Project 3 Generative Audio

Bao Hoang, bghoang@ucsd.edu

Abstract

For this project, I want to combine the two kinds of music together. The idea is simple: I will use Python libraries to get some random music samples, then I will combine those sample together and fix the combined sample to make it sounds nice. To generate these samples, I will use the Performance RNN colab to get the midi files. I will then use GANSynth to modify these files. Then I will mix these files together and see what I have at the end. Hopefully, the end result will generate something sounds good, if not, I will problably try to find different training models/ music samples in the future to make the sound better.

Model/Data

Briefly describe the files that are included with your repository:

• train_model folder
• sample folder
• result folder
• mashup file is the final result sound

Code

Your code for generating your project:

• Jupyter notebooks: https://colab.research.google.com/drive/1kscoXV4JFpTTzT5YxKu-cqxhrpH1Rz25

Results

• The train_model folder contains all the files that I used to train for Performance_RNN
• The sample folder containts all the files that were generated using Performance_RNN
• The result sample containts all the files that were generated using GANSynthOB
• Link to hear the sound:https://drive.google.com/file/d/1Yn38t84GLBCu95LMEsmVTe24fDnnfjdx/view?usp=sharing

Technical Notes

Everything I did is on Google Colab so no need to download anything. I had issue with getting the midi file from the Performance_RNN.

Reference

References to any papers, techniques, repositories you used:

• Repositories
  • Performance_RNN: https://colab.research.google.com/notebooks/magenta/gansynth/gansynth_demo.ipynb#scrollTo=EwFvRyg6xSoz
  • GANSynth: https://colab.research.google.com/notebooks/magenta/gansynth/gansynth_demo.ipynb
  • Pydub: https://github.com/jiaaro/pydub
• Blog posts:
  • https://magenta.tensorflow.org/gansynth