How tarun7r/vocal-agent works
Project type and purpose
• This is a real-time, full-stack (audio-to-audio) voice assistant written in Python.
• It captures audio, runs speech-to-text (STT), invokes an LLM agent (with tool calls), and then synthesizes speech (TTS).
• Key features: Whisper + Silero VAD for STT, Agno agent framework with Llama 3.1 via Ollama, Kokoro ONNX for TTS, plus web-search tool integrations.
File structure
• Top-level scripts:
• main.py – orchestrator of the audio pipeline
• agent_client.py – encapsulates LLM tool calls and reasoning
• vocal_agent_mac.sh – setup/install script for macOS
• Models in the root (kokoro-v1.0.onnx, voices-v1.0.bin) indicate TTS assets loaded at runtime
• requirements.txt lists Python dependencies; .gitignore, LICENSE, README.md are ancillary
README for additional insights
• Describes pipeline stages, configuration constants in main.py (sample rate, phoneme length, speed, voice profile)
• Emphasizes real-time, low-latency audio processing and extensible tool system
• Outlines external dependencies (Ollama service, System TTS engine espeak-ng)