https://github.com/l0sg/seqgan-music
Implementation of a paper "Polyphonic Music Generation with Sequence Generative Adversarial Networks" in TensorFlow
https://github.com/l0sg/seqgan-music
deep-learning generative-adversarial-network music-generation polyphonic seqgan tensorflow
Last synced: 8 months ago
JSON representation
Implementation of a paper "Polyphonic Music Generation with Sequence Generative Adversarial Networks" in TensorFlow
- Host: GitHub
- URL: https://github.com/l0sg/seqgan-music
- Owner: L0SG
- Created: 2018-07-18T04:13:21.000Z (almost 8 years ago)
- Default Branch: master
- Last Pushed: 2018-10-06T06:06:51.000Z (over 7 years ago)
- Last Synced: 2025-07-11T07:59:36.180Z (11 months ago)
- Topics: deep-learning, generative-adversarial-network, music-generation, polyphonic, seqgan, tensorflow
- Language: Python
- Size: 9.55 MB
- Stars: 28
- Watchers: 3
- Forks: 8
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
**This repo is a work-in-progress status without code cleanup and refactoring.**
## Introduction
This is an implementation of a paper [Polyphonic Music Generation with Sequence Generative Adversarial Networks](https://arxiv.org/abs/1710.11418) in TensorFlow.
Hard-forked from the [official SeqGAN code](https://github.com/LantaoYu/SeqGAN).
## Requirements
Python 2.7
Tensorflow 1.4 or newer (tested on 1.9)
pip packages: music21 4.1.0, pyyaml, nltk, pathos
## How to use
`python music_seqgan.py` for full training run.
SeqGAN.yaml contains (almost) all hyperparameters that you can play with.
5 sample MIDI sequences are automatically generated per epoch.
## Dataset
The model uses a MIDI version of Nottingham database () as the dataset.
Preprocessed musical word tokens are included in the "dataset" folder.