Skip to content
Droid Sentence MinerDroid Sentence Miner

Texts and sentence audio

The Texts tab is the permanent log. It is the GameSentenceMiner text hooker, rebuilt around Android’s replay buffer: every line that auto OCR or a manual scan accepted is stored with the moment it appeared and the moment it was replaced.

The log

Section titled “The log”

Each row shows the line, how long it was on screen, its timestamp and a speaker icon when a clip exists. Tap a line to expand it:

  • the screenshot captured at that moment,
  • a waveform of the voice clip,
  • play, copy, delete and mine actions,
  • the sentence as selectable text.

Newest lines are first. Tap a word to open the lookup popup; the card is linked to the line, so it gets the whole sentence, the screenshot and the clip.

Replay buffer

Section titled “Replay buffer”

Droid Sentence Miner captures playback audio through the same MediaProjection session used for the screen. The last 1–10 minutes (5 by default) are kept in memory as mono 24 kHz PCM.

A line’s clip is [line start − before, min(line end, next line start) + after]:

  • Before and after offsets are configurable in Settings → Sentence audio.
  • If the video is paused when a line is mined, the clip ends at the pause moment instead of running to “now”.
  • The clip is passed through voice-activity trimming (optional) and loudness normalization (on by default).

Waveform editor

Section titled “Waveform editor”

Open the editor from a line before mining to adjust the exact boundaries, or tap ✎ next to + Anki in the lookup popup to trim the sentence and the audio together. Drag the slider’s two handles, preview, and confirm. The waveform is rendered from the stored PCM, so it is the same audio that will be attached to the card.

Silence trimming is marked experimental. It is useful for visual novels with long pauses; turn it off if it clips short exclamations.

Microphone fallback

Section titled “Microphone fallback”

Some apps block playback capture and hand Android pure silence. DroidDeck (Steam games) targets Android 9, and Netflix does not allow audio capture per stream. In those cases:

  1. Enable Settings → Sentence audio → Microphone fallback (or the bubble menu toggle).
  2. Grant the microphone permission.
  3. Keep the speaker on. With headphones the microphone records nothing useful.

The microphone recorder runs alongside playback capture. If device audio is exactly zeros for a short period while something is playing, the microphone fills the buffer. Device audio takes over again as soon as a non-zero chunk arrives.

The microphone chain is: Android noise suppression → 120 Hz high-pass → calibrated speaker- correction EQ → compressor. Noise suppression, EQ and compressor are separate settings, on by default for the microphone. The EQ can be measured on the device (speaker → microphone) from the same settings page. Every finished clip is then normalized to the same loudness.

Storage

Section titled “Storage”

Text log lines, screenshots and clip metadata live in a separate Room database (textlog.db). Dictionaries live in dict.db. The two are independent, so clearing the text log never deletes dictionaries and upgrading dictionaries never touches your history.

Section titled “Related”
  • Anki mining — where audio and screenshots go on the card.
  • Auto OCR — how lines are created.
  • Settings — buffer length, offsets and normalization.