Skip to content
Droid Sentence MinerDroid Sentence Miner

Auto OCR

Auto OCR is the main mode. It reads a region of the screen continuously, waits for each new line to settle, and logs it with the moment it appeared, the moment it disappeared, a screenshot and the matching audio.

Start a session

Section titled “Start a session”
  1. Open the game or video and let it play.
  2. Long-press the floating bubble → Start auto OCR.
  3. Choose A single app in Android’s share sheet and select the app you want to read. On older Android versions, share the entire screen.
  4. Draw a text area if there is none for the current orientation.

The app reads new frames only. A static screen is not re-read, so a paused visual novel does not fill the log with duplicates.

Text areas

Section titled “Text areas”

Text areas are rectangles stored as fractions of the screen. They are kept per orientation: a portrait box drawn for a phone game points at the wrong pixels when a video goes fullscreen landscape. Draw one area for each orientation you use.

  • Drag inside the box to move it; drag a corner to resize it; tap × to delete.
  • Keep the area on the dialogue box or subtitle band. Buttons inside the box become text.
  • Multiple areas are allowed: a dialogue box plus a name box, or two subtitle regions.
  • Open the editor from the ✎ button in the Live text window or from the bubble menu.

How a line is accepted

Section titled “How a line is accepted”

The app does not log every OCR read. It tracks each line and requires repeated agreement:

  1. New frame differs from the last frame OCR’d (cell-by-cell comparison, so one typewriter glyph is enough).
  2. OCR reads the text area with meikiocr or ML Kit.
  3. The line tracker keeps the reads and waits for the text to stabilise.
  4. At 3, 5, 8 and 13 reads — and when the line disappears — the consensus of all reads is stored.
  5. Lines below the confidence threshold are dropped when a stronger line exists.

This handles fades, moving backgrounds, typewriter text and OCR noise. The stored text is the medoid of the reads, not just the last frame.

App focus

Section titled “App focus”

With A single app, auto OCR pauses while another app is in front, while Droid Sentence Miner itself is open, and while the popup or area editor is showing. That prevents the app from reading its own overlays or the home screen.

On Android 14+, onCapturedContentVisibilityChanged tells the app when the chosen app is no longer visible.

Live text

Section titled “Live text”

The floating Live text window shows new lines in real time without leaving the game. Tap a line to look up and mine it. See Live text window for the three modes and resizing.

Stop and restart

Section titled “Stop and restart”

Stop auto OCR from the bubble menu or from the notification. If screen capture ends because the captured app was closed or consent was revoked, the bubble menu offers Restart screen capture.

Choosing an OCR engine

Section titled “Choosing an OCR engine”

meikiocr is the default after the model download. It is a text-line detector plus a character recognizer trained on Japanese game text. ML Kit remains available as a fallback and alternative in Settings → OCR.

The app uses one shared OCR instance; auto OCR and manual lookups serialize their work. Benchmarks should be run with auto OCR stopped.

Section titled “Related”