🎤 AI Voice Transcriber

A modern web application that records your voice, transcribes it using OpenAI Whisper, and intelligently cleans up the transcript using a local AI model. All processing happens locally on your machine for complete privacy.

✨ Features

• 🎙️ Browser-based Recording: Record audio directly in your web browser using the MediaRecorder API
• 📝 AI-Powered Transcription: Accurate speech-to-text conversion using OpenAI's Whisper model (runs locally)
• 🧹 Intelligent Text Cleaning: Automatically removes filler words (um, uh, like), repetitions, and false starts while fixing grammar
• 📋 One-Click Copy: Easily copy cleaned transcripts to your clipboard
• 🔒 Privacy-First: All processing happens locally - your audio never leaves your machine
• 🎨 Modern UI: Beautiful gradient interface with smooth animations and responsive design

🛠️ Tech Stack

Backend

• FastAPI: Modern, fast web framework for building APIs
• OpenAI Whisper: State-of-the-art speech recognition model
• Ollama: Local LLM runtime for text cleaning (using gemma2:2b model)

Frontend

• Vanilla JavaScript: No framework dependencies
• HTML5 MediaRecorder API: Browser-native audio recording
• Modern CSS: Gradient backgrounds, animations, and responsive design

📋 Prerequisites

Before you begin, ensure you have the following installed:

• Python 3.8+ (Python 3.11 recommended)
• Ollama installed and running on your system
  • Download from ollama.ai
  • Pull the required model: ollama pull gemma2:2b

🚀 Installation

1. Clone the repository

   git clone https://github.com/Aayushmaan-24/VoiceTranscriber
   cd VoiceCleaner/VoiceTranscriber

2. Create a virtual environment (recommended)

   python3 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Python dependencies

   pip install fastapi uvicorn openai-whisper ollama

   Note: Installing Whisper will also install PyTorch and other ML dependencies, which may take some time.

4. Verify Ollama is running

   ollama list  # Should show gemma2:2b model

   If the model is not installed:

   ollama pull gemma2:2b

🎯 Usage

1. Start the FastAPI server

   cd backend
   uvicorn main:app --reload

   The server will start on http://localhost:8000

2. Open your browser

   • Navigate to http://localhost:8000
   • The application will automatically redirect to the frontend

3. Record and transcribe

   • Click "Start Recording 🎙️" and grant microphone permissions
   • Speak naturally - include ums, pauses, and filler words
   • Click "🛑 Click to Stop" when finished
   • Click "Transcribe with AI" to process your recording
   • Wait for the AI to transcribe and clean your audio
   • Copy the cleaned transcript using the "📋 Copy Transcript" button

📁 Project Structure

VoiceTranscriber/
├── backend/
│   └── main.py          # FastAPI application with transcription endpoint
├── frontend/
│   ├── index.html       # Main HTML file
│   ├── script.js        # JavaScript for recording and API calls
│   └── styles.css       # Styling and animations
├── temp/                # Temporary audio files (auto-created)
└── venv/                # Python virtual environment

🔧 Configuration

Changing the Whisper Model

Edit backend/main.py and modify the model name:

whisper_model = whisper.load_model('base')  # Options: tiny, base, small, medium, large

Note: Larger models provide better accuracy but require more memory and processing time.

Changing the LLM Model

Edit backend/main.py and modify the Ollama model:

response = ollama.generate(model = 'gemma2:2b', prompt = prompt)

Available models depend on what you've installed with Ollama. Other options include:

• llama3.2:3b
• llama3.2:1b
• mistral:7b

🐛 Troubleshooting

Microphone Access Denied

• Ensure your browser has microphone permissions enabled
• Check browser settings for site permissions
• Try using HTTPS or localhost (some browsers require secure contexts)

Ollama Connection Error

• Verify Ollama is running: ollama list
• Ensure the model is installed: ollama pull gemma2:2b
• Check if Ollama is accessible on the default port

Whisper Model Loading Issues

• First-time model download may take several minutes
• Ensure you have sufficient disk space (~500MB for base model)
• Check your internet connection for initial download

Slow Transcription

• Use a smaller Whisper model (tiny or base) for faster processing
• Consider using a smaller LLM model for text cleaning
• Ensure your system meets the minimum requirements

📝 API Endpoints

POST /transcribe

Transcribes and cleans an audio file.

Request:

• Content-Type: multipart/form-data
• Body: Audio file (webm format)

Response:

{
  "transcript": "Cleaned transcript text...",
  "raw": "Original transcript with filler words..."
}

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is open source and available under the MIT License.

🙏 Acknowledgments

• OpenAI Whisper for speech recognition
• Ollama for local LLM inference
• FastAPI for the web framework

📧 Support

For issues, questions, or contributions, please open an issue on the repository.

---

Made with ❤️ for privacy-conscious voice transcription