- Add retry logic for socket file removal and listener creation
- Optimize message writing by combining header and data writes
- Move socket paths from temp dir to /var/run
- Refactor OPUS parameter lookup to use map for better readability
- Simplify validation functions for better performance in hotpaths
Update congestion threshold multiplier and CPU thresholds to better suit single-core ARM RV1106G3 processor characteristics. Adjust memory thresholds for systems with 200MB total memory.
- Implement graceful degradation for congestion handling with configurable thresholds
- Refactor audio relay track updates to be async to prevent deadlocks
- Add timeout-based supervisor stop during quality changes
- Optimize buffer pool configuration and cleanup strategies
Significantly increase message pool, channel buffer, and adaptive buffer sizes to better handle quality change bursts. Adjust timeouts and intervals for improved responsiveness.
This change replaces all instances of GetConfig() function calls with direct access to the Config variable throughout the audio package. The modification improves performance by eliminating function call overhead and simplifies the codebase by removing unnecessary indirection.
The commit also includes minor optimizations in validation logic and connection handling, while maintaining all existing functionality. Error handling remains robust with appropriate fallbacks when config values are not available.
Additional improvements include:
- Enhanced connection health monitoring in UnifiedAudioClient
- Optimized validation functions using cached config values
- Reduced memory allocations in hot paths
- Improved error recovery during quality changes
- Add server stats reset and frame drop recovery functions
- Implement global audio server instance management
- Add WebRTC audio track replacement capability
- Improve audio relay initialization with retry logic
- Enhance quality change handling with adaptive buffer management
- Add global helper functions for audio quality control
- Replace mutex-protected refCount with atomic operations in ZeroCopyFramePool
- Implement chunk-based allocation in AudioBufferPool to reduce allocations
- Add proper reference counting with atomic operations in ZeroCopyAudioFrame
- Optimize buffer pool sizing based on buffer size
Replace mutex-protected refCount operations with atomic operations to improve performance in concurrent scenarios.
Simplify frame release logic and add hitCount metric for pool usage tracking.
Move common supervision loop logic to BaseSupervisor with configurable parameters
Simplify input/output supervisor implementations by using base template
Update function comments to be more concise
Consolidate duplicate channel and process management code from input/output supervisors into BaseSupervisor
Add new methods for channel initialization and cleanup
Standardize process termination and monitoring behavior
- Remove unused setRunning method from BaseSupervisor
- Refactor IPC input reader to use running flag and mutex
- Add atomic state management to InputSupervisor
- Implement proper channel cleanup and process termination
- Improve error handling and logging throughout
- Replace MuteMicrophone calls with StartMicrophone/StopMicrophone for clearer behavior
- Update microphone state broadcasting to reflect actual subprocess status
- Modify UI to use enable/disable terminology instead of mute/unmute
- Ensure microphone device changes properly restart the active microphone
- Implement new POST /microphone/stop endpoint
- Refactor mute handling to properly start/stop audio processes
- Add callback mechanism for audio relay to reconnect to current session
- Simplify UI microphone controls by combining mute/start-stop functionality
- Increase goroutine cache size from 4 to 8 buffers for better hit rates
- Add adaptive resize and cache warmup based on usage patterns
- Implement enhanced cleanup with size limits and better TTL management
- Optimize buffer clearing and preallocation strategies
Update config field names to better reflect their specific usage contexts in adaptive buffer and optimizer components. This improves code maintainability by making the purpose of each latency target more explicit.
The pre-warming feature was removed to simplify the audio input supervisor implementation. This feature added complexity and was not providing significant latency improvements in practice.
Alex P
Aveline