Listening to a Book With the Screen Off
Read-aloud that stops when the screen locks is not read-aloud. It is a demo. Anyone using it while walking, driving or washing up needs it to behave like a podcast: keep going, show up on the lock screen, pause when something else needs the audio.
What makes it behave like a podcast
The speech is handled by Android’s media service rather than by the reading screen. That is the whole difference. It means the phone treats the book the way it treats a music player: playback continues with the display off, and a media notification appears with play, pause and stop, showing the book title so you know which one is running.
Practically, that gives you three things:
- The screen turns off and the reading continues.
- You can pause from the lock screen or a headset button without unlocking.
- The book title shows in the notification, which matters more than it sounds when you have two books on the go.
Where synthetic narration is genuinely good
- Second passes. Re-reading something you already know, at 1.4×, is faster than re-reading it with your eyes and holds up fine.
- Anything procedural. Manuals, documentation, reports — text that carries information rather than voice.
- When your eyes are done. Late at night, after a screen-heavy day, or with a headache, listening is available when reading is not.
- Non-fiction generally. The prose is usually built to be paraphrasable, so a flat delivery costs it little.
Where it falls down
Names and unfamiliar words. A speech engine guesses pronunciation from spelling. Character names, place names and technical terms come out wrong, sometimes differently on each occurrence, and the wrongness is distracting in a way that a human narrator’s choice never is.
Anything where the voice is the point. Poetry, dialogue-heavy literary fiction, comic timing. The engine has no idea which word in the sentence carries the weight, so it puts the stress in the same place every time.
Languages with reading ambiguity. Japanese is the sharp case: the same characters can be read several ways depending on context, and a speech engine will pick one confidently and be wrong. Names are worst.
Anything without a text layer. A scanned PDF has no words to read, only pictures of words. That produces silence rather than an error message, and the same missing text layer also breaks search and reflow in that file.
Speed, and the mistake everyone makes
Do not start at double speed. Comprehension collapses well before the speech becomes unintelligible, and the failure is quiet — you will feel like you are following and retain nothing.
Start at about 1.2×, sit there for a book, then go up. The ceiling is different for fiction and for dense non-fiction, and it is lower than you think for anything you intend to remember.
An honest comparison
A professionally narrated audiobook is better than this, in every way that involves performance. If a book you love exists as one, buy that instead.
Synthetic speech wins on availability: it works on the book you already have, in the format you already have, including the report your colleague sent this morning. That is a different job, and it does it well.
In Aurora Reader, read-aloud is a Premium feature. The rest of the reading — six formats, themes, typography, bookmarks, highlights — is not.