How ezaldinwaez/audio-classifier works
Project type and purpose
• This is a standalone desktop-based machine-learning application with two main flows:
• A training pipeline that extracts features from audio and trains an SVM model
• A GUI-based inference app (Tkinter+ttkbootstrap) for real-time letter classification
• Purpose: classify spoken Arabic letters from WAV recordings, either via file upload or live microphone input.
File structure
• Top-level scripts:
• predict_gui.py – launch the GUI, handle user actions (browse, record, play, predict)
• requirements.txt, .gitignore, README.md
• src/ directory:
• feature_extraction.py – load WAV files, record via sounddevice, extract MFCCs & spectral features (librosa)
• train_model.py – load extracted features, train an SVM (scikit-learn), serialize the model with joblib
• No separate “data” folder—assume raw audio lives externally or is captured at runtime.
• No server components, databases, or microservices: this is an all-in-one desktop app.
README for additional insights
• Training steps: run feature_extraction.py, then train_model.py
• Inference steps: launch predict_gui.py
• Dependencies hint at audio I/O (sounddevice, soundfile, pygame), feature extraction (librosa), ML (scikit-learn, numpy), UI (ttkbootstrap/Tkinter).
• No external APIs or cloud services involved.