How the reader works, yapbook

Playing

Play: Starts the book from where you left off; press again to pause. Your place in every book is remembered in this browser.
Chapters: The Chapter select and the outer bar buttons move a chapter; the inner ones move a sentence.
Any sentence: Tap it in the transcript, or reach it with Tab and press Enter, to read from there. What is still ahead is dimmed.
Speed: 0.8 to 1.5 times. A voice of your own is recorded at your pace and then stretched to the speed you choose without a change of pitch, so you still sound like you.

Books

Drop in an EPUB, PDF, Word, Markdown or text file, paste text into the box under it, or paste a web address there (one line; the site's server fetches the page for you and keeps nothing, and the readable part of it becomes the chapters), paste a paper's arXiv number or address (2501.07088, or its abs or pdf address: the paper's web version is read, which does better than its PDF; a paper without one is said so, and its PDF can be dropped in), or try the sample. EPUB reads best; a PDF needs a text layer; a page that is only built by scripts, or that sits behind a login, comes back with little text, and pasting the text works instead. Books are kept in this browser only, and each can be deleted from the library. Chapters named Contents, Index, References, Bibliography, Acknowledgments or About the author are passed over with a word; the chapter list marks them skipped, and choosing one there reads it anyway. A picture with a caption is announced as a figure in the second voice, its caption read where it sits; with Describe the pictures on (under the file box), what the picture shows is read too. A formula is read the way a tutor would say it ("the sum from i equals 1 to N of x sub i", "the expectation over x drawn from p, of f of x") where the page gives its source, as a paper's web version and a LaTeX-born EPUB do, and in Markdown between dollar signs; a formula in a PDF, or one with no source at hand, is read as it appears. Please keep out documents with card numbers, Social Security numbers, passwords or other private details: when you ask a question, the part you have heard goes to the hosted brain.

Asking

The one button under the transcript does what the moment needs.

Ask by voice: Opens the microphone. Speak your question; it ends on its own when you stop talking.
The first question: Takes a few seconds longer, and so may the second: speech-to-text loads at the first question, and the voice warms up on the first answers. After that, answers come quickly.
Stop listening: Ends the recording where it is and sends what was heard.
Interrupt: Cuts into an answer being spoken, for a new question.
The player, by voice: Say "next chapter", "previous chapter", "say that again", "slower", "faster", "stop in twenty minutes", "stop at the end of this chapter", "cancel the timer" or "where was I" to the button, and the reader does it and goes on; "pause" stops it. Typed into the box, the same words do the same. Anything else is a question for the book.
Bookmarks and notes: Say "bookmark this" or "note this: remember this for the exam" to the button while listening: the spot is kept with the sentence being read, the reader says Bookmarked or Noted, and the book goes on. The card under the Ask card lists them in book order, each a jump to its sentence, with Remove; a note can be typed there too. They stay in this browser.
Ask: Sends a typed question. Enter in the text box does the same, at any time.

The answer is drawn only from the part of the book you have heard. In a long book, the passages you heard earlier that fit the question go along with the recent text, so "what did it say earlier about the lantern" works chapters later. What the book has not reached yet is not given away; the answer says so and the book goes on.

After a spoken answer

The microphone stays open for a few seconds.

Ask again, straight away.
Say continue to go on with the book.
Say stop to leave the book paused.
Say nothing: a book that was playing goes on by itself; a book that was paused waits for Continue.

Study

Study sheet: Everything you did with a book, as a Markdown file: the questions you asked with their answers, the quiz rounds with your answers and the verdicts, and the bookmarks and notes, chapter by chapter. A question asked more than once is one item with its latest answer; a question the book had not reached is listed apart, without the non-answer. The Study sheet button is on the book's card in the library and in the Bookmarks and notes card of the reader. The file is made here from what this browser kept; nothing is sent anywhere for it.
Flashcards: The same questions and answers as cards, one per line (the front, a tab, the back), in the form flashcard apps import: your questions with their answers, the quiz questions with the feedback and its verdict. Bookmarks make no cards, nor does a quiz question you never answered or one the book had not reached.
Question book: Every question you ask, every quiz question and every bookmark (a note that asks something, or "What did the book say here" with the sentence's opening) joins the book's question book, each on a schedule: a question answered right waits longer before it comes back (a day, three, a week, two, a month), one partly right comes back in half that time, a missed one tomorrow; after three right answers on three different days it is mastered and rests two months. A question asked before the book had reached its answer waits, with no date, until the book covers it. The Question book card in the library lists what is due in each book; the download of the same name is the whole book as a Markdown file, with what is due first.
Revisit: Press Revisit under the bookmarks, or in the library's Question book card, or say "revisit" or "ask me my questions" to the button. The questions due are asked back in the second voice, up to ten at a time, the ones you know least first; answer by voice (a press ends the recording) or type the answer without a microphone. Each answer is graded by the hosted brain against only the part of that chapter you have heard, so nothing ahead of you can slip into the feedback; the verdict is spoken first, and after a miss the book's own answer is read back. "I don't know" is a miss. Say "skip" or "next" to pass one (no penalty), "stop" to end the round; silence lets the question come back tomorrow. A press while a question is read, a seek, a question of your own or Play ends the round. A book that was playing goes on after it.

Keys

Space: Play or pause.
← and →: Back or forward one sentence.
A: The voice action: ask, stop listening, or interrupt, whichever the moment needs.
Esc: Cancel a question or an answer.

You can also tap any sentence in the transcript, or reach it with Tab and press Enter, to read from there.

Earbuds and headsets

One press: The voice action.
Two presses: Skip a sentence.
Three presses: Go back a sentence.

This works in Chrome and Edge, where the page can hear the earbud buttons. If a press opens another app instead, that browser is handing the buttons to its own media player.

The press asks through the laptop's own microphone, not the earbuds': opening a Bluetooth headset's microphone switches the whole headset to its phone-call profile, mono and dull, and a word of the next sound is lost on the way back. When the system's microphone is a headset's, the reader takes the first other microphone it finds and says so once under the Ask card. To use the earbuds' microphone anyway, choose it under Microphone in the reader's Settings (under the Ask card), shown when more than one microphone is found.

Voices

The book and the answers are read in the voice you choose; chapter titles and section headings come in a second voice, so they stand apart from the prose. With a voice of your own, the book alone is in your voice and the second voice takes the titles and the answers. Two voices, never three.

Your own voice

Recording: On the Your voice page, read the three sentences aloud (8 to 15 seconds) in a quiet room, close to the microphone, at the pace you want to be read to. Give the voice a name first.
Local: The recording is turned into a voice in this browser and is never uploaded. What is saved is a small voice profile, under 2 MB, never your audio; you can delete it.
Several voices: Record as many as you like, each under its own name; every one is a choice under Voice in the reader, and each can be deleted on its own.
No sound?: If the page says the microphone gave no sound, check that it is not muted (laptops and headsets often have a mute key or switch), that the system is using the right microphone, and that this site is allowed to use it. A headset or an external microphone that is plugged in is often the one taking the sound: in the reader, when more than one microphone is found, the Microphone list under Settings (in the Ask card) lets you pick another; on the Your voice page the same list sits beside Record.

Settings

Speech-to-text: Which model hears your questions: Accurate (63 MB) or Fast (41 MB). The download is paid once.
Microphone: Under Settings in the Ask card, with Voice and Speech-to-text, once this browser has found more than one microphone (their names are known after the first recording is allowed). Automatic prefers a built-in or wired microphone over a Bluetooth headset's, for the reason under Earbuds and headsets; any other choice is kept for next time, and if that microphone is gone the browser's own choice is used.
Describe the pictures: Under the file box in the library, off by default. On, each picture in an EPUB, a Word file or a paper's web version is sent once to the hosted brain when the book is added, never again, and its description, two to four plain sentences about what it shows, is read where the picture sits, after its caption, in the second voice. Ornaments and icons are passed over by size, and a book's first sixty pictures are described. The choice is made as the book is added: to describe the pictures of a book added without it, remove the book and add it again. Pictures in a PDF are passed over either way, and a picture on another site that the site does not let a browser fetch keeps its caption alone.
Quiz me after each chapter: Off by default. On, the book pauses at the end of each chapter and asks up to three questions on it in the second voice. Answer by voice (the microphone opens by itself, and a press ends the recording) or, without a microphone, type the answer in the box. Each answer is graded against that chapter alone by the hosted brain, which sees the chapter and your answer, and the verdict is spoken first; "I don't know" counts as a miss and gets you the chapter's point. Silence or an empty answer skips the rest of the round; a press while a question is read, a seek, a question of your own or Play cancels it. Chapters the reader passes over are never quizzed. The rounds are kept with the book in this browser. Say "quiz me" to the Ask by voice button at any time for a round on what you have heard of the chapter so far, whether or not this is on.
Keep question audio and transcripts: Off by default. On, each spoken question's recording, transcript and answer are kept in this browser, for checking what was heard against what you said; download or clear them from the Ask card.

Downloads and storage

The voice (93 MB on the processor, 326 MB on the graphics chip) and the speech-to-text model (41 or 63 MB) are downloaded once and kept in this browser's storage, so the next visit starts at once. Where the browser would not keep the larger voice file, the reader takes the smaller one, which stays, and says so under Play. Chrome's setting "Delete data sites have saved to your device when you close all windows" is the usual cause: it keeps about 300 MB for a site. Allowing yapbook.app to save data (Settings, Privacy and security, Third-party cookies, Site data) makes room for the larger file on the graphics chip. A private window keeps nothing, so everything downloads again there.

Not here yet

A formula in a PDF is read as its text layer has it, and a picture in a PDF is passed over: a PDF gives no source for either. Everything else the installed tool does is in the reader now. As things change, this page says so.