Using deep learning to categorize music as time progresses through spectrogram analysis.
Our aim is to create a neural network (CNN + LSTM RNN) that recognizes music genre and provides a user-friendly visualization for the network's current belief of the genre of a song.
This project uses Keras, using TensorFlow for the backend, for the neural network and Tornado for serving requests.
The rationale for this particular model is based on several works, primarily Grzegorz Gwardys and Daniel Grzywczak, Deep Image Features in Music Information Retrieval, Recommending music on Spotify with Deep Learning, and Convolutional-Recurrent Neural Network for Live Music Genre Recognition.
The paper explaining our rationale, procedures, and results can be found here: Using Deep Learning to Categorize Music through Spectrogram Analysis
The slide deck can be found on Google Slides: Using Deep Learning to Categorize Music as Time Progresses Through Spectrogram Analysis.