v0.7.0a6

What's Changed

• Update release script to use ty instead of mypy
• Fix production audit blockers: track fire-and-forget tasks, test scoring module
• Migrate examples from deprecated connect/disconnect_video to join/leave
• Add ty type checker to pre-commit hooks
• Fix ty:ignore placement on correct line in _ai_resilience.py
• Replace mypy with ty for type checking
• Refactor all examples to use shared helpers, add console mode and README
• Fix Protocol contract gaps: type erasure, dead declarations, weak annotations
• Add Protocol contracts to all 34 mixins for explicit host interfaces
• Unify voice pipeline: extract VoicePipelineMixin, strip pipeline from backends
• Add build_turn_detector helper for smart-turn ONNX model
• Add VAD model selection: energy, silero, ten via VAD env var
• Fall back to energy VAD when sherpa-onnx is not installed
• Fix SherpaOnnxVADProvider init — takes config object, not kwargs
• Add VAD to realtime Gemini example for local speech detection
• Downgrade Gemini usage log from INFO to DEBUG — too spammy
• Add debug logging to local VAD events for barge-in diagnosis
• Revert _user_speaking audio gate — broke new response playback
• Suppress AI audio during barge-in, show [BARGED] label in console
• Fix user speech streaming, add was_barge_in to transcription events
• Fix threading safety and memory leak in VAD manual mode
• Add manual VAD activity signals to OpenAI, xAI, and mock providers
• Add local VAD manual mode for RealtimeVoiceChannel
• Fire ON_BARGE_IN hook in RealtimeVoiceChannel, enable Gemini VAD
• Reduce top panel height to minimize empty space
• Fix voice state tracking for user speech in console
• Place voice activity timeline beside audio meters
• Add voice activity timeline, barge-in and skills indicators to console
• Add streaming text to console via ON_PARTIAL_TRANSCRIPTION
• Add RoomKit Console — full-screen terminal dashboard for voice agents

Full Changelog: v0.7.0a5...v0.7.0a6