GConnect v2.3.0

What's New in v2.3.0

Bidirectional Audio: Real-time microphone capture streamed to Docker container, and container audio output streamed back to client speakers.
Virtual ALSA Device: Standard Python libraries (speech_recognition, sounddevice, pyaudio) detect default virtual mic and speaker with zero code modification.
Google Meet-Style Controls: Header [🎙️ Audio] button with input/output device selectors and mute/unmute toggles.
Audio Router: Port 8001 WebSocket hub with 50ms jitter buffer, duplex echo ducking gate (-18dB attenuation during playback), and single active speaker arbitration.
Verification Utility: Standalone voice_test.py deployed at workspace root supporting 3-second recording and harmonic confirmation chime synthesis.

Dynamic Reconnect: GConnect automatically detects when Docker Desktop is launched after GConnect is already running without needing an app restart.
Physical GPU Validation: Directly queries pynvml, torch.cuda, and nvidia-smi to verify physical NVIDIA adapters before issuing Docker DeviceRequest / --gpus all, eliminating nvidia-container-cli: no adapters were found 500 errors on systems without NVIDIA cards.

Thread-Safe Connection Monitor: Real-time monitoring of clients and remote teammates connected to the GPU host.
Physical Device Grouping: Groups connections by client ID to prevent multiple tabs from inflating device counts.

Awaits Backend Health: Toggling sharing mode properly polls backend_info until sidecar is ready, preventing UI freeze.
Port 8000 Cooldown: Added 800ms cooldown for Windows socket release during mode transitions.
Token Persistence: Reuses existing auth_token across restarts.