Stream Live Responder
AI-powered live interaction bot that responds to Stream Live comments in real-time using synthesized voice.
Client:
Traf, USA
Project Overview
The client engaged Axis to create an intelligent assistant for Live streams that enables automated, spoken interactions between streamers and their audiences. The bot captures live comments, processes them with a language model, and responds using Text-to-Speech (TTS), creating a seamless, interactive experience. The system was designed to support multilingual audiences (Russian and English) and run reliably during continuous live sessions.
Challenge
Instability of early Python-only implementation due to unhandled websocket exceptions.
Difficulty managing multilingual speech synthesis from Windows registry voices.
Risk of replaying old comments when the bot connected mid-stream.
VPN dependency for access in restricted regions, affecting connection stability.
Tech Stack
AI / LLM Processing: LLaMA models via Together AI API for natural language response generation.
Speech Synthesis: pyttsx3 (Python), leveraging Windows Registry voices for local, low-latency TTS.
Live Chat Parsing: Node.js for reliable parsing and WebSocket management, Python for AI logic and TTS output.
Backend Architecture: Event-driven pipeline with inter-process communication between Node.js and Python.
Configuration: Configurable settings in
config.ini(voice index, system prompt, streamer ID, API keys).Infrastructure: VPN integration to ensure access to Live in restricted geographies.
Solution
Axis delivered a robust, two-layer architecture:
Node.js Layer: Handles stable connection to Live, filtering comments and ensuring only new messages are processed.
Python Layer: Processes text with LLM, generates responses, and converts them into real-time speech via pyttsx3.
Queue Management: Sequentially handles multiple comments, ensuring each is spoken before processing the next.
Configurable Triggers: Initially required a
/botkeyword, later upgraded to respond to all live comments dynamically.
