing repeats. It works on every book, old or new, with nothing to set up. If you captured a noise print of your room, it is used when it matches the room your lines were recorded in. Exports and the ACX file use the same pauses you hear.

Everything in 0.46.1 is in this build too.

This installer is signed.

Shift-click on the Restoration and batch processing list. Click a line, then Shift-click another to select every line between them. Shift works on the checkboxes too, and before you have clicked anything it starts from the clip shown. Shift no longer highlights the words on the page.

A Select menu. Next to Select all, Select... picks one kind of line: lines with changes, lines with no changes yet, lines that need attention, recorded lines, generated lines, or one speaker's lines. Each shows how many there are.

Beds come off the timeline. Removing a bed or its music could fail with "This cue has no audio to remove," even though the music was right there. That happened when Undo brought a piece back, or when the piece was a pasted copy. Remove now always takes the piece off, and it goes to the trash like any bed. You can also choose a bed on the timeline and press Delete; it asks first.

Everything in 0.46.0 is in this build too.

This installer is signed.

Pauses sound like your room. The pause between two lines is now your book's own room recording (your captured noise print), set to the level at the end of the line before and the start of the line after. The export lays it in every pause, and you hear the same thing when you listen, so the sound runs on through a pause instead of dropping out. The playhead keeps moving through the pause instead of jumping to the next line.

Lines load just before their turn. While you listen, StellAurator times how long a line takes to load and starts loading the next one that long, plus two seconds, before it is reached. A line is let go as soon as it has played, or when you jump somewhere else, so listening only holds what it needs.

Chapters as a whole. Right-click a chapter in the list, or use its ... button, to rename it, add a chapter after it, move it, or delete it. You can also drag chapters into a new order, and the chapter numbers follow everywhere. Delete asks first and says what goes with the chapter, then sends it to the trash as one item; Put it back returns it to its place. + Add a chapter takes a .docx, a .txt, or pasted text and splits it into lines the way a manuscript import does, and tells you what it will make before you add it. Each change is one step of Undo.

Faster on big books. Switching chapters, jumping to a line in another chapter, editing, and corrections answer much faster on a large book, and Play starts in about a tenth of a second.

Uses the power of your computer. Settings > Performance shows what StellAurator found on your computer: how much memory it may use, how many cores, and whether the built-in voices use the graphics card. Each is set to Automatic, and you can change it.

Everything in 0.45.0 is in this build too.

This installer is signed.

Restoration is now a chain under the clip. The right side of the Restoration page is the toolbox only: presets at the top, a search box, and the tools in groups (Audio Cleanup, Dynamics & Volume, Tone & Effects, Recording & Editing, Mastering & Compliance, Export & Delivery), each with a one-line description. Click a tool to add it to the chain under the clip, or drag it in between steps. Each step opens right there with its settings.

Bake. When the chain sounds right, press Bake. The chain is kept as a green, read-only stage on the clip, and a new empty chain starts on top of it. No extra audio file is written. You can start again from any earlier stage, and later stages are set aside, not lost. File as a new take is still there when you want a real file.

Detect steps and notes. Fixes that work on marked sounds (breaths, clicks, plosives, esses, pauses) get a Detect step above them automatically, and Detect floor measures the noise floor. A step that is not likely to change anything says so on its row, and after Bake each stage says what did not work and what to do about it. Nothing is ever refused.

Analyze. One Analyze button runs every detection on a clip at once: breaths, plosives, esses, clicks, background sounds and the floor. It is on every waveform in the Finish pages and can also be a step in the chain.

Several clips at once. Select several lines and they appear together as a group, each with its own Play. A large group becomes pages of small clips. The chain works the same way and each clip gets its own stage.

Before and After match loudness again, so a louder version does not sound better just because it is louder.

Performers send takes back through your Nebula. A performer you invite can send finished takes with one press, and nothing is imported until you approve it, one take at a time or all at once. Each take shows who recorded it and when, in that performer's own color. The invitation is a short 8-character code you can copy and send any way you like, and each assignment has one Send the part button. The collaboration screens were redone. Sending through the cloud needs an update to the website, which is coming; the other ways of sending keep working.

Smoother listening. Proof listen and Listen now keep one output open to your headphones or speakers for the whole session, instead of opening it again for every line, which on some audio setups caused a wait of a few seconds between lines. Listening also keeps time when the window is covered or minimized.

Smaller fixes. The spotting notes no longer show a close X. A book in Nebula now carries the noise prints and learned files its processing needs, so processed takes play on another computer.

Everything in 0.44.0 is in this build too.

This installer is signed.

Nebula has its own tab. Settings now has a Nebula tab. When Nebula is off, it is the sign-up page. When it is on, it shows your plan, Cancel, and your whole library as a tree: every book, and inside it the manuscript, the recordings and the sound effects, with a check on whatever is in Nebula. Check what you want kept there, down to one take, and press Update Nebula; nothing moves before that. Rules choose what a new book sends. A take in the delete queue is never uploaded. A book taken down on another computer marks every take that stayed behind as Not on this computer.

Share with named people. Type a person's name, and the studio makes a private link for that person with exactly the files you checked. Stopping one person's access leaves everyone else's links working.

A chapter's own musical identity. The Musical identity window has Identity for at the top: the whole book, then every chapter, with a check on each chapter that has its own. A chapter can use the book's identity, its own only (Chapter only), or the book's plus its own (Chapter modified). A chapter's own words start empty.

Sound stack. A sound-effect cue now has Sound stack: layer up to 8 sounds, each with its own start, trim, level, fades and mute, then press Use this stack to make the mix the cue's sound. The layers stay so you can change the stack later.

The voice waits for the sound effect. A sound effect placed under the voice is measured for the moment it goes quiet, and the line starts there. The row shows the wait, and you can change it or measure again.

Restoration tools with their professional names. The tool list is now eight groups in the order an engineer works a voice: Repair, EQ and Effects, Dynamics, Loudness, Editing, Voice, Chain and Export. Each tool carries the name used in professional audio software, with the plain explanation in its tooltip. Tools that did the same job are now modes of one tool.

A smooth timeline. The timeline's lanes are now drawn once and kept as pictures, and only the part that changed is drawn again. Scrolling, hovering over a line, dragging a music bed and moving the Score page's timeline no longer redraw the whole timeline, so moving around a long book stays smooth.

One Normalize tool. Choose what you are normalizing to (Loudness, RMS, Peak, True peak, or Match the lines around it) and the target, then the ceiling and what happens in the pauses. Presets: ACX, ACX limits only, Podcast, Peak and Match a pickup. Pressing Normalize again on a clip replaces its earlier Normalize instead of adding a second one. Changes you made with the old tools still open, play and edit.

Everything in 0.43.6 is in this build too.

This installer is signed.

Speech and floor shows every pause. On the Restoration page, Speech and floor now colors the quiet inside a phrase as floor too: the short pauses between phrases and a quiet hum no longer draw as speech. The colors are worked out again for every clip the first time you open it.

Sending a part directly on your network. A part sent "Directly, on this network" could reach your own copy of StellAurator instead of the performer's, and was refused with "That is not the kind of package this copy is waiting for." It now goes to the performer's copy, and a refusal says plainly what was expected.

A guest performer needs no account. Recording a part someone sent you no longer asks you to sign in. The sign-in window's Not now button is now No thank you.

Put a tool away without applying it. On the Restoration page, a chosen tool now has a close button (Close this tool). Pressing the tool's button again, or Esc, does the same: nothing is applied and the preview stops.

Every step shows its values. Each step of a processing chain, and each step of a chain change on a clip, now shows its own settings under its name.

Everything in 0.43.5 is in this build too.

This installer is signed.

The processing chain. On the Restoration page, the list of tools you build up and apply as one change is now called the processing chain: "Add to processing chain", "Processing chain · 3 steps", "Save the processing chain as a preset". Finish's Custom row reads "Or take the repairs from a saved processing chain" and "Make a processing chain in Restoration". Auto sequence calls its step a processing chain too. Your saved presets keep their names, and nothing about how they work has changed.

Musical identity, up front. On the Score page, the Musical identity button now sits at the top of the spotting notes, big and easy to find, next to Make with.

Everything in 0.43.4 is in this build too.

This installer is signed.

See what is in each clip. In Restoration and batch processing, a new switch just above the waveform reads Plain, Speech and floor, and Every sound. Speech and floor shows the words in one color and the pauses and room in another. Every sound also shows breaths, mouth clicks, pops, static, background noise and background sounds, each in its own color, with hum as a strip along the bottom. A key under the waveform names each color and how many were found, for example "3 breaths, 1 pop". Sound effect takes have their own color. The colors work in every look, and the color-blind option uses a color-blind safe set.

Noise control. The first tool in the list. Pick a target noise floor, a target loudness and a target peak level, or press ACX (Audible) or Podcast, then press Apply. For each clip, Noise control measures it, runs only the fixes it needs (background noise, the pauses, the volume, the peaks), and measures again until all three targets are met. If a target cannot be reached, the line says which one and why. Each clip gets one change, "Noise control · ACX", that you can edit, switch off or remove.

Loudness keeps your performance. By default Noise control sets the loudness of the selected lines in each chapter together, so a whisper stays quieter than a shout. To bring every clip to the target on its own, open Advanced settings.

Everything in 0.43.3 is in this build too.

This installer is signed.

Restoration and batch processing, rebuilt. The page now works left to right: the lines, the clip shown in the middle, then the tool, the toolbox and Apply to on the right.

Apply changes what you see. One button says what will happen and to how many clips, for example "Apply Reduce background noise to 3 clips". Under it, the page lists the lines it will change. On each line it changes the take shown on that line. Clicking a line inside a selection keeps the selection. Play only plays. A filter takes hidden lines out of the selection and says so.

Tool, then toolbox. The everyday tools are Remove clicks, Soften pops, Remove hum, Reduce background noise, Breaths, Soften "s" sounds, Tone, Even out the volume and Match loudness. Each has Light, Medium and Strong, a few plain sliders, and Advanced settings with every setting in plain words. Everything else is under More tools. Add to toolbox puts a tool with its settings in the toolbox. Apply runs the whole toolbox. Save as preset keeps a toolbox for the book or for one character.

Changes on this clip. Every change on a clip is in one list with Edit, Off and Remove. Edit opens the tool or the toolbox with that clip's own settings, and Apply replaces the change. Before and After play the clip as it sounds now and with the tool.

The timeline shows where you are looking. The timeline outlines the lines in view in the list. It also marks the selected lines, the clip shown and the lines that have changes. Click a line on the timeline to show it, or drag along the timeline to select lines.

Also on the page. Undo and Redo, Show me for the page and each tool, Keys, Search and the filters Has changes, Needs attention, Recorded and Generated. A bar shows what a run is doing and how many clips it changed. The page remembers your tool and toolbox for each book.

The vocal chain. The settings that apply when a recording is saved are no longer on this page. Open them with the vocal chain button at the top of the window.

Everything in 0.43.2 is in this build too.

This installer is signed.

Add a cue without copying anything. In Score, Add a cue now marks where the cue goes: drag across the lines on the timeline, click a line, or Shift+click to extend what you marked. In the manuscript, select lines and press Add a cue for these lines. The cue stops on the last word of what you marked.

See where the sound library's files are. The Sound library has Open the folder at the top and Show in folder on every piece. The desktop app opens File Explorer with the file selected.

Faster on long books. On a test book of 6,000 lines, one edit now redraws about 6 rows instead of 243, and the chapter's progress arrives in 19 ms instead of 376.

Lighter while you listen. While a chapter plays, the timeline's playhead moves on a layer of its own, so the window does about a sixth of the script work it did. The spectrum stops as soon as playback pauses, and a hidden window stops its background shimmer.

Smoother windows. Typing in the character sheet redraws the field you are typing in, not the whole sheet, and the Character Tags board under the sheet stops redrawing while the sheet is open.

Less to load at start. Screens other than the library and the manuscript load the first time you open them, so the app loads about 40 percent less code when it starts. The window now appears while the engine is still starting.

Everything in 0.43.1 is in this build too.

This installer is signed.

The music that would not pause. A music bed whose start was still waiting on the output device could begin after a pause had stopped it, and then nothing could stop it. A bed now starts only while it is still the one the player holds.

Closing the Sound library stops its sound. A piece playing from the library stops when the window closes. A line or a listen playing elsewhere is left alone.

Attributes. The + add attribute button opens a small window with the same Fill from list as the header and an Add blank button. Picking a system when rows are already on the sheet asks what they become: keep them and add the rest, keep only the filled rows and add the rest, or replace all of them. With no rows, the names come in at once.

Heights and weights. The suggestion lists now run every unit: centimeters from 90 to 260, feet and inches from 3'0" to 8'11", pounds from 40 to 550, kilograms, half stones, and meters to the centimeter. You can still type anything.

Everything in 0.43.0 is in this build too.

This installer is signed.

The character sheet is a character sheet. Each character's sheet now has two columns. On the left: Look (hair, eyes, skin, build, age, height and weight, each with a suggestion list that never limits what you type, so "tall for a dwarf" is fine), Attributes (your own rows, or Fill from a well-known set such as the six D&D abilities), and five personality systems behind checkmarks: Myers-Briggs, the Four Temperaments, DISC, the Big Five and the Enneagram. Tick a system to open its panel with sliders, a graphic, a one-line description and a link to its source; untick it and the values are kept. The main column keeps Voice, Gender, Pronouns, Description and Notes, adds a Background section (origin, family, occupation, affiliations, goals, fears, story), and folds the parts the book fills in (the counts, How you last played them, Where they are in the story) behind one In this book button. The Character Tags board has a Character sheet button beside each character.

A codex for the book. Upload your series bible (.txt, .md or .docx, up to 5 MB) from a character sheet, from the Character Tags board, or when you start a new audiobook. The codex panel lists every function that reads the book for meaning with two checkmarks, Manuscript and Codex, so you choose what each one is sent. The same two checkmarks sit at the top of each function's Open prompt window. Like the manuscript, the codex goes only to the services you chose keys for.

Find a face in the cast image picker. The category bar from the Character Tags board (gender, age, hair color, species and the rest) now sits in the image picker a character's sheet opens, and a filter there looks across every portrait set. Both bars have a Tag type choice: Portraits or Tags.

Approve all above a certainty. After detection, the confirm bar has a slider (70 to 99 percent) and a button that approves every likely line the reading was at least that sure of, in one press, with one undo step per chapter.

The teleprompter's Stop stops. Stop holds the take; the tape tile saves it and Cancel discards it. Nothing is filed until you choose.

The Chronomatic timeline. A jump's height now grows with its width, so parallel jumps no longer meet at the top. The thread changes color at each turn (red through violet, then around again) and two crossing lines never share a color. Scenes are small cards with the chapter number, the scene number and the title. Dragging a card lifts it and the other cards slide out of the way of the slot under your pointer; the slots stay put until you drop.

Everything in 0.42.6 is in this build too.

This installer is signed.

A driver that will not start is tried again in a fresh helper. RØDE's NT1 5th Gen drivers could refuse to start ("ASIO Driver open Failure!") after the other NT1 entry had been used in the same session, with nothing else holding the microphone. The app now restarts its audio helper and asks once more on its own, and says so in the Recording panel. A driver that refuses a second time is reported as before.

The update bar sits under the tab buttons. While a book is open, the "downloaded, installs when you close the app" line sits in the band under the tab buttons and covers nothing. On the library screen it stays as a bar across the top.

Chapter headings. Importing a manuscript now splits on Markdown headings (the # marks are never part of a chapter title), on number words alone ("One", "Two"), and on headings with emphasis around them. A sentence that happens to begin with "Part" or "Book" is no longer taken for a heading.

Everything in 0.42.5 is in this build too.

This installer is signed.

Signing in works again. Since 0.42.0 the sign-in form could stop the studio with "acct is not defined": when Settings opened on the Account tab before the account had been read, when you were signed out, and on a fresh install's first sign-in. The form is fixed, and a test now draws it signed out so it cannot come back.

Everything in 0.42.4 is in this build too.

This installer is signed.

Clip gain is one press. In the waveform editor, select a stretch, set the gain, press Apply. The change is kept on the take at once, as one of its changes, with a red line beside the tile that says what was applied. Nothing is saved as a new take. Every clip gain shows as a band on the waveform: click its label to select exactly that stretch, then Delete, Reset, or Remove this clip gain to take it off. The edits panel has a Remove beside every change too.

The editor's tools stay usable after fixes are applied. When a fix landed on a take, the editor used to clear its selection and go grey. It now keeps the selection, the cuts and the undo.

A crash in the studio is fixed. The track beside the clip pool could lock up the window on some screens ("maximum update depth exceeded"). It no longer can.

Switching between ASIO drivers works. Choosing a second ASIO driver while the first was open for monitoring used to fail with "another ASIO driver is open in this helper", with nothing to do about it. The app now closes the open driver and reads the new one itself, and monitoring follows. If the sentence ever shows again, the Recording panel has a tile for it: Close the open driver and read this one. A driver that reports no input channels now says so instead of "Unknown capture device".

The microphone list behaves. It draws above everything and follows its button, a helper that did not answer is asked again, and one missed poll no longer empties the list.

The WPM readout opens its settings. Click the WPM numbers on the top bar to open Settings at the pace section.

First run. On a fresh install you can look around and sign in before the license agreement is shown.

StellAurator for Mac and phones is on the way. A line in Settings says so, with a Get notified link.

This installer is signed.

The license agreement now names the usage reports. Version 1.5 of the agreement says which reports the app sends on its own: the alpha usage reports, for alpha accounts while the alpha runs, and the usage reports you choose to turn on in Settings > Privacy. It also says that books you put on Nebula are uploaded sealed, so we cannot open them. Nothing the app sends has changed. You will be asked to accept the agreement again the next time you open the app.

This installer is signed.

No more blank window when a screen hits an error. If Proof listen or another screen runs into a problem while it is drawing, the window now says which screen stopped, shows the error, and offers Close and Reload instead of going blank. The error is also saved in the engine log, so a bug report from Settings carries it.

This installer is signed.

Guides on the website. The Help menu in the library has a new entry, Guides on the website. It lists the buying and how-to guides on stellaurator.com: choosing a microphone, audio interfaces, headphones, Audible's ACX requirements, and treating a room. Each guide opens in your own browser.

Two shortcuts to the guides. In Settings, Audio, a line under the microphone chooser opens the microphone guide. The noise print panel ends with a line that opens the room treatment guide.

This installer is signed.

Ready for the One-Book Pass. This release carries the machinery for a one-time purchase that opens the whole studio for one book, for good: record, edit, finish and save that book with no deadline, while other books stay read-only. Nothing changes in the app until the pass goes on sale on the account site. When it does, the Subscribe buttons name its price, a pass can be put on a book from the book's menu in the library (it stays on that book for good), the book's row wears a One-Book Pass tag, and the account panel says which book the pass is on.

Small fixes. The read-only wording and every existing plan are unchanged.

This installer is signed.

Problems counted while you record. The recording bar now counts clicks, plosives, harsh S sounds, breaths, hum, background sounds and the room's noise level as you record, with a mark on the strip where each was heard. It is a running estimate that costs the computer almost nothing, it never changes the recording, and it pauses itself if the computer is busy. Settings, Recording, Count problems while recording turns it off.

A report the moment you stop. The take shows a report: what was found, how loud it is, the room's noise, and a verdict in plain words. One button, Apply recommended fixes, applies the whole suggested set as layers you can take back with Undo. Review still opens the before-and-after window.

Fixes applied automatically, if you want. Settings, Recording, Apply the recommended fixes automatically after each recording applies them the moment the check finishes and says so on the take.

Outside services, explained once. On the first run, after the welcome questions, a screen lists every outside service the studio can connect: what each one adds, next to what the studio does for that job without it, who bills you, a button to the service's own key page, and a button that opens Settings on the exact field for the key. None of them is required. Open it any time from Settings, Services, or from Help.

This installer is signed.

Every clip says what was done to it. A small icon now sits on every place a clip appears: the take box in the clip pool, the waveform editor, the clip editor, Restoration and Comp takes, Proof listen's line, and the timeline's blocks and track pieces. It is grey when nothing has been done to the clip and lit when the clip has been edited or repaired. Hover it to read the list, such as de-clicked, de-essed, trimmed. Click it to open the panel.

The panel. Each edit or fix is a chip with when it was made and by what: by hand, by the suggested repairs, or by Finish. Click a repair's chip to switch that one repair off or on. The Original and Processed switch is there too. Show tag locations draws where each fix landed on the open waveform or clip view, in the fix's color. Edit in editor takes you back to where the edit was made: Restoration with that repair chosen, or the clip editor.

The same words everywhere. The Masks row on a take, the Original and Processed switch, and the Finish results all use the same words as the tags.

This installer is signed.

Playback waits for nothing. While a line plays, the next four lines are made ready one after another, so a line whose repairs still had to be rendered is ready before its turn. The listen in the manuscript and Proof listen both do this. A line made ready once is not made again.

Proof listen moved. The Proof listen button now sits between Record a part and Score at the top.

A silence can start a sound. A silence row has two new buttons, Sound effect here and Music here. Each opens the sound library, and the piece you choose starts at that silence. The silence stays.

Editing stays quick. The slowdown that built up as you edited is measured and fixed: lines redraw only when they change, the quiet color drift no longer restyles the whole page, every play button shares one clock, and suggested repairs are found on a separate thread so the app keeps answering. The Running jobs box now shows what the app is holding in memory, with a Free memory button.

The owner's Console left the app. It was never shown to authors. Nothing changes for you.

This installer is signed.

Tracks beside the clip pool. Every track your timeline has, Sound effects, Music and any numbered track, gets its own narrow column beside the clip pool and its own check mark directly above it in the clip pool bar. Tick any number of them. A track shows in its column with its name written along the top; click it to open it wide. Drag a take onto a column to make a sound effect at that moment, drag a piece up or down to move when it plays, or drag it back onto a line to make it a take again. The old dropdown is gone.

Sound effects and music from the line. Insert spot effect and Add background track now open the sound library instead of a file picker: Sound effects with Place after this line, Music with Start here. Importing from your computer lives inside the library, with Import files and Import a folder, and a file you import while placing is offered for that line right away. The library is sorted into tabs: Sound effects, Music, Ambience and beds, Stings and buttons, Other.

The timeline's tracks. Each track has a header with its name and two buttons: M mutes it, S isolates it so only the isolated tracks play while you listen. One track is current, highlighted along its length and named on the playhead; a click on a header or an empty spot chooses it, and the Up and Down keys move it. Insert at playhead puts a piece on the current track. Right-click an empty spot for Insert here, Put the playhead here, mute, isolate, Add a track and Rename; right-click a piece for Play, Edit in the clip editor, Copy and Remove.

A muted track and the export. A muted track stays out of the export. When one is muted, the Export dialog shows a red block, "A track is muted. Unmute the track or confirm export.", names the tracks and chapters, and offers Unmute and export or Export without the muted tracks.

Memory. The window and the engine now run with a 4 GB memory ceiling. The slowdown that builds up while editing is measured and fixed in the next release.

This installer is signed.

The noise print panel. Every recording area now has a small panel beside the level meter. It shows the room the engine believes in, the saved noise print, and the room the microphone hears right now, the live print, both drawn over the frequency display. When the two drift apart it tells you. Press Capture a new noise print: it counts down from three, records the room alone for ten seconds while asking you to stay quiet, and keeps the result as the book's print. The automatic pass and Finish clean with your print instead of guessing the room from each take, and a take whose room no longer matches falls back to its own pauses and says so.

Record a part is the stage. The second stage button says Record a part and opens the part screen. Its row keeps Whole cast, Continue and Pickups.

End breaths. A quiet breath at the start or end of a line that still stands above the room is now treated as a breath and made quieter. Before this, anything more than 40 dB under the voice was left alone as faint, even when it was plainly audible.

Mark a problem. On any take's waveform, in Comp takes, the clip editor or Restoration, and while a chapter plays in Proof listen, press M or Mark a problem to mark where something is wrong and what it is: a missed breath, a click, a harsh ess, mouth noise, a cut word, a gap, a level jump, or sound the waveform hides. Under each mark the engine writes what its own detectors saw at that spot. Problems marked lists them, and Export the teaching set writes the marked moments and the engine's readings to a folder, so the detectors can be checked against your ear.

This installer is signed.

Who says this line. When your manuscript comes in, the studio now works out who speaks each line on its own, by looking at who is in the scene and how the conversation goes back and forth. The lines it is not sure about are ranked, with the most likely name first, and Review in order walks you through them one by one. Confirm likely lines lets you accept the probable ones in a batch.

The stage strip. Four plain buttons: Manuscript, Record, Score, Finish. Record offers Record a part, Whole cast, Continue and Pickups. A Proof listen button sits at the right end on every stage.

Gaps between lines. The gap after each line shows as a small number between the lines; click it to set the pause by hand, or press Back to the plan. With Gaps follow the pace on, a passage read fast gets shorter gaps and a slow one longer gaps.

Sound effects on the timeline. Click a block to move the playhead. Drag a clip down to the Sound effects lane, or press Make this a sound effect on a line, and choose Over the lines, with a quiet space. The clip editor can copy and paste a stretch between layers. A sound track can now run down beside the clip pool: tick the check mark by the Clip pool heading, drag a take onto it, drag a piece up or down to move when it plays, or drag it back onto a line to make it a take again.

Breaths and esses. The breath cleanup now finds the breaths in the quiet spots between your phrases and at both ends of a line, including breaths that run straight into the next word. The suggested repairs after a recording, and the Finish pass, soften harsh S sounds when a take has them.

Finish and read-throughs. A read-through is now processed whole, as one, instead of being left out of the Finish pass.

Nothing stops when you leave a window. Leaving a window or screen while something is still working asks Continue in the background or Cancel. A pill at the top shows each job still running, and the result is waiting when you open the window again. Closing the app asks you to wait or cancel.

Listen to Character. On Record a part, each character chip has a Listen to Character button: hear their earliest lines, their most recent lines, or one chapter, with the sound shown by frequency as it plays. Record a character sample keeps a short example of how they sound.

The frequency display also shows beside the meter while you record, over Before and After in Restoration (click it to set the De-esser's, Hum's or EQ's frequency), and as a By frequency toggle on the player.

The Chronomatic timeline. A long character no longer fails to build: the reading is split into parts as needed. Place the scenes by hand lets you drag a character's scenes into the order they happen, mark flashbacks, and keep those placements through a rebuild.

This installer is signed.

Musical identity. On the Score screen, the spotting notes gain a Musical identity button. It opens a window where you write the music of the book in your own words, click tags into it (low droning dramatic, orchestral adventurous, synth pulse, solo piano, and more), add references to films, series and games whose music is close to what you want, and press Describe to have your text model lay out each one's music in musical terms, which you can add to the identity. Suggest references names works that would suit this book from its plot and genre. The identity then guides every cue's music prompt loosely (beds most, stings lighter, sound effects not at all) and Find the beats considers it. A Follow the identity slider on the Score bar sets how closely each chapter keeps to it, from loosely to tightly. The music service only ever receives musical terms, never a title or a composer's name.

This installer is signed.

Nothing of ours silences anyone. Monitoring now opens your headphones in shared mode, so every other program keeps its sound while you hear yourself. The shortest path is still there: tick Reduced delay beside the Monitoring slider; its hover tip says what it costs: it disables shared output from other apps. With Reduced delay on, the app plays your clips itself through the headphones' own stream, so playback never goes quiet because the monitor is on.

Pin what you use. With a line's More list open, hold Alt: every item shows a box; Alt + click pins it to the tools row on every line. The More button is the same menu icon as the clips'.

The Auto-process gear runs your sequence. Its ▾ opens the sequence: file as ALT 2, voice-change, lock by default; add any process from the take steps, the Clean, Tone, Dynamics and Level tools or your chains, order them, switch them off, reset. Kept per book.

Review suggested repairs has two play buttons, Original and Repaired, with the themed play icon and the one sounding lit; a playhead that moves and seeks on click; "Preparing the preview..." until both sides are ready; a legend for the marks and a tip on each.

The disclosure says what is true. "Contains AI-generated narration" only for synthetic speech. A reading you recorded and voice-changed is tagged "Contains Speech to Speech modified human performance".

Finish results show each take's peak and floor and whether it is leveled, normalized or as recorded.

Gaps at the top of Proof listen. The book's gaps between lines (clause, sentence, paragraph, separator, heading) are edited from the Proof listen header while you listen; the next gap uses the new value, and the export writes the same.

Permission documents. Attach the email in which a platform authorized something (a file or pasted text), tick "Applies to this book", and the ACX caution, the export summary, the export folder and its manifest all carry it. The files keep their tag.

Restoration: the tool opens under the clip. A line's row shows one expanded clip and every other take as a mini card (click one to swap); Lock works on the mini cards and on the expanded clip. Choose a tool on the right and it opens under the clip in its own row: the clip's view, Before | After, Compare, Done so far, Settings and Keep, with Done. The top of the row never changes, only one row is open at a time, moving to another clip moves the tool there, and Esc or Done folds it. The right side keeps the tool tiles, the clip's numbers and history, and the Results. A folded toolbox shows a Tools tab.

Background sounds. A new Clean tool finds sounds that are not you: a sound in a pause, a higher voice under the narration, a bright sound under the narration. It lists them with Play, Flag for a pickup and Not a problem, the Review window draws them, and Finish counts them. Nothing pretends to remove another voice from under yours: flag the line and record it again.

In doubt, the AI flags the line. When dialogue attribution cannot settle who speaks a line, it no longer picks someone. The line gets a Who says this? chip with the names it weighed; one click settles it. The Cast tab counts the lines to review, the jump menu has Next line to review, and the progress line says how many speakers are left to review.

The Input gain slider follows your hand. It no longer jumps while you drag it.

Score stays in its chapter. The Score bar has the chapter picker, playback stops at the chapter's end instead of running into the next one, and Add space puts empty seconds after a line for music with no dialogue: the timeline draws it, the spotting notes list it, Proof listen and the export honor it.

This installer is signed.

Finish says what it is doing. A chapter or the whole book sits in its own card above Diagnose with its stats, so the subject is never in doubt. While Diagnose, Apply or Check runs, the step expands with a live list of every line as it is checked, repaired or measured, with a running tally; when it ends the list stays behind "What was checked". Leave the screen while a job runs and come back: the progress and Stop are still there. Custom lets you custom: tick the repairs you want, Force into ACX compliance, and the export target, or take the repairs from a chain, and the choice is remembered per book. Every step has its own Show me tour; the 5-minute walkthrough is untouched. The Finish stage's steps now run in working order: Finish, Restoration and batch processing, Proof listen, Pickups, Export, because pickups are made in Proof listen.

Proof listen chooses its chapter in its own row, with the chapters before and after. The sidebar draws Chapters only on Finish and on Restoration and batch processing; Restoration's own chips name the characters, and a voice is changed in Write & Cast. The manuscript text size opens at the slider's middle.

The Score timeline floats over the sheet a quarter up from the bottom. Drag its grip up or down; it stays the page's full width, and the place you leave it is remembered.

The Teleprompter is a real window. A title strip you drag, edges you resize, a normal X at the top right, and Escape closes it, with the frame remembered. Slide mode centers the start line when the session starts. The window stays open while the clips are cut after a restart. Recorded bars beside the text show the span of every continuous recording, each later recording in its own column, so where two overlap you see it.

This installer is signed.

The Monitoring slider always does what it says. On the Native (studio) engine the slider in the top bar is the one switch for hearing yourself: set it to anything but Off and Low-latency monitoring turns on with it. The app now watches the engine's own state and turns the monitoring back on by itself whenever the microphone was closed or opened again behind it (a format change, a take, the helper starting again). When it cannot start, the pill says so: "MONITORING · NOT ON" in red, the reason as the first sentence of its tip ("The microphone is in use by another program: RØDE Connect. Close it, or choose another microphone in the Recording panel."), and one message the first time. An app that started before the native helper answered no longer stays on the fallback engine for the whole session: it asks again and switches to Native (studio) the moment the helper is there.

The Recording panel never swallows a click. A setting pressed with no microphone chosen says "Choose a microphone first." instead of doing nothing; a setting being saved shows it; a saved "Keep original captures (WAV)" shows its check mark at once and stays checked after the engine answers.

This installer is signed.

Spools. What the app called a version column is now a spool. Takes you have recorded or generated wait in a line's clip pool; the one chosen for each line is wound onto a spool, and a spool is one complete reading of the book that you can switch to or export whole. Spool A, spool B and as many as you add work exactly as the columns did: the headings, the Add spool button, the lock sentences, the Export dialog's sources and the tour all say spool now. Nothing you saved changes.

The Finish check says what it assumed. The ACX check judges a chapter as the export will write it, which includes the export's own leveling. A quiet chapter now reads "ACX: passes after the export's +17.2 dB" instead of a bare "passes" beside its recorded numbers. When your export target is Custom with Loudness off, the export writes the chapter as recorded, so the check judges it as recorded and says so, and the level sentence tells you what to turn on.

This installer is signed.

Finish: one preset, five steps. The Finish stage now opens on a Finish screen. Choose the scope (This chapter | Whole book) and a preset: Clean narration, Audiobook master (ACX), Podcast or Custom (your own chain). Then the five steps fill in as they run: 1 Diagnose checks every take and measures the chapter ("Chapter 3 · 42 lines: 12 clicks · 3 plosives · 9 breaths · 2 noisy takes · −19.2 LUFS · peak −2.1 dBTP · floor −58 dB · ACX: 2 problems"); 2 Plan says what Apply will do; 3 Apply does it, every repair a step on the take with its own Undo, and for the ACX preset forces the chapter into compliance; 4 Check gives the ACX verdict per chapter, each problem naming the tool that fixes it; 5 Export opens Export on the preset's target. Results below list every take it touched, with Review before and Open in Restoration after.

Your own key and mouse clicks are erased from recordings. While a take records, every key press and mouse click in the app is stamped on the recording's clock. When you stop, the last quarter second and the moment around each stamp are checked for a click: one after your last word (the stop key) or before your first is trimmed off; one inside your words is mended in place so the word keeps its timing. The recorder shows a tick at each stamp as you record, and the take's details say "2 key clicks erased · 1 mended". Settings → Recording: "Erase my own key and mouse clicks while recording", with "Mend them inside the words" or "Cut them out".

This installer is signed.

Stages instead of modes. The strip at the top now names where your book is: Write & Cast → Record → Score → Finish, with the stage's steps on a lighter row beneath (Record: Studio · Record a part · Teleprompter; Finish: Restoration · Pickups · Proof listen · Export). Every screen is the one you know; only the way in changed. A progress line says how far along the book is: "3 of 12 chapters recorded · 2 chapters finished · 4 pickups open." In Write & Cast the line's structure tools come first; in Record, the everyday ones.

Suggested repairs after every recording. When a take lands, StellAurator checks it with the Clean narration preset (clicks and pops, plosives, breaths quieted, light noise reduction when the room is loud; never a level change) and the take box says what it found: "Suggested: 2 clicks · 3 breaths · noise" with Review. Review plays the repaired audio immediately, Original | Repaired to switch, every repair marked on the waveform. If something was changed that you meant to keep, drag over that stretch and press Keep breath, Keep plosive or Keep 2 clicks: that spot stays as you recorded it and the preview plays again. Apply keeps the repairs as steps on the take, each with its own Undo. Settings → Recording holds the switch and the preset.

Takes lock themselves. A new recording becomes the line's locked take unless you locked one yourself, in which case the app asks once: "Lock the new take instead? You locked v2 yourself." A voice-converted take locks itself into its own column. Switch it off in Settings → Recording ("Lock new takes automatically").

This installer is signed.

Restoration has one work surface. The list on the left is for choosing lines; the panel on the right is for doing. Play on a card only plays. Edit this clip opens the waveform editor inside the panel, in the inspector's place, with Done to come back. Every tool says what it works on: the strip has a row "For the selected lines" and a row "For the clip in the inspector", each tile carries "N lines" or "1 clip", and the line above the tool reads "Applies to: 16 lines selected" or "Works on one clip: Line 2 · Narrator". Spectral repair opens the inspector large with the spectrogram up: "Drag a box over the sound you want gone."

Before | After, then Keep. Each tool shows the clip as it sounds now (every kept step included) over what the tool would do, each with its own play tile and readings, and the change in words. The one accent button is Keep ("Keep for this clip", "Keep for 16 lines"). After Keep, the After moves up to become the Before, and the step appears under Done so far: the clip's kept steps in order, each with a switch and Undo.

Tags on every clip card. Each kept step puts a small tag on the clip's card ("v3 · Clicks · Noise · Gain"). Press a tag to switch the step off or remove it; the clip plays and draws without it at once, one undo step.

One step per tool per clip. Keeping a tool that is already on the clip changes that step in its place instead of stacking another: level changes add up (+4 then −3 is one step at +1), time stretch multiplies, spectral repair adds its boxes, everything else takes the new settings. The confirmation says so before you press Keep ("Gain is already on this clip (+4): this Keep changes it to +1, in its place."), and the Before | After split always plays the real result.

Moving a line. The grip is on the top-left corner of the line itself. While you hold it, a card of the line follows the cursor, the line you took it from dims, and the gold line shows where it will land.

This installer is signed.

Character menus you can find your way around. The menu on a line's character chip lists the cast with the Narrator first and then A to Z, whatever order the book keeps them in. With eight characters or more it opens with a search box: type part of a name, and Enter picks the first character left. The same order drives the selection toolbar's character dropdown and Record a part's "Add or change a part…" list and chooser.

Record a part can make a character. "+ New character…" in the "Add or change a part…" list and on the chooser asks for a name, adds the character to the cast, and reads for them.

This installer is signed.

The Restoration list selects like Windows Explorer. A click on a clip card selects only that line. Ctrl + click adds or removes a line and leaves the rest. Shift + click selects every line from the last one you clicked to this one. Ctrl + A selects every line listed; Esc clears the selection; the arrow keys move through the list (Shift extends, Ctrl moves the focus only). The whole card is the click target. A checkbox changes only when you press it or press Select all. Play, the clip gain fader, Go to the line and Generate again never select.

Smaller cards. The clip cards are Compact by default, so more lines fit on the page; a Card size switch at the top of the list (Compact | Comfortable) brings the roomier cards back, remembered on this computer.

This installer is signed.

An easier first twenty minutes. StellAurator asks one question first, "What are you here to do?", and builds your path from the answer. One guide at a time: the plan stays folded to its pill until you open it, and the walkthrough waits until you ask for it. Once you have a book, the library is your shelf: covers, progress, last opened.

The Studio reads by scope. Book, chapter, line and take controls are separated so you can tell what a button affects. The line's tools come in four groups with the everyday ones first and the rest behind More. Take cards say what their numbers mean, "166 WPM" instead of a bare number. On a narrow window the mode tabs go compact instead of wrapping.

Export starts with the target. Audible / ACX, a chaptered M4B, both, or custom. Choosing ACX sets the compliant settings and says so; every control from before lives under Advanced.

Smaller things that matter. The New Audiobook form asks only what it needs up front. Restoration says its loop in words: lines selected, choose a tool, preview, apply. The rail's "Export" is now "Export selection", so it cannot be mistaken for exporting the book. The ElevenLabs card sits after your cast and reads as what it is, an optional way to get another voice, with its disclosure.

A size for every screen, and a full window to start. StellAurator now opens maximized, like a studio app should, and remembers your window from then on (a "Start maximized" switch in Appearance). The interface picks a size preset from the window's size, compact to extra large, and you can pin your own choice per size; "Interface size: Automatic" is the default, your old slider is now "Custom size", and Ctrl+0 returns to Automatic.

The Restoration panel explains itself. Every tool group shows its tools as tools, with one sentence saying what the chosen one does. "Applies to" says where it acts. Before and after are drawn as two waveforms with the change in words. Empty results name the button to press. Clips is now Results, and the player row is Compare.

Working files beside your books. Recordings in progress, export pieces and conversions are made next to your library, not in Windows' temp folder. A recording from an interrupted session is never deleted; Recover takes found on disk offers it.

Themes. Decorative themes get a "Working surface" slider so the artwork behind the manuscript can be as quiet or as loud as you like. The icons are untouched; an audit of every pack at real size is in the docs for the owner's eye.

This installer is signed.

Native audio. StellAurator now records through your microphone's own ASIO driver, or through Windows exclusive mode, in the format the driver delivers and with the smallest buffer it allows. On the RØDE NT1 5th Gen with RØDE's driver that is real 32-bit float, proven in this release for the first time. Any maker's ASIO driver you install appears in the Recording panel, with its own control panel one press away. The old paths remain as Direct (legacy) and Browser.

Low-latency monitoring. Hear yourself through your headphones with no processing at all: no gate, no filter, no compressor, unity gain unless you turn it down, through the shortest path Windows has. The panel shows the measured round trip. A wireless headset adds its own delay that no software removes.

What the driver says goes. Sample rates, bit depths and buffer sizes come from the driver's own answers. When a driver cannot run at the buffer you ask for, the app opens at the one it can and says so under the buffer control. A take never begins with the digital silence some drivers emit on start.

When something holds the microphone. The record row names the program using it. If your antivirus asks about stella-audio.exe when you first record, press Allow, then press Refresh.

Export of a fresh book. Export now starts on "Latest-take fallback": a line with takes but nothing locked in exports its newest take, the missing-lines box lists only lines with no audio at all, "(costs)" appears only when a paid voice is involved, and a chapter with no audio yet is skipped with a note instead of stopping the export.

Tidier screens. The "Your path" plan sits in the tabs strip instead of covering Export; popovers and cue menus stay inside the window and above the bars they belong to; first run shows the interview alone; the teleprompter and the Export, Assign and Voice & pace boxes read correctly in the light themes; stale words fixed ("Settings → Recording", the ••• menu, a new Record a part tour step).

Corrections. The NT1 5th Gen has no headphone output; the app no longer says otherwise. Third-party notices now cover every component in the installer, with the full GPL text for ffmpeg, and a license scan guards every release.

This installer is signed.

Masks, not copies. A repair or process no longer makes a new copy of your take. It is a mask over the take, played instead of the original while it is on, with an Original | Processed switch on every take. Undo removes the mask. "Bake into a new version" makes a version only when you ask, and the export renders the masked sound. Masks are rendered into a cache outside your book that you can size in Settings.

Encoded once. Every export, MP3 and M4B included, is now made from a single float master and encoded one time. Generated lines arrive as PCM where your plan allows, and the voice changer is sent the original capture.

Session gain. The recorder no longer levels each take to its own peak. One session gain applies to every take you record, so your lines keep their relative level. The old behavior stays as a switch, marked not recommended.

Pauses as you read them. The silence between lines is measured by a speech detector that works in ordinary rooms, pause lengths follow Audible's standard with a heading kind, a 20 percent band leaves a pause alone unless it is well off, seams can take the pauses from your own read-throughs, and a lengthened gap is your room tone, never digital silence.

Clean, refined from the research. Click repair works in the click band only and reports only what it changed, breath quieting never presses the room below its floor, plosives are found under vowel onsets and only the excess is removed, background noise borrows the room of neighboring lines for short takes, spectral repair handles short clicks, the tools run in the restoration order with a book-wide floor, and the room tone fill never repeats.

Dynamics and Level. The de-esser cuts the range you set at the "s", the compressor's release is the same at every depth, pickups match by the median of their neighbors with a short-line measure, the readout shows the ACX Check floor beside ours, Force into ACX compliance uses true time constants, the Podcast preset replaces Spotify/Apple, EQ bells over 6 kHz match their analog shape, crossfades match the sound, time stretch fits by the pauses first, and a slow leveler is available as a tool that is never a default.

Help make StellAurator better. An opt-in on first run and in Settings → Privacy sends how you use the app: which buttons you press, which features you use, where something fails. Never what you record, upload or write, and no name, file or key. You can see what would be sent, send it now, delete what is waiting, or turn it off at any time.

This installer is signed.

Restoration and batch processing, rebuilt. The tab is now a full editor: a tool rail of Inspect, Clean, Tone, Dynamics, Level, Edit, Voices, Chains and Export, a clip list with the inspector and a spectrogram, and one grammar for every tool: preview A/B on one clip with matched loudness, then apply to the selected lines. Every run makes new versions, keeps the old ones, reports what changed on each clip, and can be put back in one press.

Clean. Background noise with a learned noise print, mouth de-click, de-pop, hum removal that learns the exact frequency, breaths inside a line, a room floor expander that never gates speech, spectral repair, and room tone matching.

Tone and Dynamics. A five-band parametric EQ with its curve, a real de-esser, character effects (telephone, mobile, radio, megaphone, far away, next room, small room, hall, cave), a compressor with a gain-reduction meter, an expander and a true-peak limiter.

Edit. Crossfades at seams, time stretch, punch-in with pre-roll, and comping from several takes in the clip editor. Chains save a set of tools and settings and run as one.

Level. Measure loudness per chapter and line with an ACX pass or fail, match a pickup to its neighbors, clip gain on every take, a read-through leveled as one, and Force into ACX compliance (Clip or Smooth). The export levels each chapter with one gain and a true-peak ceiling, with presets for ACX, Findaway and Spotify/Apple, so every line keeps its relative level.

Export. Export selected lines as one file played as the book plays them or one file per line, the ACX retail sample in one press, WAV and FLAC masters with dither, stems, and Export this line as a file beside Import on every line.

Read-throughs are played as you read them. A read-through is kept whole; nothing inside it is trimmed or re-gapped, and the pause setting applies only where lines were recorded separately. "Show trims and gaps" on the timeline marks every automatic trim in red and every added silence in blue. The cut between two parts can be moved on the waveform.

ElevenLabs models. Choose Eleven v4, v4 Turbo, v3, Multilingual v2 or Flash per book and per character. The line's direction goes to v3 and v4 as an audio tag. Compare models on a line with presets like Absolutely terrified, Romantic or Screaming.

Chronomatic timeline. A character's arc, Chronological or As written, with a Morph on every scene, as a side panel on the character sheet and a popup on Record a part. Auto Direct directs by the Morph.

Spotting in Proof listen. Drop a cue at the playhead with a key per kind or a general button while you listen, and move it after.

This book's sounds. A scan of your book finds every sound effect it could use, in its own words, and generates them with your ElevenLabs key, one at a time, with the cost named first.

Terms. Labels keep plain words; the craft's terms sit in the Ctrl tips, and a few labels took the craft's word where it reads better.

This installer is signed.

Recover takes found on disk. In Repair and process. If the old chapter Undo took recordings off their lines, this finds them on your computer and puts them back as new takes, with the original capture attached. The library clean-up never deletes them.

Timeline. A built-in Sound effects lane and a built-in Music lane, with your own tracks below. Click the ruler to move the playhead, and Add at playhead to place any sound from the library there. Beds now line up with their lines and the playhead. Click an empty bed or any placed piece, on the timeline or on the cue sheet, to open its options: generate, import, replace, move, change lane, volume, edit the arc, edit, copy, paste, save a copy.

Clip editor. Edit any take or music or sound piece in a window of its own: add channels from the library, other tracks, other takes or a file, trim, fade, mute and solo, then save it as a new edited version. Copy and paste with Ctrl+C and Ctrl+V.

Auto Direct. Reads the scene with your AI service and sets a mood on every line from what came before and who the character is at that point in the book. Shows its progress, never touches a line you directed by hand, and one Ctrl+Z takes back one line.

Surprise. A ninth emotion on the direction web, with blends like Pleasant surprise, Startled and Elation. Every mood you saved before is unchanged.

Record a part. Remembers who you are reading for in each book. An × on each character's chip and a Back to character selection button return you to choosing.

Cast bar. Wraps to two rows and scrolls past that. "On this page" shows only the characters with lines on the page, on by default.

Account. Sign out automatically after a period with no activity or when the app closes, and optionally require sign-in to open books on this computer.

Looks. The mode tabs are drawn in each look's own style with larger icons. Your flag icon marks pickups in every look.

This installer is signed.

Celestial. A new settings gear: a starry gear with its own aura.

This installer is signed.

Undo and redo. Undo and Redo are always in the chapter header, with Ctrl+Z and Ctrl+Y. Each step is one action, and 250 steps are kept each way, so an undo takes back only the last thing you did. The sparkle button is now called Generate again by rule.

Repair and process. Click a clip to select it. Every repair and voice pass runs on the selected lines, this chapter, or the whole book, and the characters in the chapter can be chosen at the top. Every line shows all of its clips, with the unassigned ones on the right as in the studio.

Pickups. Every flagged line is laid out as a full studio row with recording, repair and upload.

Score. A written sound effect in the text can be placed as a sound and replace the spoken clip. Generated and imported pieces land on the timeline, overlapping sounds take a new lane, and the timeline is visible in Score. A piece in the library can be added back with Add to timeline bed. A music piece can carry several cues on one bed, with their times measured from the start of the bed. The cue rows take the height between the controls and the timeline, and the sheet as text opens only when asked.

Proof of performance. A line replaced by a sound effect is listed as a sound effect, never as a recording. Inserted sound clips and silences are not counted as recorded lines. The AI disclosure names the music and sound providers as well as the voices.

Looks. The Plain look is one card at the top of the theme list with a Day/Night toggle. The Brightness and Contrast sliders move again. New direction, score and tape icons in every look.

This installer is signed.

0.26.0. A larger step: tabs for every mode, Repair and process, music and sound effects on the cue sheet, a score library, and the direction panel. Everything from 0.25.15 is in it.

Modes are tabs. Studio, Record a part, Proof listen, Score, Pickups and Repair and process sit as large tabs at the top of the window, on the main page and inside every mode. Record a part opens first and asks who you are reading for.

Repair and process. A list of your lines with the clips assigned to each version column, their waveforms, and any pickup flags. Clean up breathing sounds, Normalize, Merge chapter, Regenerate, Voice changer and Revoice work on the selected lines, this chapter, or the whole book.

Score. Every cue on the cue sheet has a Generate music or Generate sound effect button, and Import audio. Generated pieces go to a score library shared by all your books, where you can import folders, sort, tag, and generate from your own prompts. Suno opens in your own browser, and a download saved to your Music folder appears on the cue.

Direction. On Record a part, every line has a direction button that opens a side panel with the emotional direction control. Narrator lines offer simple choices like Excited, Fast, Slow and Interested.

Record a part. A take's silence trim now lands the moment the take does.

ElevenLabs. The ElevenLabs link shows their official logo and their trademark notice.

This installer is signed.

Modes are tabs. Studio, Record a part, Proof listen, Score, Pickups and Repair and process sit as large tabs at the top of the window, on the main page and inside every mode. Record a part opens first and asks who you are reading for.

Repair and process. A list of your lines with the clips assigned to each version column, their waveforms, and any pickup flags. Clean up breathing sounds, Normalize, Merge chapter, Regenerate, Voice changer and Revoice work on the selected lines, this chapter, or the whole book.

Score. Every cue on the cue sheet has a Generate music or Generate sound effect button, and Import audio. Generated pieces go to a score library shared by all your books, where you can import folders, sort, tag, and generate from your own prompts. Suno opens in your own browser, and a download saved to your Music folder appears on the cue.

Direction. On Record a part, every line has a direction button that opens a side panel with the emotional direction control. Narrator lines offer simple choices like Excited, Fast, Slow and Interested.

Record a part. A take's silence trim now lands the moment the take does.

ElevenLabs. The ElevenLabs link shows their official logo and their trademark notice.

This installer is signed.

Appearance. Settings has a new Appearance tab: the theme, interface size, manuscript text size, icon size, brightness, contrast, higher contrast, keyboard focus, reduced motion and reduced transparency. Accessibility choices too: a hyperlegible font, color-blind friendly status colors, status words beside colors, no flashing, line and letter spacing, dimming the lines you are not on, mono playback, listening speed, and screen-reader announcements.

Standard and Standard Night. Two plain themes in the Windows look, light and night, with Windows-style icons. Turn on "Follow Windows light and dark" to switch between a daytime theme and a nighttime theme with your Windows setting. Any two themes can be the pair.

Next unread. The button goes to the earliest line with no recording in the chapter you are in, or in the whole book; the menu beside it chooses which.

Console. The Numbers pages set aside the owner's own activity. The app reports its own update checks and downloads from an owner's account only.

This installer is signed.

Clean up breathing sounds. A new button under Normalize audio in the Chapters sidebar. It listens to the start and end of every recorded take, finds breathing sounds, and sorts them into kinds: before the line, after the line, running into the line, running out of the line, might be part of a word, and faint. Each kind has a menu: Silence, Quieter, or Leave. Any single breath can be unchecked. Each row plays the clip, the breath alone, and after Apply, the adjusted clip.

Silence presses the breath down to the room's own floor and leaves the room tone as it is, so timing never changes and nothing is cut. A cleaned take is saved as a new version with its record, and the original stays.

This installer is signed.

Timeline. A bracket on the timeline outlines the lines you can see in the workspace and slides along as you scroll. When the timeline is zoomed in, it scrolls to keep the bracket in view.

ACX warning. The warning on the main page has a "Do not show this again on this computer" box. Turn it back on in Settings, Recording tab, ACX warning. The warning before an ACX export still shows every time.

Recording a part. The empty beat line reads "Optional: add a beat here."

This installer is signed.

Recording a part. "Add a part" is now "Add or change a part": change whose part you are reading, or add another part to read alongside it. "Show as" sits right beside it. You can change who says a line in Sides as well as in Scenes and Full script.

Sides with context. In Sides, choose how many lines to show before and after each of your lines, from 1 to 3 each.

Direction. The box under each line now reads "Optional: actor direction (Emotional direction, Context, Manner of speech (yelling, whispering))".

Celestial Aura. The Record button's label is readable again; main buttons with an icon show their words in cream.

This installer is signed.

Play buttons that agree. The play button on a take's box, the line's play button and the sound track's play buttons now work the same way: press to play, press again to pause, press again to carry on. Pressing play on a line stops whatever else was playing. Playback starts at the playhead when you have placed it inside the line.

Waveforms redraw after a trim. When a suggested trim is accepted, the sound track, the take box and the timeline now show the trimmed recording. Before, they kept drawing the recording from before the trim, so the last words could look cut when they were not.

The Recording panel. The buffer, input gain and monitor sliders are sized to the row in every theme, the knob and the fill line up, and a click lands where you point. The descriptions are rewritten in plain words, and the input gain note says what to aim for: peaks between -12 and -3 dB.

Recording a part. The Record button sits next to the line you are recording, and every other line of the part has its own record button. Click a speaker's name to give the line to another character.

Generate lines while listening is now a switch with a number beside it, in Settings, Recording. Off skips lines that have no recording; on generates the line being played and the number of lines ahead you choose.

Also: hold Alt with a clip's tools open to see which tools are always visible; the Horror theme's "Show me" and "Lightning" buttons are gone.

This installer is signed.

The built-in voices are back in the installer. Every 0.25 release left out the 26 built-in voices (the installer was about 360 MB instead of about 640 MB), so the installer offered "Built-in voices" but nothing arrived. This build carries them. If you installed any 0.25 version, install this one and the voices are on your computer; the Windows voices and the paid services keep working as before.

Every package is back, and the error box has a way out. 0.25.6 and 0.25.7 were packaged without some of the parts the audio engine depends on, so the engine stopped on launch with "Cannot find module 'body-parser'" and StellAurator showed an error box. When the audio engine cannot start and there is no newer version to offer, that box now offers "Reinstall StellAurator" (using the installer already on your computer when it is still intact, otherwise downloading it), "Open the download page" and "Show the log", instead of only OK.

Celestial Aura: the glass rims light up only where a light lines up. The bevels no longer glow all the way round. An edge the sun faces carries a wide, soft band along its whole length; an edge the sun runs along stays dark; the reflection moves as the camera swings.

This installer is signed.

Every package is back in the installer. 0.25.6 and 0.25.7 were packaged without some of the parts the audio engine depends on, so the engine stopped on launch with "Cannot find module 'body-parser'" and StellAurator showed an error box instead of opening. This build carries them all again. If you have 0.25.7 installed, opening it offers this update: press "Download and install".

The error box has a way out. When the audio engine cannot start and there is no newer version to offer, the box now offers "Reinstall StellAurator" (using the installer already on your computer when it is still intact, otherwise downloading it), "Open the download page" and "Show the log", instead of only OK.

This installer is signed.

Celestial Aura scrolls smoothly again. A per-frame style recalculation that every theme paid for is gone, the star field draws lighter while you scroll, and the glare fades its reflections during fast scrolling instead of lagging behind the cards. The sun is a sun now: a white-hot core, a warm bloom and a faint lens flare that moves with the sky.

ElevenLabs. Wherever the app or the website mentions connecting ElevenLabs, there is a link to create an account, with the disclosure beside it that StellAurator is an independent affiliate and is paid a commission.

This installer is signed.

The alpha, in the app. Alpha accounts see their pass and the alpha price under Settings › Account, and after each session the app sends basic usage counts — how long it was open and which features were used, never your text, audio or names. The last report is there to read under Settings.

Celestial Aura. The glare on the glass is wider and brighter and lingers after a scroll; the bevels are rounder so it travels a longer arc. The sun moves with the stars. The settings gear keeps its shape.

This installer is signed.

Celestial Aura: scrolling turns the craft. Scroll and the craft turns with you and stays turned: the stars ahead drift at a tenth of your pace, the reflection slides the other way, and the sky behind you shows the celestial objects your chapter's characters wear — or a random few for chapters without them. Bodies hang off-center like a real view; the rounded rims stretch what they reflect.

Horror: the house is not empty. Now and then a face appears over a portrait for a blink, or takes an icon's place until you reach for it. Lightning flashes across the page every so often, and once in a while something is standing in it. Show me and Lightning on the Horror theme card in Settings preview both.

Settings. The version row carries Update now, Restart now and When I quit. Celestial's slider fill stays inside the tube.

This installer is signed.

Celestial Aura: the planet ahead is far away. Seen from a great distance, it is a small bright disc about a degree and a half across, hanging just below the view; a modest scroll brings it up, and it sinks back when you stop.

This installer is signed.

Celestial Aura: you are aboard a craft. Scrolling pitches the whole craft: the star field ahead slides with it and settles back when you stop. Behind you hang a distant sun, the Earth and the Moon; the glass reflects them, they slide across the panes as the craft pitches, and their light glints along every panel's rounded edge. Scroll far enough and a planet rises into view below — a real sphere with its own clouds, a haze on the lit edge, and a night side in shadow. Every planet in the theme is shaded from the same sun.

Settings. The version row carries the update's own buttons: Update now, Restart now, When I quit.

This installer is signed.

Celestial Aura, refined from your notes. Panel faces are black. The band that crossed them is now the window of a reflection: reflected stars appear only inside it, at a steady brightness, and the window slides as the view swings. The rounded rim bends the stars behind it, so a star slips and stretches as it crosses a panel's edge. The sky is brighter, drawn live by the theme. The icons' auras are thinner, so they read as lines of starlight.

Clip library. The Settings text says plainly what cleaning does.

This installer is signed.

Celestial Aura: glass edges and a real sky. Every panel is a pane of glass with a slightly rounded edge, and a line of light runs along its rim wherever the light catches it. Move the window and the view swings for a moment, then settles; while it swings, the panes catch the reflection of a second star field. Behind everything is a black sky with stars placed by the theme itself, not a picture.

Clip library. The Settings text says plainly what cleaning does: it deletes recordings and generated takes that no line in any book uses anymore, never music or sound effects.

This installer is signed.

Celestial Aura, a twelfth theme. The studio becomes a window onto open space: no panel surfaces at all, a night sky behind everything, icons drawn from starlight with their dark bodies see-through, sliders and handles to match. Choose it under Settings › Themes.

Celestial objects, a new cast gallery. One hundred worlds, gas giants, stars, moons, asteroids and comets, nebulae, black holes and galaxies, available as cast images in every theme. Pick Celestial objects in the cast image picker or on the tag board, and a Kind list appears beside it to narrow the set. Every object has a description, so Best fit can match one to a character.

Small fixes. Dropdown lists in Celestial Aura have a solid ground; a test that failed once in 256 runs is fixed.

This installer is signed.

Nebula is on sale. Our cloud storage option carries your books between your computers. Three sizes - Nebula 25, Nebula 100 and Nebula 500 - billed monthly ($7, $13, $21) or yearly at half price ($42, $78, $126). It is bought on the Purchase page like the plans, and it appears in Settings › Nebula once it is on. Nebula needs a paid plan.

Monthly or yearly. The Nebula offer in the app and the Change plan dialog have a Monthly / Yearly switch. Changing size opens the payment page: the new size starts as soon as you pay for it, the size you have stays until the end of the period you paid for, and until then you have the larger of the two. After canceling, Turn Nebula on again buys a new period that begins the day the current one ends, so nothing is charged twice.

The Purchase page now offers Nebula above the plans, with No Nebula for now chosen unless you choose otherwise. Choosing a size buys it right after the plan, in a second checkout.

This installer is signed.

Recording: a proper panel for your hardware. Settings › Recording shows the microphone as a rack: the device with a status light (ready, in use by another program, not connected — and Record waits until it is back), the format it can do and the one it is set to (sample rate, bit depth, mono/stereo/left/right, buffer), an Input gain slider that sets the device's Windows level from inside the app, a real meter (peak and RMS on a dBFS scale, peak hold with the number, a CLIP light you clear by pressing it, the noise floor), a ten-second Level check with a verdict (including when room noise is above the −60 dBFS ACX limit), a five-second Test recording through the real capture path, monitor volume, and a profile remembered per device, with a built-in one for the RØDE NT1 5th Gen. Keep original captures keeps the untouched WAV of every take beside it (on by default for 24-bit and higher devices): re-renders and repairs work from the original instead of the MP3, and the original can be played or exported. A mono microphone is now encoded mono; a read-through keeps its capture format end to end; every take records exactly how it was captured.

Sign in the standard way. Settings › Account has Sign in and Create an account: each opens your browser to stellaurator.com, you sign in with Google or your email, and the browser hands the app a code that only this copy of the app can use — nothing to read aloud, nothing to type. Typing a code stays as a fallback for a computer that cannot open links, and codes now live ten minutes. The email-only "Start my free trial" button is gone: the free trial starts the first time you sign in with a confirmed email address. Signing in by email and password inside the app is still there, under a disclosure.

Also. A trim committed on a mono take no longer turns it stereo. The app's own account of the RØDE NT1 5th Gen is corrected: through its Windows driver the app captures 16-bit, not 32-bit float, and the panel now says what actually opened.

This installer is signed.

Portraits. In Character Tags and every other gallery list, the romance portraits and the masquerade couples now come last; the pool opens on another set.

The cue sheet. The science-fiction genre note reads: "Beds set the location, and the location changes scene by scene."

This installer is signed.

Generated audio and ACX. ACX does not accept text-to-speech narration. The app now says so in a pop-up at two moments: the first time a book holds both recorded takes and generated (text-to-speech) takes in a session, and every time an export includes a generated line — the export dialog tells you how many of the lines about to go out are generated, and asks you to export anyway or go back. A performance you recorded yourself and converted with speech-to-speech is still your performance, and the app now proves it (below). Also fixed: a tuned or repaired copy of a generated take was not being marked as AI-generated narration in the exported file; it is now.

Every converted take is signed. When a take is made from another take — a voice conversion, a re-render, a repair, a wave edit — it carries a record: which take it came from, the original recording at the root of the chain, the hashes of the source and the result, the provider and voice, the performer and release version when known, the consent it ran under, and the computer that made it, signed with that computer's key. The record is also appended to a per-book log. On the take: a SIGNED mark, and Verify in its details. Every export writes proof-of-performance.json beside its manifest; a book can also write a proof-of-performance.txt you can hand to a distributor.

An original with dependents cannot be deleted. A take that another take was made from is kept while that take exists: deleting it, flagging it for deletion, finalizing it away, or emptying it from the bin is refused with the reason ("v2 is kept: v3 (converted) was made from it. Delete v3 first, or keep both."), and its delete controls show why. Deleting a whole line still works: the original and its dependents go to the bin together.

The website says what the studio does now, with every screenshot re-taken on a real published book, and describes StellAurator Nebula (opening soon) and its prices.

This installer is signed. It carries a code signature issued through Azure Artifact Signing to the publisher, with a timestamp, so Windows no longer shows the "unknown publisher" SmartScreen warning for it. Nothing in the app itself changed from 0.22.0.

Nebula prices. When Nebula opens, its three sizes are Nebula 25 at $7 a month, Nebula 100 at $13 a month, and Nebula 500 at $21 a month. The app shows the prices the service sends it, so this note is for the record; Nebula is still switched off until it is opened for sale.

StellAurator Nebula is built into this version, and switched off until it is opened for sale. Nebula is our cloud storage option: it carries your books between your computers. You put a book in Nebula from one of your linked computers and take it down on another, whether or not the first one is on. Everything is sealed on your computer with keys that only your linked computers hold, so we cannot open what you put there. It is not a backup: the master copy of your work stays on your computer. Nebula is a monthly add-on to a paid plan in three sizes (25, 100 and 500 GB), with Change plan and Cancel in the app; when Nebula ends you have 60 days to take your books down, then its contents are deleted for good. You can also share one book or one file by a Nebula link: the link carries its own key, opens only that one thing, and can be stopped at any time or ends by itself at the date and download count you chose. None of this appears in the app until the service is opened; when it is, a Nebula button appears in the library and a Nebula block in Settings › Account.

The license agreement is version 1.4. One sentence was added: if you turn on Nebula, the books and files you choose are uploaded sealed, as the Privacy Policy describes, and we cannot open them. The app asks you to accept it once at the next start. The billing terms (1.5) gain a Nebula section and the privacy policy (1.3) says exactly what Nebula holds about you: sizes, dates, counts and which of your computers hold the key — never a title, a file name, a word, a recording or a key.

Also. The installer is still unsigned, so Windows SmartScreen warns; the signing chain is complete except for one sign-in on the build machine.

The license agreement is version 1.3. The app shows it once at the next start and asks you to accept it again. What changed: which license you hold (trial, subscription, or Gold) and how long it lasts is now stated in the agreement itself, and a Gold license is named as perpetual; a bug report or a contact message is the only way any of your content reaches us, and the agreement says exactly what one carries; the machine identifier is described exactly (a one-way hash of hardware details, never of a voice or a face); the in-app confirmation that you hold a person's consent to clone their voice is recorded as a confirmation, not as the consent itself; the compliance marks and notes the app shows are described as aids, not determinations; the liability limit is the greater of the last twelve months' payments, a Gold license price, or $100, with fraud and willful misconduct never excluded; you must be 18, or of age where you live, to accept it.

The performer release is SVR-1.1. Outtakes are a fourth separate No: a performer's false starts and alternate takes may be published as bonus material only when that box is granted. Releases already accepted under SVR-1.0 keep their exact words. The header no longer says "not yet reviewed by an attorney"; it says what a standard form is.

Also. The billing terms (1.4) and the privacy policy (1.2) on the website say the same things in the same words: refunds stated as the rule first and the exceptions second, a Gold withdrawal recorded with its ground and appealable within 30 days, and each kind of record with its own retention. The Windows build is ready to be signed with Azure Artifact Signing; this installer is not yet signed, so Windows SmartScreen still warns.

Proof listen and sign-off. Under the chapter header, Proof listen plays one chapter of one version exactly as the export will, ticks each line as you hear it, and offers a sign-off once every line has been heard. A sign-off records the takes that were heard; an edit that changes one of them un-ticks that line and says so, and "Play the changed lines" restores it. Nothing is ever locked against editing. Flag any line for redo with the ⚑ on it, or with F while it plays; the Pickups list gathers the flags with who, when and why, and can regenerate, record or send them as a batch. The chapter list counts them, and the export dialog says which chapters were proof-listened — it never blocks.

Bring your recordings. A chapter you recorded elsewhere, as one file, becomes a chapter of takes: drop the file in, say which chapter it is, choose how to divide it, and review what matched, what came from an earlier draft, what could not be placed (listed with its timecode, never silently dropped) and any line whose words differ from the manuscript. Your file is never changed and the audio is never re-mastered. Lines that could not be placed are flagged for redo; the import finishes in Proof listen.

Auto Score, the cue sheet. A cue sheet per chapter: add a cue by hand, set its kind and mood, print or copy a sheet a composer could work from, and see the cues on the timeline. With a text model configured, "Find the beats" reads each scene and proposes where music belongs — a bed under the sentence before the moment, a sting on it — each with one sentence of why. Accept, move or reject each one. No music is generated yet; that comes later.

Every AI prompt is yours to edit. Beside each function that asks a text model — finding the characters, describing them, briefing the narrator or the illustrator, casting faces, finding the beats — a button opens the prompt: the instruction in plain prose, one field per piece of information it is given, a preview of what will be sent, and Reset to default. Changes are kept per book.

Where do I start? A first run asks four questions — where you are coming from, what stage the book is in, what you want, where you want to end up — and pins a checklist whose steps point at the real controls. No account and no key are needed for it.

Also in this release. The 3-2-1 count-in does not start until the microphone is confirmed recording. Auto-stop waits 3 to 10 seconds of quiet, your choice, or can be turned off. The sound track's tools are icon tiles like every other clip control, and play turns into pause. A version column can be deleted from its settings when it is empty. "Read again the lines that…" regenerates dialogue that runs into a speech tag, the tags, or lines containing a word. Column A now exports the take you locked in, not the newest one. The update rescue for a copy whose engine cannot start no longer stalls under a second dialog, and its wording says what happens in order. Gilded's textures no longer show a seam.

Recording keeps your first words. The take used to be timed from the moment the microphone was asked to open, not the moment it was actually recording — about half a second later on a USB mic — so the first words of a line could be cut off as count-in. The 3-2-1 now does not start until the microphone is confirmed recording, and the quiet before your first word is measured, not assumed. The take also no longer stops itself mid-sentence: a long held line could push the "is someone talking" threshold above your own voice, and the auto-stop clock treated 80% of the reading-speed target as "the line is over". The auto-stop now waits 3 to 10 seconds of quiet (your choice, under Settings → Prompter), or can be turned off.

A trim is suggested, then cut. After a take, the quiet at each end is shown on the waveform as dashed handles with an Approve button instead of being cut out of the file on the spot. Press Approve, drag a handle, or simply move on — record the next line, change chapter, close the book, or export — and the trim is accepted as shown. By default the quiet is then removed from the file; a setting keeps the whole take on disk and skips the quiet in playback and export instead.

Hold Ctrl to see what every icon is. With the pointer over nothing in particular, holding Ctrl shows the name of every icon on screen. Over a tooltip it still opens the full explanation.

Deleting is safer. Clean out orphaned clips was deleting background beds and stacked sound layers that were still in use. It now only removes spoken takes that no line points at, and never touches sound effects, music or a book's bin. A chapter is labelled as containing AI narration only if that chapter does; before, one generated line early in a book tagged every chapter after it.

The app says what it does. Choosing which tools show on a clip now applies to every clip, as it was always meant to. Several hundred tooltips and messages were rewritten to say plainly what happens and what to press — six of them were wrong, including two that promised things the app does not do. What ACX requires is stated as ACX states it. People without an ElevenLabs account get a sign-up link, with the disclosure beside it. Contact from the website goes to mike@stellaurator.com.

Deleting no longer destroys anything. A deleted take, clip, layer, background bed or whole line now moves into a bin inside the book, with a note of where it came from and why it left - play it, put it back, or empty the bin when you are sure. Chapter Undo brings the sound back with the record, which is what everyone assumed it already did. You will find the bin in Settings, under Files and folders, with a five gigabyte allowance per book. While building it we also found and fixed a long-standing bug: Clean out orphaned clips was deleting live background beds and stacked layers. And on each character card there is now a Convert all button - every line you recorded for that character in this chapter, read again in their cast voice, one conversion per line, added under the originals.

Version columns can now take their own clips. Tell a column to take generated, voice-changed or recorded clips and they land there as you make them - the column header shows which kinds it takes, and anything you placed by hand is left alone. Turning it on offers to claim what is already in the book, in one undoable step. Exporting a version that is missing clips now explains itself instead of just warning: which characters are short, by how many, and what another column could supply - then fill from that column, generate the missing ones, or export anyway knowing the count. Also: background lanes can be deleted (empty ones without being asked), a version column with clips in it can no longer be deleted by accident, and the add-column button no longer pushed the version headings out of line with their clips.

Side panels you can drag wider or narrower, or fold away (the sidebar, the timeline and more) - they remember how you left them. No more clicks at the start and end of lines in Listen and in exports. A quote and its "she said" now read as one sentence: a new, short "Inside a sentence" pause (Settings > Audio), and the built-in voices are told whether the sentence carries on. See exactly where a line is cut, on its waveform, and drag the handles - past the end to hold it longer. If the audio engine ever cannot start, the app now offers an update and shows the real reason. Cast tab buttons no longer get cut off in a narrow sidebar.

Fixes 0.16.0, which could not open: a file the audio engine needs was left out of the installer. If 0.16.0 tells you the audio engine did not start, download and run this installer over it - your library and settings are untouched. Everything new in 0.16.0 (your own computers, linked) is in here.

This build cannot open: a file the audio engine needs was left out of the installer. Use 0.16.1 instead, which is this release with that fixed. If you already installed 0.16.0, run the 0.16.1 installer over it - your library and settings are untouched.

Invite an actor by email: they get a link, download StellAurator free, open the invitation, and from then on you send the part and they send their takes directly - no files to juggle. You can withdraw an invitation at any time. Also: fixes to sending over the internet, and safer license management.

Collaboration, end to end. Review an actor's takes side by side (Preferred, Approve, Reject, Again please) with a completion line per part; pickup requests travel as small packages that open into the actor's existing project; Voice the scene and Run the Scene. Send: choose some or all takes and the app tries every way to deliver them - the route already set up, directly on your network, directly over the internet, a shared Dropbox / OneDrive / Google Drive folder, a USB drive - and tells you what each method needs; if none works it walks you through setting one up, with a message to copy for the other person. Back up this book: one .stellabook file holds a whole production and opens on another computer. Choose where things are saved (Settings > Files & folders). American spelling throughout.

Speaker detection fixes: one Dr. Hart instead of three, a line tagged 'she screamed' can no longer go to a man, and the person being spoken to is no longer mistaken for the speaker. Settings now shows How detection decides, takes your own instructions for the book, and has Run detection again - press it once on books you already imported. Sign in through your browser (Google or any way you use on the website) and the app signs itself in. Collaboration: assign a part to an actor, choose what they may see, send them a package, import their takes; performer agreements and permissions; Cast tab: Record a part, Character Tags.

StellAurator Gold: perpetual licences. A Gold key (GOLD-XXXX-XXXX-XXXX) turns an account into the full studio for good - redeem it from the link you were sent, or in Settings > Account. Also in this release line: the update bar answers clicks everywhere (0.12.1) and one character's part can be recorded on one screen with direction and cues (0.12.0).

Fixed: the update bar (and the Stop button of the generating bar) often ignored clicks - most of each button was being treated as part of the window's title bar. They now answer wherever you click, and show at once that the press was taken. Also in 0.12.0: record one character's part on one screen (the microphone on a cast card), with scene, character, beat and line direction and the line before yours played as a cue.

New: record one character's part on one screen. On the Cast tab, press the microphone on a character: their scenes, the direction for them (scene, character, beat and line - write it right there), the line before each of theirs played as a cue, then record, listen back, lock in, next. Show as Sides or Blind previews exactly what a remote actor would be given. This is the first step of collaboration; nothing leaves your computer yet.

The licence agreement now names its governing law (Arizona) - the app asks you to accept it once more. Nothing else changes.

The licence agreement now names its licensor (Jathtech Digital Consulting LLC) and is no longer a draft, so the app asks you to accept it once more. Nothing else changes.

A free 7-day trial - no card - that starts the first time you sign in. When a trial or plan ends the app becomes read-only instead of locking: every book still opens and plays, and your recordings stay yours. Settings > Account now lists the computers on your account (up to three) and lets you remove one.

Keeps signing in, invitation keys and bug reports working while the account pages move to account.stellaurator.com.

Signing in, invitation keys and bug reports no longer depend on the website being up. The Purchase button opens the purchase page directly. The readthrough no longer names a transcription service you have no key for.

Windows such as a character's profile now always open fully on screen, above the top bar, on small laptop screens too. Settings > Account has a place to enter an invitation key.

Recording a single line: the auto-stop now judges speech against your room rather than a fixed level, so it works with a hot microphone in a lived-in room, and the take is trimmed to exactly the quiet the recorder counted before your first word and after your last. A stray click no longer counts as having spoken.

Update a manuscript without losing your work: choose Update manuscript on a book, and the app compares the edited text with the book, keeps every unchanged line's takes and approvals, and walks you through what changed - accept, keep the old audio, match a rewritten line by hand, or decide later. Afterwards, changed and new lines wear badges with Regenerate and This-take-is-good, and Next changed jumps between them. Copies and editions of a book share their audio files instead of duplicating them; Settings has Clean out orphaned clips. Also: the Voice audit lists takes that no longer fit their line.

Right-click on any text field now offers Cut, Copy, Paste and Select all. Recording a single line: the take now starts a quarter-second before GO, stops by itself after a quiet spell once you have finished the line (Settings > Prompter, 0-3 seconds, default 2), and the silence at both ends is trimmed to a small gap. Pressing Stop still works at any moment.

A gold bar shows whatever is generating, with a Stop button. Listening generates only the line it plays unless you ask for more (Settings, Audio). The practice book uses the built-in voices when they are installed. The installer's finish page can start the app, and a desktop icon is a choice.

The installer now asks what to install: the app (required), the built-in voices (311 MB), and all eleven themes (otherwise Gilded, Vivid and Gilded Ivory). The choice is remembered across updates. Also: re-running speaker detection keeps characters you created and speakers you chose by hand.

Re-running speaker detection now keeps characters you created and speakers you chose by hand. A line's speaker menu shows 'Set by you' with a way to hand it back to detection.

Updates now carry a machine-local data-folder setting across an install. Otherwise the same as 0.9.2.

First release with in-app updates. From 0.9.2 on, the app finds new versions itself, asks before downloading, and installs when you say so - never during a recording.