kenngare/swahili-transcription

use llm to transcribe swahili audio # Swahili Audio Transcription A Python tool for transcribing Swahili audio files to text using OpenAI's Whisper model with CUDA acceleration support. ## Features - 🎯 **Optimized for Swahili**: Pre-configured for Swahili language transcription - 🚀 **CUDA Support**: Automatic GPU acceleration when available - 🔧 **Flexible Configuration**: Multiple model sizes and customization options - 📁 **Batch Processing**: Transcribe multiple files at once - 🖥️ **CLI Interface**: Easy-to-use command-line interface - 📝 **Multiple Output Formats**: Text files with automatic naming - 🔍 **Device Information**: Check CUDA availability and device specs ## Installation ### Prerequisites - Python 3.8 or higher - CUDA-compatible GPU (optional, for acceleration) - FFmpeg (for audio processing) ### Install FFmpeg #### Ubuntu/Debian ```bash sudo apt update sudo apt install ffmpeg ``` #### macOS ```bash brew install ffmpeg ``` #### Windows Download from FFmpeg official website or use chocolatey: ```bash choco install ffmpeg ``` ### Install the Package 1. Clone the repository: ```bash git clone cd swahili-transcription ``` 2. Install dependencies: ```bash pip install -r requirements.txt ``` 3. Install the package in development mode: ```bash pip install -e . ``` ### CUDA Setup (Optional) For GPU acceleration, ensure you have: - CUDA Toolkit installed - Compatible PyTorch with CUDA support Install PyTorch with CUDA: ```bash pip install torch torchaudio --index-url download.pytorch.org ``` ## Usage ### Command Line Interface #### Basic Transcription ```bash swahili-transcribe audio.wav ``` This will create `audio.txt` with the transcription. #### Specify Output File ```bash swahili-transcribe audio.wav -o transcription.txt ``` #### Use Different Model Size ```bash swahili-transcribe audio.wav -m large ``` Available models: `tiny`, `base`, `small`, `medium`, `large`, `large-v2`, `large-v3` #### Force CPU Usage ```bash swahili-transcribe audio.wav -d c …