
What it is
The work that comes before the voice
An audiobook is heard in a few hours and prepared over many days, and most of those days are spent on things a listener never notices. Somebody decides where each chapter begins and ends, and what it is called when the narrator announces it. Somebody reads the text for the marks that make sense on a page and nonsense in the ear: the stray heading, the broken line, the footnote number sitting in the middle of a sentence. Somebody makes the cover that will sit in a listener's library for years. AudioBook AI is built for that stretch of the work, the part that happens before anyone presses record.
The order it keeps is plain. A person brings a book in, by pasting its text or uploading a file. An AI reads the text and divides it into chapters, each with a title, and the person reviews those titles before the book is saved. On the book's own page, an AI then goes through the chapters one at a time and prepares each one for reading aloud, cleaning the text so that it flows when spoken. An image generator makes a cover for the book and a picture for every chapter. Every book the person has brought in is kept in a library, searchable by title and author, and each one opens to a gallery of its chapters.
It is a book prepared for listening: divided, titled, cleaned for the ear, and dressed with a cover and a set of chapter pictures, held in order until a narrator or a voice takes it up.
It matters to say plainly what the app does not yet do, because its name suggests more. It does not make audio. No voice reads the prepared text, no speech service is called, and no sound file is written to any chapter. A note left in the code names Azure's text-to-speech service as the step meant to follow the preparation; that step is not built. The app has a player, drawn at the foot of the book page with its controls in place, and it is silent, because there is nothing yet for it to play.
So the honest description of AudioBook AI is the one its own record keeps. It is a book prepared for listening: divided, titled, cleaned for the ear, and dressed with a cover and a set of chapter pictures, held in order until a narrator or a voice takes it up. That is a smaller claim than the name makes, and a useful one. The preparation is real work, it is the work that most often goes wrong, and here it is done in one place, with every chapter's state written beside it.

A day with it
Thursday, a memoir in nine parts
The production editor of a small regional press arrives on a grey Thursday with one book on her list. A retired harbour pilot has written a short memoir of forty years bringing ships into the estuary, and the press has agreed with him that it should exist as a recording before it exists in print, because most of the people who will want it are his old colleagues, and many of them now read more easily with their ears than with their eyes. He has sent the manuscript as a Word file. Her job today is to get it ready for the reader the press has booked for next month.
She opens AudioBook AI and the home screen shows her the four most recent books she has prepared, each with its cover and its status, and a button that says Create Audiobook. She presses it. The first step is called Import Your Book, and it asks for a title, an author and the text. She types the title the pilot chose and his name, then moves to the Upload File tab, which names the five kinds of file it reads, and chooses the manuscript. A loader turns while the file is sent up and its text drawn out, and then a line confirms that the content has loaded.
Out of curiosity she looks at the third tab, Connect, which shows a button for Google Drive and one for Notion beneath a line about connecting accounts. Neither does anything when she presses it, and she did not need either today. She returns to her text and presses Detect Chapters.
The loader turns again while an AI reads the memoir and divides it. When it returns, the screen is called Review Chapters, and it says the AI has found nine chapters in her book. Each one sits on its own row with its number, its title in a field she can type into, and its length in characters underneath. Most of the titles are the pilot's own. The fourth, where he had written no heading at all, has come back as a plain section number, and she renames it after the storm it describes. She presses Back once to look at the text again, finds it all still there, and returns. Then she presses Create Audiobook.
The book's page opens. At its top are three buttons: Back, Generate Images and Generate Audio. Below them is an empty frame where the cover will be, with a quiet line inside it saying the cover is made by clicking. Beside it sit the title and the pilot's name, three small tiles counting nine chapters, a status of Draft and nothing yet ready, and a bar called Audiobook Progress standing at none of nine.
She clicks the empty frame. A little while later a cover arrives, made by an image generator from the title and the author's name, in the cinematic register the app asks for. It is not quite right, so she clicks again, and a second cover takes its place. She keeps the second.
Then she presses Generate Images. Below the cover, a section called Chapter Gallery holds nine tiles, each one saying No image yet. One by one, in order, they fill: a pilot's ladder against a hull, a harbour in fog, a chart table under a lamp, each made from the chapter's title and its opening lines. The page keeps each picture as it lands, so she can watch the gallery become a sequence while she answers an email.
Last, she presses Generate Audio, knowing what it does and what it does not. Below the gallery is a list called All Chapters, a card for each chapter with its picture, its number, its title, the first lines of its text and a small chip that says Pending. As the AI works through the book, each chip changes to Generating and then to Ready, and the progress bar counts them up. Ready, here, means the chapter's text has been prepared for narration: its stray marks cleaned away and its punctuation set for the ear. When all nine have turned, the status tile beside the cover changes from Draft to Ready.
She opens the chapter about the storm from the gallery. A player rises from the foot of the page with the chapter's picture and its title, and its controls are all where a listener would expect them. It makes no sound, and she knows it will not, because the voice will be the reader's next month. She closes it, goes to the library, types the pilot's surname into the search box, and finds the memoir sitting among the press's other books, marked Ready, waiting for its narrator.

Bringing it in
A text, a file, and the chapters found inside it
Every book begins on the same screen, and the screen asks for very little. There is a field for the title and one for the author, and below them three tabs. The first, Paste Text, is a single large box for a person who already has the book open somewhere and would rather copy it across than look for the file. The second, Upload File, takes a document directly, and it says which kinds it reads: plain text, PDF, the two Word formats and EPUB. The third, Connect, shows two buttons for Google Drive and Notion. Those two buttons are drawn and not yet wired, so for now a book arrives by pasting or by file.
When a file is chosen, the app uploads it and asks an extraction service to read three things out of it: the full text, the title and the author. If the file names its own title and author, those fill the two fields at the top, in place of anything typed there. The text is then held ready for the next step, and a line on the screen confirms that it loaded. Until there is text in hand, the button that moves the work forward stays dim, so the next step cannot be started on an empty book by accident.
A short book, a long essay, a novella or a set of lectures fits inside that, and a full-length novel does not.
That button is Detect Chapters, and it is the first place an AI does work in this app. The text is sent to the AI with a short instruction: divide this book into chapters, give each chapter a title and its content, and where the chapters are not clearly marked, make sensible sections of roughly equal length. The AI is also asked to suggest a title and an author. Those suggestions are used only where the fields are still empty, so a title a person has typed is never replaced by a guess.
There is one limit here that a person should know before they begin. The AI is given the first fifty thousand characters of the text, and no more. A short book, a long essay, a novella or a set of lectures fits inside that, and a full-length novel does not. For a long book the honest way to use the app is a part at a time, each part brought in as its own entry in the library, so that nothing is silently left off the end. The chapter text the app keeps is the text the AI returns for each chapter, which is why the review that follows matters.
The review screen is called Review Chapters, and it opens by saying how many chapters were found. Each chapter sits on its own numbered row. Its title is in a field that can be typed into, so a section the AI has called by a number can be given the name its author would have given it, and a title that is too long for a narrator to announce comfortably can be shortened. Under each title is the chapter's length in characters, which is the quickest way to notice that something has gone wrong: a chapter of a few hundred characters where the rest run to many thousands is usually a heading that was mistaken for a chapter.
From the review a person can go Back to the first step, where the text is still waiting, or forward with Create Audiobook. The review lets a person rename chapters and nothing more; the text of a chapter is not edited here, and chapters are not merged, split, reordered or removed. When the book is created, it is saved as a draft with every chapter marked pending, and the app opens the book's own page, where the rest of the preparation happens.

Before any voice reads a chapter, somebody has to decide where the chapter ends. That is the first work this app does.

Preparing it
Text for the ear, a cover, and a picture for each chapter
The book page carries three pieces of preparation, and each one is started by a person, when they choose, from a button or a frame on that page. Nothing runs by itself. A book can sit as a draft for a month with its chapters found and nothing else done to it, and that is a perfectly good state for it to be in.
The first piece is the text itself, and the button for it is called Generate Audio. What it does is narrower than its name, and worth describing exactly. The book's status becomes processing. Then the app goes through the chapters in order, and for each one it marks the chapter as generating, sends the chapter's title and the first ten thousand characters of its text to an AI, and asks the AI to act as an audiobook narrator preparing a script. The instruction is specific: remove any formatting artifacts, fix the punctuation, and make sure the text flows naturally when it is read aloud. The AI returns the cleaned text, the app writes it into the chapter, and the chapter is marked completed. When every chapter has been through, the book is marked completed too.
No voice reads it and no audio file is written, so Ready on a chapter means its text has been prepared for a voice, not that it can be heard.
That cleaned text is the whole of what this step makes. No voice reads it and no audio file is written, so Ready on a chapter means its text has been prepared for a voice, not that it can be heard. The app sends the prepared text back with the chapter's record, in a field the book's own schema does not declare, and no screen in the app shows it yet. If the AI fails on one chapter, that chapter is marked with an error and the work moves on to the next, so one difficult passage never holds up the other eight. And because no chapter ever receives a sound file, pressing the button a second time prepares every chapter again from the start.
The second piece is the cover. An empty cover frame on the book page says that a click will make one, and a click does: the app asks an image generator for an audiobook cover built from the book's title and its author, in a register the prompt describes as elegant, modern and cinematic, with rich colour and dramatic light. The cover is saved to the book and appears in the library and on the home screen from then on. If it is not right, a second click on the cover makes a new one in its place, and a small turning mark appears over the frame when the pointer rests there, to say that it can be done.
The third piece is the chapter pictures, started by Generate Images. The app walks through the chapters in order and, for each one that does not yet have a picture, asks an image generator for a book illustration drawn from the chapter's title and its opening five hundred characters, in an atmospheric, cinematic style. Each picture is saved to the book the moment it is made, so the gallery fills one tile at a time and a person can watch the book acquire its faces while they do something else.
Two small rules in this step are kinder than they look. A chapter that already has a picture is left alone, so running the step again after a failure fills only the gaps and never paints over a picture somebody liked. And a picture that fails to arrive is simply skipped; the rest of the chapters still get theirs. The result, after a few minutes, is a book with a cover and a sequence of images, one for each chapter, that belong together because they were asked for in the same register.

The book's page
A gallery, a card for every chapter, and a silent player
The book's page is where a prepared book is looked at, and it is arranged so that its state can be read in a few seconds. On a wide screen the cover sits on the left. Beside it are the title and the author, or the words Unknown Author when none was given, and three small tiles: how many chapters the book holds, its status, and how many chapters are ready. Under the tiles a bar called Audiobook Progress shows the ready chapters against the total, in a fraction a person can read at a glance. The status tile says Ready once the book is completed and Draft at every other moment, including while its chapters are being prepared.
Below that is the Chapter Gallery, the part of the page that shows what the pictures were for. Each chapter is a tile with its illustration filling it, or the words No image yet where a picture has not been made, and when the pointer rests on a tile its title and its number appear across its foot, with a small play mark beside them once the chapter's text has been prepared. Seen together, the tiles let a person check the shape of a book the way an editor checks a contents page: whether the sequence holds, whether one picture sits wrongly among its neighbours, whether a chapter has been given a title that does not fit.
Under the gallery is a longer list, All Chapters, in which every chapter has a card of its own. The card carries the chapter's picture, the words Chapter and its number, its title, the first hundred and fifty characters of its text and a small chip that names its state: Pending before anything has been done, Generating while the AI works, Ready when the text has been prepared, and Error if that chapter's preparation failed. A Ready card also shows a play button. The cards carry each chapter's artwork and its place in the work. They carry no audio, because none has been made.
The play button, like any tile in the gallery, opens the player. It rises from the bottom of the page and holds the chapter's picture and title on the left, a row of controls in the middle with play and pause at its centre, a position slider with a time at either end, and on the right a mute button, a volume slider and a way to close it. Most of those controls answer a touch: the pause turns to play, the slider moves, the volume falls to nothing when muted.
It is right to be exact about the player, because it looks finished. There is no sound behind it. The position slider and the volume move only what the screen shows, the length it displays is the same five minutes for every chapter, and the buttons for shuffle, skipping and repeating are drawn without anything behind them. It is the frame of the player a finished audiobook would need, set in place on the page where it belongs, waiting for the voice step that the code names and does not yet take. Closing it returns the page to the gallery exactly as it was.

The shelf
A library to search, and a home screen that remembers
Every book a person brings in is kept, and the library is where they are all found together. It opens on a heading and a count of how many books it holds, and a button for a new audiobook sits at the top right so that starting another never needs a second screen. Below the heading are two ways to narrow the shelf, which work together rather than one at a time, and a third control that changes how it is shown.
The first is a search box. It reads both the title and the author of every book as a person types, so a press with forty books in preparation can find the one by a particular writer by typing part of the surname, without remembering what the book was finally called. The second is a filter for status, set to every book by default and narrowed to the drafts, to the books being processed or to the completed ones. The third is a pair of buttons that change the view. The grid shows each book as its cover with its title, author and status beneath. The list shows a smaller cover beside the title and author, adds the number of chapters the book holds, and puts the status at the end of the row, which is the better view for a long shelf.
It is the screen a person sees first every morning, and it is arranged so that the books begun most recently are already in front of them.
Every book in the grid has a small menu. One item in it, Delete, removes the book from the library at once and asks nothing first, so it is worth using with the same care as any other permanent act. The other item, Download, is drawn in the menu and not yet wired to anything; nothing in the app can be downloaded today, and the prepared text and the pictures live inside the app's own records. Each cover in the grid also shows a Play button when the pointer rests on it, and that button simply opens the book's page, as the cover itself does.
The library is careful about its empty and its waiting states. While books are loading, eight quiet placeholder shapes stand where the covers will be, so the page does not jump when they arrive. When a search or a filter finds nothing, the page says so, suggests creating a first audiobook and offers the button to do it. A book whose address is wrong opens to a plain line saying it was not found, and nothing else.
The home screen is the other way into the shelf. It opens on a short statement of what the app is for, a button to create an audiobook and three cards describing the work, two of which name things still to come: a connection to Google Drive and Notion, and a finished audiobook to download. Once a person has books, a section below them shows the four most recent, each as its cover with title, author and a status that reads Draft, Processing or Ready, and a link to view them all. It is the screen a person sees first every morning, and it is arranged so that the books begun most recently are already in front of them.

A picture for every chapter, so the book has a face at each of its turnings and not only on its cover.

In the suite
One room of the publishing floor in B-Corps
AudioBook AI is one of the nineteen applications that make up B-Corps, the house's own operating system, and it is installed on the house's own Mac by its own method, against its own cloud backend. It sits on the same desk as the other eighteen. That desk starts each app, opens it and lets it rest, and every two minutes it restarts any app that has stopped, so the room is there when a person comes to it. A stock-take of every app's records and calls is kept beside them, and in this app's row it reads simply: books, the people who sign in, and the four services it calls for files, text, chapters and pictures.
It belongs to the publishing floor of the suite, and its neighbours there do different parts of the same work. Chapter to Blog takes a finished book, has an AI find its chapters and writes each one as a post for the open web; a manuscript brought into both apps is divided twice, once for reading on a screen and once for listening. Luminous BookFlow is the house's publishing house in software, the place a book is carried through its life; AudioBook AI is the room where that book's chapters and pictures are prepared for an audio edition. ContentSpark sets one source out across every channel and into a course, and the chapter pictures made here are natural companions for what it makes from the same book.
The suite is arranged to be used side by side, and the connections between these apps are the ones a person makes by keeping them open together. Nothing passes from one to another automatically. A person who keeps Resonance Canon open beside this app, where each book is read as a course with others and a coach, can prepare that same book here chapter by chapter for the members of the course who would rather listen. A person who keeps OmniWriter open, the writer's desk with the writing, the posts and the book's own page, has the book's public face in one window and its listening edition in preparation in the other.
At the centre of the suite is The Luminous Book, the brand studio that reads a business and writes down its voice. A press that has written its voice there has something to hold the chapter titles against when it renames them on the review screen, and something to hold a generated cover against before it keeps one. SquareDash, the house's marketing workspace for websites, is where a book's page is built and its articles written, and a cover and chapter pictures made here can be placed on that page by hand. The house's reading room reads every app in the suite and every book the house holds, and this app is among the ones it reads.
Each app in B-Corps keeps its own sign-in today, and AudioBook AI is no exception. A person signs in to it on its own, and its books belong to its own records. The value of the suite is not that its apps talk over a person's head. It is that each one does a single part of the work well and plainly, and that a person running a small press or a studio can keep the whole of a book's life within one set of rooms, each built to the same standard and kept running on the same desk.

Who it is for
A press, a studio, an author with a finished text
It is for the person who has a finished book and has decided it should be heard. That might be the production editor at a small press preparing a title for its narrator, or an independent author who intends to record the book themselves and wants it divided and cleaned before they sit down at the microphone. It might be a studio that produces audio editions for other people's books and needs every manuscript brought into the same order before the work is scheduled. It might be a teacher turning a short course text into something students can listen to on the way to class.
What these people share is a manuscript that was set for the page and must now be set for the ear, and a stretch of preparation between the two that usually happens in a word processor, in a spreadsheet of chapter titles and lengths, in an image folder somewhere else and in a running list of which chapter is at which stage. AudioBook AI puts those four things in one place. The chapters and their titles live in the book, the prepared text lives with each chapter, the cover and the pictures sit on the same page, and the state of every chapter is written beside it in a word.
What it replaces, then, is not a recording studio or a narrator. It replaces the folder of loose parts that a book becomes on its way to the studio. The difference is felt in small moments. The editor no longer has to open the manuscript to remember what chapter seven was called, because the gallery shows it with its picture. She no longer has to keep a separate list of which chapters have been cleaned, because the chips say so. And when the narrator asks how the book is divided, the review screen and the chapter list are the answer.
There are good tools for producing finished audio, and many of them begin at the point where this app stops: with a clean script and a known set of chapters. That is why the preparation matters. A voice, human or synthesised, reads exactly what it is given, so a chapter boundary in the wrong place or a stray heading left in the text becomes a mistake in the recording that someone has to find and fix later, at a higher cost. AudioBook AI is built so that those mistakes can be seen while they are still text.
This page tries to be honest about the app's edges, because its screens do not yet say them: nothing tells a person that the AI reads only the first fifty thousand characters, and the Generate Audio button, the Audiobook Progress bar, the play buttons and a player that shows five minutes for every chapter all speak as though the voice were there. A press that knows exactly what a tool does can build a week around it. A press that has to discover what a tool does not do, on the morning a narrator arrives, cannot.
How it is offered
Installed for an organisation, and kept
AudioBook AI is offered to an organisation in three ways: on its own, as part of the publishing floor, or inside the whole of B-Corps, where every application on these pages is installed together, set up in the organisation's own voice and with its own records, and kept running for as long as the organisation keeps it. The house carries each application into its suite for good, so what is set up now is still there, and still mended, in the years after.
The fee reflects the work of installing and shaping it, and the size of the organisation it serves. It is set with each organisation, after a conversation about what the application will carry for it.
Arranged in conversationAudioBook AI · B-Corps

The care it takes
What is kept, what is said, and what waits
The app asks a person to sign in before it shows anything, and a person who has not been given access is told so plainly, with what to do about it, rather than shown an empty screen. Every book belongs to the app's own records. Nothing in it is published, posted or sent anywhere: the prepared text, the cover and the chapter pictures stay inside the book until a person takes them out by hand. Where an AI does the work, the page says so; the chapters it finds and the pictures it makes are laid out on the screen to be judged before anyone relies on them, and the prepared narration text is the one result kept out of sight for now.
Its small rules are written in a person's favour. A title someone typed is never replaced by the chapter-finding AI's suggestion. A picture someone kept is never painted over by a second run. A failure in one chapter's preparation is marked on that chapter alone, a picture that fails simply leaves its tile waiting, and neither stops the rest. Its loading and empty states are drawn rather than left blank, and a missing book is named as missing. The one sharp edge is deletion, which is immediate and asks nothing first, and the shelf is the place to treat it with care.
Long before the first word is spoken into a microphone, somebody has to sit with the text and decide where each chapter ends.
This page has tried to say only what is true, including where the app's own labels run ahead of it. A chapter marked Ready has had its text prepared for a voice. The voice itself is the step still to come, named in the app's own code and not yet built, and when it arrives it will have a book that is already divided, titled, cleaned and illustrated, waiting in the library in its proper order.
A recording lasts as long as its listeners keep it, and the best of them are kept for decades. Long before the first word is spoken into a microphone, somebody has to sit with the text and decide where each chapter ends. This is the room for that.
B-Corps
