v1.2.4

v1.2.4 — Notification UX, thread safety, and memory fixes

Notification improvements, thread safety hardening, and memory optimizations.

What's changed

• Boot notification UX — Tap the "start services" notification to start them instantly (invisible trampoline activity), no app screen opens. Auto-cancels on tap, lowered to IMPORTANCE_LOW
• Service notification — Now swipeable (IMPORTANCE_MIN, no stop button). Service keeps running in the background as intended
• Thread safety — ReadWriteLock protects native sherpa-onnx resources from concurrent transcribe/release. ConcurrentHashMap for FileObserver debounce jobs
• Memory — Replace boxed MutableList<Short> with ShortArray chunks in AudioRecorder and AudioDecoder, avoiding ~8x heap waste during recording
• Text injection fix — Remove aggressive fallback that could inject text into wrong field
• Service state tracking — AudioMonitorService uses static isRunning flag instead of optimistic local boolean
• Dark/light mode — Recording overlay adapts background and text colors to system theme
• Lazy OkHttpClient — Only created when model download is triggered
• i18n — Extract remaining hardcoded English strings to resources

For the full feature list, see the v1.0.0 release.

v1.2.3

v1.2.3 — Code quality and stability

Threading fixes, deprecation cleanup, and improved robustness.

What's changed

• Thread safety — Use AtomicBoolean for recording state flags, @Volatile for singleton instance, and CopyOnWriteArrayList for file observers
• Deprecation fixes — Replace deprecated onBackPressed(), remove AccessibilityNodeInfo.recycle()/obtain() calls (auto-managed since API 26), add explicit foreground service type for Android 14+
• Resource management — Fix mutex scoping, properly cancel debounce jobs, use response.use{} for HTTP streams
• Async handling — Replace Handler.postDelayed with coroutines, fix race conditions in recognizer initialization
• Cleanup — Remove non-functional language setting, dismiss audio notification when closing transcription

For the full feature list, see the v1.0.0 release.

v1.2.2

v1.2.2 — Android 15 folder monitor fix

Fixes the 6-hour timeout for folder monitoring on Android 15+.

What's changed

• Fix folder monitor timeout — Switch from dataSync to specialUse foreground service type, which has no time limit on Android 15+
• Code cleanup — Remove dead code branches for API levels below minSdk 26
• Documentation — Add Android 14/15/16 FGS restrictions reference to BootCompatHelper

Android version support

Android	Behavior
8.0–13	Services start directly at boot
14+	Notification prompts user to open app
15+	Folder monitor now runs indefinitely (was limited to 6 hours)

For the full feature list, see the v1.0.0 release.

v1.2.1

v1.2.1 — Android 14+ boot fix

Fixes service startup on Android 14+ (API 34+) where foreground services are blocked from starting at boot.

What's changed

• Harden service startup — Catch ForegroundServiceStartNotAllowedException, SecurityException, and IllegalStateException when starting foreground services
• Restart notification — When floating mic or folder monitor fails to start at boot, a notification prompts you to open the app (which restarts the service automatically)
• Notification permission — Request POST_NOTIFICATIONS permission on Android 13+ when enabling services

For the full feature list, see the v1.0.0 release.

v1.2.0

v1.2.0 — Local model import

Load the speech model from a local folder instead of downloading it over the internet. Useful for users on metered connections or devices without internet access.

What's new

• Import from folder — Pick a folder containing the model files and copy them directly into the app
• Separate download and import — Clear UI with "Download model" (from HuggingFace) and "Import from folder" options
• Copy progress indicator — Shows "Copying to app storage" with progress during local import
• User-friendly error messages — Checksum mismatches, missing files, and network errors now show localized messages instead of raw exceptions
• Model file validation — App re-checks model files on resume and before loading, recovering gracefully if files were deleted externally
• NVIDIA attribution — Added CC-BY-4.0 attribution for Parakeet TDT model

Bug fixes

• Wrap browser intent in try-catch for devices without a browser app
• Prevent UI from getting stuck if model files are deleted outside the app

For the full feature list, see the v1.0.0 release.

v1.1.1

v1.1.1 — Boot stability fix

Harden boot-time code paths to prevent crash after system update.

Changes

• Catch UnsatisfiedLinkError (JNI native library loading) in speech recognizer initialization
• Guard foreground service startup in audio monitor against IllegalStateException on Android 12+
• Prevent Application.onCreate() crash if background services fail to start at boot

For the full feature list, see the v1.0.0 release.

v1.1.0

v1.1.0 — Localization

UI localization for 23 languages. All hardcoded English strings have been extracted to Android resources and translated.

What's new

• 23 language translations — Bulgarian, Croatian, Czech, Danish, Dutch, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish, Swedish, Ukrainian
• Store listings in German, French, Spanish, Italian, and Portuguese
• Localization note in README inviting translation contributions
• Fixed language count in descriptions (25, not 26+)

Notes

• Translations were machine-generated. Pull requests to improve phrasing are welcome.
• German uses neutral/passive language; French, Italian, and Spanish use informal "tu" form (standard for mobile apps).
• The app language follows your Android system language setting.

For the full feature list, see the v1.0.0 release.

v1.0.2

What's Changed

• Fix floating button crash when overlay permission is missing
• Overlay permission setting now always visible with tap-to-manage shortcut
• Folder monitor automatically restarts when adding or removing watched folders

See v1.0.0 release notes for full feature list.

v1.0.1

What's Changed

• Add accessibility service disclosure dialog for Google Play compliance
• Fix model download source in privacy policy (HuggingFace, not GitHub)

See v1.0.0 release notes for full feature list.

v1.0.0

TranSlander v1.0.0

Offline voice typing for Android — your voice never leaves your device.

Speak into your phone and text appears in any app. No internet required, no cloud processing, no tracking.

Highlights

• Fully offline — Uses Parakeet neural model running locally via sherpa-onnx
• Works everywhere — Injects text into any app via Accessibility API
• Multiple input methods — Floating mic, accessibility button, or keyboard integration
• Voice message transcription — Share audio files or monitor folders for automatic transcription
• 25 languages — With auto-detection
• Privacy first — Zero telemetry, zero analytics, zero cloud

Getting Started

1. Install the APK
2. Download the speech model (~600MB, one-time)
3. Grant accessibility permission
4. Tap the mic and speak

Downloads

File	Description
translander-v1.0.0.apk	Install directly on your device

Notes

• First launch requires downloading the ~600MB speech model
• After download, you can revoke network permission for complete offline operation
• Requires Android 8.0+ (API 26)

---

This is the first public release. Feedback and bug reports welcome!

Full Changelog: v1.0.0-alpha3...v1.0.0