🎙️ Sarvam AI Voice Bot

Real-time voice AI bot powered by Sarvam AI and Pipecat framework. Supports phone calls via Twilio with multilingual support.

---

🚀 Quick Start

# 1. Activate virtual environment
venv\Scripts\activate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run the Twilio server
python twilio_server.py

---

✨ Features

• 🎤 Speech-to-Text - Sarvam AI STT (saarika:v2.5)
• 🧠 Language Model - Sarvam AI LLM (sarvam-m)
• 🔊 Text-to-Speech - Sarvam AI TTS (bulbul:v2)
• 🌐 Multilingual - Telugu, Hindi, English, Gujarati
• 📞 Twilio Integration - Phone call support with IVR
• 🎯 Language Selection - User chooses language via IVR menu
• ⚡ Real-time - Low latency responses (2-4 seconds)

---

📁 Project Structure

AI_Voice_Agent/
├── twilio_server.py          # Main Twilio server
├── sarvam_ai.py              # Sarvam AI API integration
├── audio_utils.py            # Audio format conversion
├── requirements.txt          # Dependencies
├── .env                      # Configuration
├── docs/                     # Documentation
│   ├── LANGUAGE_SELECTION.md # IVR language menu
│   ├── AUDIO_CONVERSION.md   # Audio format details
│   ├── ARCHITECTURE.md       # System architecture
│   └── TROUBLESHOOTING.md    # Common issues
└── README.md                 # This file

---

⚙️ Configuration

Edit .env:

# Sarvam AI (Required)
SARVAM_API_KEY=your_api_key_here

# Twilio (Required)
TWILIO_ACCOUNT_SID=ACxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
TWILIO_AUTH_TOKEN=your_auth_token
TWILIO_PHONE_NUMBER=+1234567890

---

🌐 Language Support

Users select their language via IVR menu:

• Telugu (te-IN) - Press 1
• Hindi (hi-IN) - Press 2
• English (en-IN) - Press 3
• Gujarati (gu-IN) - Press 4

See docs/LANGUAGE_SELECTION.md for details.

---

📊 Performance

• STT Latency: 1-2 seconds
• LLM Latency: 0.4-0.8 seconds
• TTS Latency: 0.6-1.4 seconds
• Total Response: 2-4 seconds

---

📚 Documentation

Document	Description
LANGUAGE_SELECTION.md	IVR language menu flow
AUDIO_CONVERSION.md	Audio format conversion details
ARCHITECTURE.md	System architecture and flow
TROUBLESHOOTING.md	Common issues and solutions

---

🎓 How It Works

User Calls → IVR Menu → Language Selection → Voice Conversation
                                              ↓
                                    STT → LLM → TTS
                                    ↓     ↓     ↓
                              Sarvam AI Services

---

🐛 Troubleshooting

See docs/TROUBLESHOOTING.md for detailed solutions.

Quick fixes:

• Server won't start: Check Python 3.10+, activate venv
• No audio: Check Twilio webhook URL configuration
• API errors: Verify Sarvam API key in .env

---

📞 Deployment

1. Deploy to Render/Railway/Heroku
2. Configure Twilio webhook: https://your-domain.com/voice/incoming
3. Test by calling your Twilio number

---

Built with ❤️ using Sarvam AI