In this project completed for a Machine Learning course, my team built an audio preprocessing pipeline in Python by converting 5,000+ raw music tracks into model-ready features. We trimmed each to a 30-
second clips, extracted mel-spectrograms with Librosa, and stored them as compressed HDF5 for batch loading in Google Colab.
We built a five-model comparison benchmarking different approaches, KNN, LSTM, CNN, and Transformer, for multi-label genre
classification. I personally implementing the KNN and LSTM models and evaluating results with precision, recall, Hamming loss, and AUC.
