Variational Autoencoder for Timbre...

beta.0 (December 2020) version

Documentation

See documentation here

Dependencies

• Numpy
• torch
• plotly
• Librosa
• Flask
• Flask cors
• Coverage

Latent Space Visualizations

CQT Reconstructions

Dataset

• Training: London Philarmonic Orchestra dataset

Repository Structure

Directories:

• vae: module containing VAE architecture, training and data uitilities.

  • .py: main module.
  • configs.py
  • data_utils.py
  • loss.py
  • model.py
  • train.py
  • plot_utils.py

• plots: latent space and reconstruction plots.

• tests:

  • .py

• notebooks:

  • 1-.ipynb
  • 2-.ipynb
  • 3-.ipynb

• webserver: integration of audioanalysis module in Flask.

Installation

cd .path/to/timbre-vae
python setup.py install

References

[1]

Authors

Carlos Hernández - carloshero@unizar.es

Department of Electronic Engineering and Communications, Universidad de Zaragoza, Calle María de Luna 3, 50018 Zaragoza