🎙️ AiTranscript

Voice transcription tool with AI-powered cleanup capabilities. Transcribe audio from YouTube videos, uploaded files, or live recordings, with intelligent summarization and message refinement using local LLM or OpenAI GPT.

🎥 Demo

1. Record Transcription

2. YouTube Transcription

3. Audio File Transcription

✨ Features

• Multiple Input Methods:

  • YouTube URL transcription
  • Audio file upload (mp3, wav, m4a, ogg, flac)
  • Live voice recording

• Dual AI Processing Modes:

  • Summarize Mode: Get clear, concise summaries of transcripts with key points extraction
  • Refine Mode: Transform voice recordings into well-structured, professional messages

• Flexible AI Providers:

  • Local LLM (Default): Run AI models on your machine via Ollama - free, private, no API key needed
  • OpenAI GPT: Cloud-based option for GPT-4/3.5 models

• Local Transcription: Uses Whisper model locally for privacy and cost-effectiveness

• User-Friendly Interface: Built with Streamlit for an intuitive web experience

🏗️ Architecture

AiTranscript follows a service-oriented architecture with clear separation of concerns:

• Services Layer: YouTube extraction, audio transcription, AI summarization
• Utils Layer: Input validation, file handling
• UI Layer: Reusable Streamlit components

📋 Prerequisites

• Python: 3.11 or higher
• FFmpeg: Required for audio processing
• Ollama: For local LLM (recommended, free)
• OpenAI API Key: Optional, only if using OpenAI provider

Installing FFmpeg

macOS:

brew install ffmpeg

Ubuntu/Debian:

sudo apt-get update
sudo apt-get install ffmpeg

Windows: Download from ffmpeg.org and add to PATH.

Installing Ollama (for Local LLM)

macOS/Linux:

curl https://ollama.ai/install.sh | sh

Windows: Download from ollama.ai

Pull a model:

# Start Ollama
ollama serve

# Pull a model (in another terminal)
ollama pull llama2
# or
ollama pull mistral

🚀 Quick Start

1. Clone the Repository

git clone <repository-url>
cd AiTranscript

2. Install uv (if not already installed)

# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

3. Create Virtual Environment and Install Dependencies

# Create virtual environment
uv venv

# Activate virtual environment
# On macOS/Linux:
source .venv/bin/activate
# On Windows:
.venv\Scripts\activate

# Install dependencies
uv pip install -e .

4. Configure Environment Variables

# Copy the example environment file
cp .env.example .env

# Edit .env if needed (optional - defaults work out of the box)

Environment variables (all optional):

• AI_PROVIDER: Choose 'local' (default) or 'openai'
• LOCAL_MODEL: Local model to use (default: llama2)
• OPENAI_API_KEY: Your OpenAI API key (only if using OpenAI)
• OPENAI_MODEL: OpenAI model to use (default: gpt-4-turbo-preview)
• WHISPER_MODEL_SIZE: Whisper model size (default: base)

5. Run the Application

streamlit run app.py

The application will open in your default browser at http://localhost:8501.

📦 Project Structure

aitranscript/
├── src/
│   ├── __init__.py
│   ├── common/              # Shared services and utilities
│   │   ├── __init__.py
│   │   ├── ai_processing.py # AI processing logic
│   │   ├── ai_service.py    # AI service integration
│   │   └── audio_service.py # Audio transcription (Whisper)
│   ├── recording/           # Voice recording feature
│   │   ├── __init__.py
│   │   ├── service.py
│   │   └── view.py
│   ├── ui/                  # UI components
│   │   ├── __init__.py
│   │   └── components.py
│   ├── upload/              # File upload feature
│   │   ├── __init__.py
│   │   ├── service.py
│   │   └── view.py
│   ├── utils/               # Utilities
│   │   ├── __init__.py
│   │   ├── config.py
│   │   ├── file_handler.py
│   │   ├── time_utils.py
│   │   └── validators.py
│   └── youtube/             # YouTube feature
│       ├── __init__.py
│       ├── provider.py
│       ├── service.py
│       └── view.py
├── app.py                   # Main Streamlit application
├── pyproject.toml           # Project configuration
├── .env.example             # Environment variables template
└── README.md                # This file

🔧 Configuration

Whisper Model Sizes

Choose the appropriate model size based on your needs:

Model	Size	RAM	Speed	Accuracy	Use Case
tiny	39M	~1GB	Fastest	Lowest	Quick drafts
base	74M	~1GB	Fast	Good	Default choice
small	244M	~2GB	Medium	Better	Quality focus
medium	769M	~5GB	Slow	High	Professional
large	1550M	~10GB	Slowest	Best	Maximum accuracy

Set in .env:

WHISPER_MODEL_SIZE=base

AI Provider Models

Local LLM Models (via Ollama) - Default & Recommended

Model	Size	RAM	Speed	Quality	Use Case
llama2	3.8GB	~8GB	Fast	Good	Default, balanced
llama3	4.7GB	~8GB	Fast	Excellent	Best quality
mistral	4.1GB	~8GB	Fast	Very Good	Great alternative
phi	1.6GB	~4GB	Fastest	Good	Low resource
codellama	3.8GB	~8GB	Fast	Good	Code-focused

Advantages:

• ✅ Free - no API costs
• ✅ Private - data stays on your machine
• ✅ No API key needed
• ✅ Works offline

Setup:

# Install Ollama
curl https://ollama.ai/install.sh | sh

# Pull a model
ollama pull llama2

# Run the app (Ollama will start automatically)
streamlit run app.py

OpenAI Models (Optional)

Model	Cost/1K tokens	Speed	Quality	Use Case
gpt-3.5-turbo	$0.0015	Fast	Good	Cost-effective
gpt-4-turbo	$0.01	Medium	Excellent	High quality
gpt-4	$0.03	Slow	Best	Premium quality

Setup:

# Set in .env
AI_PROVIDER=openai
OPENAI_API_KEY=your-api-key-here
OPENAI_MODEL=gpt-4-turbo-preview

🛠️ Development

Using uv for Package Management

# Add a new dependency
uv pip install package-name

# Add a development dependency
uv pip install --dev package-name

# Update dependencies
uv pip install --upgrade package-name

# Sync dependencies from pyproject.toml
uv pip sync

Running Tests (Coming Soon)

pytest

Code Formatting

# Format code with black
black .

# Lint with ruff
ruff check .

📝 Usage

Processing Modes

Summarize Mode - Get clear summaries of content:

• Perfect for YouTube videos, podcasts, or long recordings
• Extracts key points automatically
• Choose from concise, detailed, or bullet-point styles

Refine Mode - Transform your voice into professional messages:

• Record what you want to say naturally
• AI refines it into a clear, well-structured message
• Choose tone: professional, friendly, formal, or casual
• Optionally specify recipient context for better refinement

Step-by-Step Guide

1. Configure Settings (in sidebar):

   • Select processing mode (Summarize or Refine)
   • Configure mode-specific options
   • Note: AI Provider is configured via .env file

2. Choose Input Method:

   YouTube Transcription:

   • Navigate to the "YouTube" tab
   • Paste a YouTube URL
   • Click "Get Transcript"
   • View transcript and AI-processed result

   File Upload:

   • Navigate to the "Upload File" tab
   • Upload an audio file (mp3, wav, m4a, ogg, flac)
   • Click "Transcribe File"
   • View transcript and AI-processed result

   Voice Recording:

   • Navigate to the "Record Audio" tab
   • Click the microphone button to start/stop recording
   • Click "Transcribe Recording"
   • View transcript and AI-processed result

3. Download Results:

   • Use the download buttons to save your transcript and AI-processed output

🔒 Privacy & Security

• Local Transcription: Audio is transcribed locally using Whisper (no data sent to external services)
• No Data Storage: Transcripts are not stored permanently
• Temporary Files: Automatically cleaned up after processing
• API Key Security: Store your OpenAI API key securely in .env (never commit to version control)

🤝 Contributing

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

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• OpenAI Whisper for local transcription
• Ollama for local LLM support
• Streamlit for the web framework
• OpenAI for GPT API

📞 Support

For issues and questions, please open an issue on GitHub.

---

Built with ❤️ using Streamlit, Whisper, Ollama, and OpenAI GPT