The Flow Engine
How your audio is made
SpokenText Reader generates your audio as you listen, always a few sentences ahead, so the flow never has to stop and catch up.
What’s actually happening
1. Your text is split into natural chunks
Roughly 130–200 characters each, snapped to sentence boundaries so a chunk never cuts off mid-thought.
2. Each chunk is synthesized moments before you reach it
Not the whole book at once. Just a short rolling window, kept a few chunks ahead of wherever you're actually listening.
3. Playback always trails the buffer, never the synthesis
By the time you reach a sentence, its audio has already been sitting ready for a moment, so there's never a gap to wait through.
The buffer, visualized
A short rolling window of chunks is always kept ready ahead of playback. The window moves with you; it never needs to catch up because it was never behind.
Diagram: text is split into chunks. A playhead moves through already spoken chunks, the chunk currently playing, a short rolling window of chunks already synthesized and waiting, and chunks not yet generated further ahead.
- Already spoken
- Playing now
- Synthesized, waiting
- Not yet generated
Motion is reduced because your system prefers reduced motion. The diagram above shows a single still frame of the buffer.
Why we built it this way
I lived through the era when the only option was pre-recorded audio. Every reading tool I tried back then had the same problem: the moment you wanted to keep something, you had to stop. Switch to some other screen, find the passage again, drag a selection until you’d caught the right words, and by the time you had, you’d often forgotten why you wanted it. The flow was gone every single time.
Real-time generation exists so that keeping something does not mean stopping. There are three ways to keep it, and audio keeps playing while you do.
- Bookmark. One quick tap saves your place.
- Note. Tap the note button at any time. It saves a note stamped with the current time code. For example: “this whole paragraph is good, come back to it and read it more later.”
- Clip note. Mark Clip In and Clip Out, then save. Use the Save button, or long press the Gesture Pad to open a note. This captures a chunk of text, for example a passage you want to cite in a paper later. From the Notes screen you can play that passage back. The same passage is highlighted in the Text view.
Marking Clip In and Clip Out does not save anything. A clip is saved when you save it.
Key facts about how audio is generated
- characters per spoken chunk
- 130–200
- generated on your device
- 100%
- pre-recorded audio files
- 0
characters per spoken chunk
generated on your device
pre-recorded audio files