Skip to content

About

An AI-powered voice command component using OpenAI Whisper + GPT-4. Built with Next.js, Tailwind CSS, and TypeScript. Record voice, transcribe it, and get smart AI replies — built for EmpireUI and submitted for Vibe-a-thon 1.0.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

🎙️ Voice Command Button – EmpireUI Component

An AI-powered microphone component that listens to your voice, transcribes it using OpenAI Whisper, and responds with GPT-4. Built with 💡 simplicity, 🔐 security, and 🎨 modern UI.

🧠 Created for the Vibe-a-thon 1.0 by Geek Room and contributes to the EmpireUI open-source library.


✨ Features

  • 🎤 Mic button with pulsing animation and state transitions
  • 🧠 Transcription via OpenAI Whisper (voice-to-text)
  • 🤖 Smart AI response via OpenAI GPT-4
  • 🔐 API key secured with environment variables
  • 🎨 Tailwind CSS + TypeScript + Next.js
  • 💬 Live feedback, loading states, and error handling

🚀 Live Demo

🧪 Visit the demo page

Or run locally:
http://localhost:3000/demo


📁 Project Structure

empire-voice-button/ ├── components/ │ └── ui/ │ └── VoiceCommandButton.tsx # Main mic component ├── pages/ │ ├── api/ │ │ └── voice/ │ │ └── route.ts # API route: Whisper + GPT-4 │ └── demo.tsx # Usage demo ├── lib/ │ └── openai.ts # OpenAI config using env key ├── styles/ │ └── globals.css # Tailwind directives ├── .env.local # Your OpenAI key (DO NOT COMMIT) ├── .env.example # Shared env template ├── .gitignore # Prevents secret + build files ├── package.json # Dependencies and scripts └── README.md # This file


🔧 Installation & Setup

  1. Install dependencies:

npm install Create .env.local file in the root and paste:

OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxx (Replace with your actual OpenAI API key)

Run locally:

npm run dev ➡ Go to http://localhost:3000/demo

📦 How It Works User clicks the mic button → voice is recorded

Audio is sent to pages/api/voice/route.ts securely

Whisper transcribes the audio → GPT-4 generates a response

The UI displays both transcript and AI reply

🔐 Security Notes API key is never exposed to frontend

Audio files processed securely via backend

.env.local is ignored in Git

Temporary files deleted after processing

💡 Sample Voice Prompts “Summarize this document.”

“Tell me a joke.”

“Translate this sentence to Hindi.”

🧠 Tech Stack Frontend: React, Next.js, Tailwind CSS

Backend: Next.js API Routes

AI: OpenAI Whisper (whisper-1) + GPT-4

Language: TypeScript

📮 Submission Instructions ✅ GitHub: github.com/your-username/empire-voice-button

✅ EmpireUI PR includes:

VoiceCommandButton.tsx

API route (voice/route.ts)

README, .env.example, and .gitignore

🙌 Author & Credits Made with ❤️ by Tanmay Shinde For Geek Room Vibe-a-thon 1.0

🎧 Code the vibe. Build the future.


📌 Reminder

  • ✅ Paste this into a file named README.md in your root project folder.
  • ✅ Update the Vercel demo link and GitHub link before final submission.
  • ✅ Push to GitHub or submit in your ZIP/Pull Request.

Let me know if you want the .env.example too — or want help submitting the GitHub PR. You're fully ready now 🔥

About

An AI-powered voice command component using OpenAI Whisper + GPT-4. Built with Next.js, Tailwind CSS, and TypeScript. Record voice, transcribe it, and get smart AI replies — built for EmpireUI and submitted for Vibe-a-thon 1.0.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages