Fine-tuned HuBERT Transformer classifying music genres from noisy audio. Uses Systematic TTA across 5 overlapping 10-second windows for robust accuracy.