Piano Music Generation using GRU

Generating music using a neural network has always been a fascinating task. This repository showcases one such neural network which is able to produce somewhat nice results.

Model Architecture

The model leverages the GRU architecture. For indepth analysis do checkout the training notebook or checkout the model on Kaggle

Training

The model was trained on the midi music of Frédéric Chopin from the Classical Music MIDI available on Kaggle. It contains 48 music pieces.

The model was trained for 46 epochs.

Results

These are the training and validation loss curves respectively.

Training_loss

Validation_loss

Predictions

To generate your own pieces I would suggest you to open the prediction.ipynb in Google Colab and run it there.

You can git clone the repository there.

Here are some of the generated music pieces.

• Prediction 1

test-0.-.Made.with.Clipchamp.mp4

• Prediction 2

test-1.mp4

• Prediction 3

test-2.mp4

• Prediction 4

test-3.mp4

• prediction 5

test-4.mp4

Help Me Benchmark It

If you have listened to all these prediction please share your review by filling this form

Thank You