# RECCoon TODO

## Build / distribution

- [ ] **The sherpa-onnx binaries are too large to keep in git.** The Android
      AAR is ~50 MB and the onnxruntime `.so` files push the APK to ~72 MB. The
      AAR and the built APK are currently gitignored. Decide how to distribute
      them later: Git LFS, a small Maven repository, a Gradle task that
      downloads the AAR on first build, or Play Asset/Feature Delivery for the
      Whisper models.

## Requested

- [x] **Separate left/right volume meters** — done: stereo capture with a
      mono fallback, plus a two-channel peak meter (`LevelMeterView`).
- [x] **Mono switch** — done: the **Mono** checkbox records a single channel.
- [x] **Option to disable notifications and vibrations while recording** — done
      via Do Not Disturb total silence (`ACCESS_NOTIFICATION_POLICY`), restored
      on stop. Persisted as a checkbox.
- [x] **Live waveform** — done: scrolling `WaveformView` with red clipped
      columns.
- [x] **WAV markers** — done: the **Mark** button embeds `cue `/`adtl` chunks in
      the WAV and stores a JSON sidecar; the player shows marker chips.

## Backlog (from earlier discussion)

- [ ] Export transcripts as `.txt` / `.srt` and/or embed them in WAV metadata
      (`LIST`/`INFO`, `bext`).
- [ ] Foreground service so recording continues when the app is in the
      background.
- [ ] Library actions: delete, rename, share, "open with".
- [ ] Word-level timestamps and tap-to-seek inside a transcript segment.
- [ ] Read embedded markers back from the WAV when there is no JSON sidecar.
- [ ] Add a JUnit regression test for `StreamingResampler` (it could throw
      `IllegalArgumentException: <n> > <n-1>` when the trim overshot the buffer).