music genre classification

Technologies used

• Librosa for extracting audio features
• Scikit-learn for machine learning
• Streamlit for deployment
• PyDub for converting mp3 to wav
• Docker for deployment

Features extracted

• chroma_stft
• rmse
• spectral_centroid
• spectral_bandwidth
• rolloff
• zero_crossing_rate

Model used

• OutputCodeClassifier with Logistic Regression for multiclass classification

How it workes

• First the user has an option to either view demo or upload their own audio
• Then if the user selects classify, the web app calculates all the above mentioned features and lassify the audio into one of 10 genres using our model
• If user selects spectogram, it display the mel spectogram to the user

link of the deployed app

https://music-type-predictor.herokuapp.com