feat(livekit-plugins-dataspike): add real-time deepfake detection plugin

[adwells-ds]

Summary

Adds the Dataspike plugin for real-time detection of AI-manipulated (deepfake) content in LiveKit video streams.

The plugin performs video sampling on remote participant tracks, sending selected frames to the Dataspike API for analysis and publishing detection events back into the room or a custom callback.

About Dataspike

Dataspike provides an AI Manipulation Detection API for analyzing visual and audio content in real time.
The service detects signs of synthetic or altered media, including face swaps, compositing, and AI-generated artifacts within live or recorded video streams.

Key Features

• Manipulation Recognition – Detects a wide range of synthetic media techniques, including face swaps and ai-generated overlays.
• AI Model Detection – Identifies content generated or altered by modern AI systems such as Veo, Sana, Flux, and other generative models.
• Adaptive Video Sampling – Samples only selected frames for analysis, maintaining real-time responsiveness while minimizing bandwidth and processing load.
• Realtime Deepfake Alerts – Returns lightweight detection signals via the Dataspike API, enabling instant response to suspected AI manipulation.
• Privacy & Security by Design – No video data or metadata is stored on the plugin or server side.
  All frame transmissions are end-to-end encrypted, and only short-lived analysis signals are returned to the client.

Implementation

• Follows standard LiveKit plugin patterns
• Integrates with AgentSession and rtc.Room APIs
• Implements adaptive video sampling via InputTrack
• Includes internal WebSocket client for Dataspike streaming and result handling

Testing

• Verified end-to-end using valid Dataspike API keys
• Tested with multiple LiveKit rooms and multiple participants
• Confirmed stable reconnection, result delivery, and adaptive sampling performance

Files Changed

Plugin: livekit-plugins/livekit-plugins-dataspike/

• Core plugin: detector.py, __init__.py, schema_pb2.py
• Logging and version metadata

Example: examples/deepfake-detection/dataspike/agent_worker.py

[CLAassistant]

All committers have signed the CLA.