Audionome: Music Genre Classification

Audionome is an AI-powered music genre classifier built for the Machine Learning course at VIVES University of Applied Sciences. Upload an audio clip and the app predicts its genre in seconds — no music theory knowledge required.

What I Built

Working with a teammate, I trained and compared several Scikit-learn classifiers — including logistic regression, SGD, and Support Vector Machine (SVM) — to find the best-performing model for genre prediction. The SVM classifier came out on top and became the foundation of the final application. The entire pipeline runs inside a Streamlit web app, making it accessible to anyone with a browser.

Audio Feature Extraction

Raw audio files are not fed directly into the model. Instead, Librosa extracts a compact set of meaningful features from each clip:

• MFCCs (Mel-frequency cepstral coefficients) — capture the overall timbral texture of the track
• Spectral centroid and bandwidth — describe where the energy sits in the frequency spectrum
• Zero-crossing rate — a simple but effective proxy for percussiveness and noise
• Chroma features — encode harmonic and melodic content

Pandas and NumPy handle feature aggregation and normalization before the data reaches the classifier.

Key Features

• Upload any audio clip and receive an instant genre prediction with confidence scores
• Model comparison — multiple classifiers were evaluated on the same feature set so results are evidence-based
• Fully interactive Streamlit UI — no installation needed, runs at audionome.streamlit.app
• Clean data pipeline from raw audio through feature extraction to classification output

This project gave me hands-on experience with the full machine learning workflow: feature engineering, model selection, evaluation, and deployment.