Humanize Synthesized Music Generation Using RNNs

This project is an extension to an existing base project, "Jazz Improvisation Using Deep Learning". This extension deals with the implementation of a network that predicts the strength of a music note playing, given a sequence of preceding notes. For the training of the network, we used a database of jazz improvises containing accurate information about the playing strength of each note. This repository has been adapted to our needs and made into a high-quality dataset that will fit the format of the base project. During this project we used RNN networks and experimented with language generation models.
All parts of the project are implemented in Python language and the network training process was done with the help of Amazon's cloud services (AWS). After processing the information, we experimented several different intermediate models to ascertain the reliability of the network and finally we designed the final and complete network model. Analyzing the results revealed significant success with low loss and r2 values tend to 0.
We hope that integrating this project with the base project will produce results that will sound more “human created” and will be a significant addition to the project we have sought to expand.

Humanize Synthesized Music Generation Using RNNs
Humanize Synthesized Music Generation Using RNNs