Release Notes
Version-specific changes for the Klleon Ondevice SDK.
v0.5.0 (Latest)
Release date: April 2026
Breaking Changes
pauseVideo()/resumeVideo()/isVideoPaused()-> Renamed toenterStandby()/exitStandby()/isStandby()registerKeywords()/clearKeywords()/keywordDetected-> Renamed toregisterWakewords()/clearWakewords()/wakewordDetected
New Features
- Standby mode: Full pipeline flush + timing reinitialization to prevent fast-forward/ANR
- Wake-word matching: Detect registered keywords from STT recognition results -> Deliver callbacks (active only during standby)
setIdleTimeout()/disableIdleTimeout(): Automatic standby entry after inactivity following TTS_ENDisResponding: StateFlow<Boolean>: Track avatar response state (PREPARING_RESPONSE ~ TTS_END)STANDBY_ENTERevent: Emitted via sdkState on automatic standby entry
Bug Fixes
- VideoDecoder
stop()race condition (pauseToken/stopToken order inversion) - ASYNC mode consumer not checking pauseToken
- Added AudioPlayer
pause()/resume()(fixed issue where only video paused while audio continued playing)
v0.4.0
Release date: April 2026
Breaking Changes
play(),finish()-> Changed tosuspend fun(must be called within a coroutine)- Removed
KlleonOndeviceSdk.stoppingcompanion property -> Replaced withisStoppinginstance property
Changes
- Removed
BaseElement.stop()runBlocking:withTimeout(2s)+ forced cancel fallback - Introduced
stopMutex: Prevents concurrentstop()entry (prevents ANR/deadlock/double release) - Added
awaitFinish(): Prevents resource contention during quick re-entry (finish -> play) - Removed unused
bridgeJobcode
v0.3.6
Release date: April 2026
Changes
- Added
avatarIdparameter toResourceManager.init() - Separated YUV buffer lifecycle: Buffer reuse on same avatar re-entry (0ms)
- 3-second cooldown on avatar switch (OOM defense for low-end TV SoCs)
- Buffer retention on
finish()(resetState) + complete release on app exit (deinit) - Automatic background memory release based on
onTrimMemory/onLowMemory - IO coroutine transition within
init()(removed Main thread blocking)
v0.3.5
Release date: April 2026
Changes
- Implemented RAW YUV video decoding pipeline (MP4 -> YUV pre-conversion)
- Automatic NV12/I420 color format detection and conversion
- Activity lifecycle-based native memory release
- Added Android memory monitoring skill
v0.3.4
Release date: March 2026
Changes
- Disabled StyleEncoder runtime inference (skip model load when .style is pre-loaded)
- Restored SdkInitializer.validate() call
- Improved sample app resource download flow
v0.3.3
Release date: March 2026
Changes
- Pre-computed .style file loading (added StyleFileLoader)
- Deleted style_encoder_float32.bin asset (~35MB AAR size reduction)
- Internalized play() parameters (removed ttsModelUse, mmapEnabled, tokenServerUrl)
- Added WebSocket voice signals (SPEECH_START, SPEECH_END, STOP_RESPONSE_MESSAGE)
- ResourceManager .style file download support
v0.3.2
Release date: March 2026
Changes
- Changed Oboe input format from Float to I16 (improved USB microphone compatibility)
- Added I16 -> Float32 runtime conversion logic
- publish.sh support for per-STT-provider deployment
v0.3.1
Release date: March 2026
Changes
- Added ALWAYS_ON voice mode (auto-starts after play(), session maintained)
- Removed CONTINUOUS mode and cleaned up dead code
- Removed unnecessary Azure STT START_VOICE/END_VOICE signals
v0.2.18
Release date: March 2026
Changes
- Azure STT token server integration
v0.2.17
Release date: March 2026
New Features
- Added voice recognition status notification (
STT_STATUS) -- Notifies app of token waiting/reconnecting/recovery complete states
Bug Fixes
- Fixed audio input stream open failure on some devices
- Added
mmapEnabledparameter toplay()function
v0.2.16
Improvements
- Added STT interim results (
STT_PARTIAL) support - Changed SDK distribution to bundled AAR (unified dependencies)
v0.2.15
New Features
- Voice recognition improvements
- Automatic voice recognition pause during avatar speech
v0.2.14
Improvements
- Improved USB microphone compatibility (MMAP disabled by default)
v0.2.13
Improvements
- GPU compatibility fixes
v0.2.12
New Features
- Automatic silence detection
- Low-latency audio input
- Automatic USB microphone detection and switching
v0.2.11
Bug Fixes
- Fixed staircase artifact during screen rotation rendering
v0.2.10
New Features
sendStopSpeech(): Avatar speech interruptionsendChangeLanguage(): Language change during conversation
v0.2.9
New Features
- Multi-language support (Korean/Japanese/English)
- Android TV UI improvements
v0.2.8
New Features
- Android TV 270-degree rotation support
- Dynamic
characterIdpassing per character
v0.2.7
Improvements
- Improved USB microphone device selection
v0.2.6
New Features
- Added
sendChangeLanguage()API - Improved STT result handling
v0.2.5
New Features
- Low-latency microphone recording
- Automatic USB microphone switching
- Silence detection
- WebSocket binary transmission
v0.2.4
New Features
play()API change: Switched tosdkKey,characterId,languagebased
Bug Fixes
- Sony TV / Android 8.1 compatibility fixes
- Improved long-term usage stability
v0.2.3
Release date: December 2025
Improvements
- Security enhancements
v0.2.2
Release date: November 2025
Improvements
- AI inference performance optimization
v0.2.1
Release date: October 2025
Bug Fixes
- Improved server connection stability
- Enhanced WebSocket reconnection logic
v0.2.0
Release date: September 2025
Improvements
- Enhanced model security
- Video processing performance optimization
- Improved GPU acceleration performance
v0.1.x (Initial Versions)
Core Features
- AI-powered real-time video processing
- Lip Sync functionality
- WebSocket real-time communication
Supported Environments
- Android 8.0+ (API 26)
- arm64-v8a, armeabi-v7a
Supported Versions
| Version | Status | Notes |
|---|---|---|
| 0.5.x | Active support | Latest |
| 0.4.x | Active support | |
| 0.3.x | Active support | |
| 0.2.x | Maintenance | |
| 0.1.x | End of support | Ended Q1 2025 |
We recommend using the latest version.