Author:
This project has two main files
- main.py - Contains the main executable code.
- features.py - This contains helper functions like to extract audio features.
The name of the dataset is GTZAN dataset and it can be downloaded from the given link (https://www.kaggle.com/datasets/andradaolteanu/gtzan-dataset-music-genre-classification)
The data will be downloaded in zip form and contains multiple files. We'll be using only the audio files present in the extracted folder. So we just need the genres_original folder from the zip. Rename the folder "genres_original" to "genres" and put that folder in a separate folder with the two files mentioned above.
To see the working of code just run the ipynb file named main.ipynb.
Features Used:
Audio features are primarily from 2 domains:
• Time-domain Features:
1. Amplitude Envelope
2. Root Mean Square Energy
• Frequency-domain Features:
1. Mel Frequency Cepstral Coefficient (MFCC)
MFCC was used in this project because unlike other features. MFCC has the capability to differentiate between various frequencies just like humans would perceive those sounds.
Techniques and Models Used
• K-Means Clustering
• K-Nearest Neighbours
• Multiclass SVM
Metrics Comparison and Results
KMC performed very poorly for the given feature values of the selected dataset because it tries to create clusters using the nearby values.
KNN performed well over the same given dataset because it takes only k values nearest to itself, which is a much better measure to differentiate between different genres.
SVM : In case of polynomial model, the accuracy slightly decreases compared to the KNN model. Whereas the linear models shows slight improvement in accuracy.
