🎤 Mantri AI — Real-Time Voice Conversation Assistant

Mantri AI is a real-time, speech-to-speech conversational AI assistant that listens, understands, and responds like a human.
It supports English, Hindi, and Hinglish conversations with natural flow, interruption handling, and continuous interaction.

---

🚀 Features

• 🎙️ Real-time Speech Recognition (STT)
• 🧠 AI-powered conversational responses
• 🔊 Text-to-Speech (TTS) replies
• 🌐 Multilingual support (English, Hindi, Hinglish)
• 🔁 Continuous conversation loop
• ✋ Interrupt AI while speaking
• 💬 Chat UI with message history
• 🧠 Short-term conversation memory

---

🏗️ Tech Stack

Frontend

• React.js
• Web Speech API (Speech Recognition + Speech Synthesis)
• Axios

Backend

• FastAPI (Python)
• Groq API (LLaMA 3.3)
• Uvicorn

---

🧠 How It Works

1. User speaks → Speech Recognition converts voice to text
2. Text sent to backend (FastAPI)
3. Groq LLM generates response
4. Response sent back to frontend
5. Speech Synthesis speaks the reply
6. Loop continues for real-time conversation

---

✨ Special Features

• 🗣️ Natural conversation tone (human-like responses)
• 🎯 Language detection (auto Hindi / English / Hinglish)
• 🔁 Smart listening flow (AI speaks → mic restarts)
• ✋ Interrupt system (stop AI mid-speech)

---

⚠️ Important Notes

• .env file is excluded for security
• Never expose API keys in frontend
• Works best on Chrome (Web Speech API support)

---

🚀 Future Improvements

• Better natural voice (advanced TTS integration)
• Long-term memory system
• User authentication
• Deployment optimization (Render + Netlify)
• Mobile support

---

👨‍💻 Author

Sukhad Sharma
BTech CSE (Full Stack Software Development)

---

⭐ Show Your Support

If you like this project, give it a ⭐ on GitHub!