Skip to content

Latest commit

 

History

History
65 lines (38 loc) · 2.5 KB

README.md

File metadata and controls

65 lines (38 loc) · 2.5 KB
Raw

Music Genre Classification Project

This repository is based on the recognition of musical genres through supervised and unsupervised learning.

apr_project_architecture

Plugins:

$ pip install -r requirements.txt

Information:

the dataset used for built this project is the notorious GTZAN dataset, recovered from kaggle (link to database: https://www.kaggle.com/datasets/andradaolteanu/gtzan-dataset-music-genre-classification).

In the utils directory, there are all that classes used for preprocessing the dataset and performing data augmentation (I did not use the csv file available at the previous link, I built my own).

  • features_computation: computation of the various features to extract from audio files.
  • features_extractions: extraction of the computed features to a csv file in a proper directory.
  • features_visualizations: visualization of the single audio signals and the visualization of the various extracted features with a confrontation of the different genres.
  • prepare_dataset: check the duration of audio files and perform data augmentation (30s long file -> ten 3s long chunk).

Then we have the core classes of the project:

  • main: main class of the project that calls all the other.
  • genres_ul_functions: performs k-means clustering and then performs its evaluation.
  • genres_sl_functions: performs various classification algorithms (Neural Network, Random Forest, K-Nearest Neighbors, Support Vector Machine) and evaluate their performances with confusion matrix, roc curve and metrics (accuracy, F1-score,...).
  • plot_functions: used for defining all the plot functions.
  • constants: contains all the constants used in the project.

Performace Summary:

MULTILAYER PERCEPTRON RANDOM FOREST K-NEAREST NEIGHBORS SUPPORT VECTOR MACHINE
ACCURACY 84.80 79.33 89.80 89.40
F1-SCORE 0.85 0.79 0.90 0.89
EXECUTION TIME (sec) 63.72 52.76 7.73 21.40