Music Genre Prediction System using Audio and Lyrics Ensemble Models

Purpose: To create a model, which automatically predicts the genre of any new music developed by a user. An ideal use case would be - song streaming websites using the system to easily identify song genres for unlabelled songs.

Techniques used: SVM, KNN, Logistic Regression, Random Forest, Gradient Boosting, Multilayer Perceptron

Tools Used: Jupyter Notebook, Python

Overview of each code:

1. audio_analysis/Audio EDA v0.1ss.ipynb: The code contains:

• Exploratory audio data analysis

2. audio_analysis/Audio Models+Ensemble v0.1ss.ipynb:

• Audio Modeling - QDA, Logistic, SVM

• Ensemble models

3. Text Modeling (5000 words, 1000 word) - text_analysis/final_text_analysis.ipynb The code contains:

• Text Modeling - Random Forest, Gradient Boosting, Multilayer Perceptron

• Ensemble Model

4. Text EDA - data_eda/word_cloud.ipynb

5. Ensemble Boosting.ipynb: The code contains:

• Boosting for ensemble

6. LyricBagofWordsClassifier.ipynb This code contains:

• Text modeling (500 word) - Random Forest, SVM, KNN, Logistic Regression

Datasets Used:

1. Genre Annotations: http://www.tagtraum.com/msd_genre_datasets.html (File name: msd_tagtraum_cd2c.cls.zip)

2. Audio Dataset: https://labrosa.ee.columbia.edu/millionsong/tasteprofile

3. Lyrics Dataset: https://labrosa.ee.columbia.edu/millionsong/musixmatch