Music Genre Classification using CNNs

Authors: Redouane Dziri, Arnaud Stiegler

Base paper: Local-feature-map Integration Using CNNs for Music Genre Classification

Data: GTZAN dataset

This is an ongoing project

Tips

• Requires Python >= 3.6
• Before connecting to the Cloud Storage Bucket, make sure you set the GOOGLE_APPLICATION_CREDENTIALS environment variable to point to your Google Cloud credentials e.g.

os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "/Users/red/Documents/ADL/Project/My First Project-9a16875b5624.json"

Reproducing the experiments

• Collect the GTZAN dataset and store it somewhere you can access it

In our case, we had the dataset on a Google Cloud Bucket. For all subsequent scripts, we fetch the data from the bucket and process it locally.

• Generate train and test set using python early_preprocessing/train_test_split.py
• Run the preprocessing using python feature_engineering/preprocess_full_data.py
• Proceed to EDA on the generated feature-maps using the notebook: feature_engineering/exploration.ipynb

To reproduce the paper results:

• Use training/training_GLCM.ipynb To train our models:
• Use improvement/training_stacked.ipynb or improvement/training_stacked.ipynb