موسیقی کی صنف کی درجہ بندی

Music genre classification turns raw audio into a genre label. Early systems used features such as MFCCs, spectral centroid, zero-crossing rate, and tempo with classifiers such as SVMs. MFCCs are derived from log mel-frequency energies using a cepstral transform; a mel-spectrogram instead preserves time-frequency energy on a mel-scaled axis. GTZAN contains 1,000 thirty-second clips across ten genre labels and has documented repeated excerpts, artists, and annotation issues. If repeated material crosses random train/test splits, measured performance may overstate generalization to new recordings. Modern approaches use mel-spectrograms or learned audio representations. Self-supervised methods can pretrain on unlabeled music and then use labeled examples for downstream genre classification; studies such as S3T evaluate label efficiency on specific benchmarks, not a universal ability to train on millions of unlabeled commercial tracks. Genre labels remain fuzzy and culturally situated, so evaluation should describe the taxonomy, split protocol, and dataset limitations.