VDSA: Voice Dangerously Smart Agent

Real-time speech processing system with dynamic LangGraph agent responses using Kafka communication.

Features

• Real-time Speech Recognition
  • Voice Activity Detection (VAD)
  • Multi-device microphone support
  • Configurable audio processing
• VDSA Agent System
  • LangGraph workflow orchestration
  • Code generation & execution
  • Docker-based sandboxing
• Distributed Architecture
  • Kafka message streaming
  • Modular microservices
  • Horizontal scalability

Prerequisites

• Conda (Miniconda/Anaconda)
• Python 3.10
• Local Kafka instance
• NVIDIA GPU (recommended) or Apple Silicon
• Microphone

Installation

1. Clone Repository

   git clone https://github.com/Teachings/vdsa.git
   cd vdsa

2. Create Conda Environment

   conda create -n vdsa python=3.10
   conda activate vdsa

3. Install Dependencies

   pip install -r requirements.txt

4. Start Kafka

   docker-compose up -d

Configuration

config.yml:

device: "cuda"  # "mps" or "cpu"
voice-model: "large-v3-turbo"
kafka:
  broker: "localhost:9092"
  topic_vdsa: "vdsa.agent.requests"
  topic_response: "vdsa.agent.response"

Usage

Unified Launcher

conda activate vdsa
python launcher.py

Manual Execution

Terminal 1 - Speech Recognition:

conda activate vdsa
python stt/main.py

Terminal 2 - Agent System:

conda activate vdsa
python langgraph_dynamic_agent/dsa.py

Architecture

┌────────────┐       ┌─────────────┐
│ STT Engine │       │ VDSA Agent  │
└─────┬──────┘ Kafka └──────┬──────┘
      │                     │
      ▼                     ▼
 Microphone Input      Agent Responses

Logging

• Launcher:

  • logs/stt.log
  • logs/vdsa.log

• Manual Mode:

  tail -f stt/debug.log
  tail -f langgraph_dynamic_agent/agent.log