jarvas01/asr-framework-comparison

Multi-framework ASR comparison platform for evaluating acoustic modeling performance across high and low-resource languages # Multi-Framework ASR Comparison Platform A comprehensive research platform for comparing Automatic Speech Recognition (ASR) frameworks across multiple languages, with emphasis on low-resource language acoustic modeling performance. ## 🎯 Research Objective Investigate whether transformer-based ASR models (Whisper) maintain superiority over CTC-based models (Wav2Vec2) when applied to low-resource languages with limited training data. ## 🚀 Features - **Multi-Framework Support**: Compare Whisper, Wav2Vec2, and SpeechBrain - **Multilingual Testing**: 10+ languages including low-resource varieties - **Comprehensive Metrics**: WER, Verb/Noun Ratio, Average Utterance Length, Real-Time Factor - **Video Support**: Automatic audio extraction from MP4, AVI, MOV, MKV, WebM - **Professional UI**: Streamlit-based interface with visualization dashboards - **Data Export**: Results exportable as CSV/JSON for analysis ## 📦 Supported Languages **High-Resource:** - English, Spanish, Portuguese, Mandarin Chinese, French, German **Low-Resource:** - Haitian Creole, Creole variants **Medium-Resource:** - Italian, Russian, Arabic, Japanese ## 🛠️ Installation ### Prerequisites - Python 3.11 - macOS, Linux, or Windows ### Setup ```bash # Clone repository git clone github.com cd asr-framework-comparison # Create virtual environment python3.11 -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate # Install dependencies pip install --upgrade pip pip install -r requirements.txt # Download spaCy model python -m spacy download en_core_web_sm ``` ## 🎬 Quick Start ```bash # Activate virtual environment source venv/bin/activate # Run the application streamlit run app.py ``` The app will open in your browser at `localhost` ## 📊 Usage 1. **Select Language**: Choose from 10+ supported languages 2. **Load Models**: Click buttons to load Whisper, Wav2Vec2, SpeechBrain 3. **Upload Audio/Video**: Drag and drop …