S26: Add Stronger Guardrails to the AI Voice Bot

[notzabir]

Description

Decibel-Based Voice Detection

Basic RMS detection is okay.
Explore WebRTC

• VAD (highly optimized)
• Silero VAD (PyTorch-based)
• Pyannote (speaker diarization)

Advanced Upgrade: Interrupt Detection

• If user speaks while bot talking: Immediately stop TTS
• Switch to listening That makes it feel intelligent.

B. Logging + Telemetry System Track:

• Latency
• Memory usage
• Crash events
• Conversation length

Store metrics for optimization.

C. Fail-Safe Mode
If:

• CV crashes
• LLM times out
• RAM overload

System switches to:

• Basic fallback voice assistant mode

WebRTC VAD

• WebRTC Voice Activity Detection in Python
  https://www.youtube.com/watch?v=Gz7uJv1uY7Q

Silero VAD (PyTorch)

• Silero VAD Real-Time Implementation
  https://www.youtube.com/watch?v=ZfY0M9wE7C8
  These directly help