PersonaPlex Serverless Worker

RunPod Serverless worker for real-time voice AI using NVIDIA PersonaPlex.

Features

• Full-duplex speech-to-speech conversation
• WebSocket streaming for low latency
• Persona customization via text prompts
• Voice selection (multiple natural and variety voices)

Deployment

1. Push to GitHub (triggers build via Actions)
2. Add HF_TOKEN secret to GitHub repo settings
3. Create RunPod Serverless endpoint with the image
4. Configure TCP port 8998 exposure

Environment Variables

• HF_TOKEN - HuggingFace token (required for model download)

API

WebSocket endpoint: wss://<worker-ip>:<port>/api/chat?text_prompt=<persona>&voice_prompt=<voice>

Parameters

• text_prompt - Persona description (URL encoded)
• voice_prompt - Voice ID (NATF0-3, NATM0-3, VARF0-4, VARM0-4)