https://github.com/fdch/kwgan
A GAN architecture to generate raw audio
https://github.com/fdch/kwgan
Last synced: 5 months ago
JSON representation
A GAN architecture to generate raw audio
- Host: GitHub
- URL: https://github.com/fdch/kwgan
- Owner: fdch
- Created: 2020-04-17T01:27:12.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2022-04-24T16:33:08.000Z (about 4 years ago)
- Last Synced: 2025-05-22T11:35:09.196Z (about 1 year ago)
- Language: Python
- Size: 87.9 KB
- Stars: 3
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# KWGAN
A GAN architecture to generate raw audio based on a dataset. This work is based on [wavegan](https://github.com/chrisdonahue/wavegan) and it is made using Tensorflow and Keras.
## Usage
For now just edit the variables within the python script to match your needs.
```
epochs_number = 40001
model_save_interval = 1000
audio_export_interval = 400
audio_export_per_epoch = 3
audio_samplerate = 16000
TRAIN_BUF = 2048
TEST_BUF = 128
BATCH_SIZE = 256
LATENT_DIM = 128
DIMS = (2**14,1)
gen_learning_rate = 0.0001
disc_learning_rate = 0.0002
number_of_disc_layers = 22
```
Then, simply use `python kwgan.py` to begin training. GPU usage is a must. This work was tested using *pitzer* at [https://www.osc.edu](https://www.osc.edu)