Music-Genre-Classification

Classifies the genre of music files (.wav) using ML models. Trained using the GTZAN Dataset: https://www.kaggle.com/datasets/andradaolteanu/gtzan-dataset-music-genre-classification

Repo link: https://github.com/break-through-19/Music-Genre-Classification

Steps to execute

1. Download the dataset from Kaggle: https://www.kaggle.com/datasets/andradaolteanu/gtzan-dataset-music-genre-classification?resource=download

2. From the downloaded dataset, move the contents of genres_original (contains folders of .wav files) folder to Music-Genre-Classification/dataset folder.

3. Navigate to Home -> Add-Ons -> Search for "Audio Toolbox" -> Install.

4. Run MATLAB script:

   • scripts/generate_mel_feature_csv.m

5. To normalize the extracted feature CSV and generate 3D scatter plots:

   • scripts/normalize_and_plot_feature_space.m

6. Navigate to Home -> Add-Ons -> Search for "Statistics and Machine Learning Toolbox" -> Install.

7. To write PCA transformed features and train/test SVM models on original, PCA, and curated feature variants:

   • scripts/iteration4_svm.m

8. Navigate to Home -> Add-Ons -> Search for "Deep Learning Toolbox" -> Install. To train/test CRNN models on original, PCA, and curated feature variants:

   • scripts/crnn.m

Feature Extraction Pipeline (MATLAB)

The MATLAB pipeline is modular and organized as:

• scripts/generate_mel_feature_csv.m : main entry point.
• src/pipeline/processDatasetToMelFeatureTable.m : dataset traversal + orchestration.
• src/io/listGenreAudioFiles.m : class/file discovery from folder structure.
• src/audio/splitAudioIntoFixedSegments.m : 30s audio -> 3s segments.
• src/features/extractMelSpectrogramFeatureVector.m : extract 16 important features (4 per category).
• src/features/extractHarmonicFeatureVector.m : compute harmonic feature block.
• src/features/getImportantMelFeatureNames.m : fixed names for the 16 features.
• src/io/buildFeatureRowTable.m : metadata + feature row construction.
• src/io/alignColumnsToExampleCsv.m : optional schema alignment with example CSV.
• src/preprocessing/readFeatureDataset.m : read and validate the feature CSV.
• src/preprocessing/meanCenterAndNormalizeFeatureTable.m : mean-center and z-score normalize feature columns.
• src/visualization/getFeatureCategoryDefinitions.m : define the four feature categories.
• src/visualization/generateCategory3DScatterPlots.m : write category-wise 3D scatter plots.

Output CSV:

• data/features/mel_spectrogram_features.csv
• data/features/mel_spectrogram_features_normalized.csv

Generated plot folder:

• data/features/plots_3d_scatter

CRNN output folder:

• data/model_outputs/crnn

Included metadata columns:

• class_name (from folder name)
• wav_file_name (from WAV file name)
• segment_index (1-based segment id within each WAV file)

Feature columns:

• Spectral/timbral level features (4):
  • timbral_log_mel_mean
  • timbral_log_mel_median
  • timbral_band_energy_spread
  • timbral_frame_peak_mean
• Spectral variability features (4):
  • variability_log_mel_std_global
  • variability_band_std_mean
  • variability_frame_energy_std
  • variability_band_range_mean
• Temporal dynamics features (4):
  • temporal_frame_energy_delta_abs_mean
  • temporal_frame_energy_delta_std
  • temporal_zero_crossing_rate_mean
  • temporal_rms_std
• Harmonic features (4):
  • harmonic_voiced_frame_ratio
  • harmonic_f0_mean_hz
  • harmonic_f0_std_hz
  • harmonic_energy_ratio_mean

Iteration 4 SVM Outputs

Input CSV:

• data/features/mel_spectrogram_features_normalized.csv

Generated outputs from scripts/iteration4_svm.m:

• PCA transformed feature CSV:
  • data/transformed_features/mel_spectrogram_features_pca_r4.csv
• Confusion matrix plots:
  • Iteration 4 plots/SVM_Phase1_Baseline.png
  • Iteration 4 plots/SVM_Phase2_PCA.png
  • Iteration 4 plots/SVM_Phase3_Curated.png

SVM experiment phases:

• Phase 1: Baseline SVM on all 16 normalized features.
• Phase 2: SVM on top-4 PCA components.
• Phase 3: SVM on curated 14-feature set (drops timbral_log_mel_median and variability_band_range_mean).