tifresi: Time Frequency Spectrogram Inversion

'tifresi' to be pronounced 'tifreeezy' provide a simple implementation of TF and spectrogam suitable for inversion, i.e. with a high quality phase recovery. The phase recovery algorithm used is PGHI (phase gradient heap integration).

Installation

This repository use the ltfatpy packages that requires a few libraries to be installed.

1. Be sure that cmake is installed
   • On debian based unix system:

   sudo apt-get install cmake
   

   • On MacOS X using homebrew:

   brew install cmake
   

   • On MacOS X using port:

   sudo port install cmake
   

2. Install fftw3 and lapack
   • On debian based unix system:

   sudo apt-get install libfftw3-dev liblapack-dev
   

   • On MacOS X using homebrew:

   brew install fftw lapack
   

   • On MacOS X using port:

   sudo port install fftw-3 fftw-3-single lapack
   

3. Install cython (required for installing ltfatpy):

   pip install cython
   

4. Install the requirements (You probably wants to create a virtual environment first)

   pip install -r requirements.txt
   

5. Clone this repository and install it

   git clone https://github.com/andimarafioti/tifresi
   cd tifresi
   pip install .
   

Starting

After installation of the requirements, you can check the following notebooks:

• demo.ipynb illustrates how to construct a spectrogram and invert it.
• demo-mel.ipynb illustrates how to compute a mel spectrogram with the setting used in this repository.

License & citation

The content of this repository is released under the terms of the MIT license. Please consider citing our papers if you use it.

@inproceedings{marafioti2019adversarial,
  title={Adversarial Generation of Time-Frequency Features with application in audio synthesis},
  author={Marafioti, Andr{\'e}s and Perraudin, Nathana{\"e}l and Holighaus, Nicki and Majdak, Piotr},
  booktitle={International Conference on Machine Learning},
  pages={4352--4362},
  year={2019}
}

@article{pruuvsa2017noniterative,
  title={A noniterative method for reconstruction of phase from STFT magnitude},
  author={Pr{\uu}{\v{s}}a, Zden{\v{e}}k and Balazs, Peter and S{\o}ndergaard, Peter Lempel},
  journal={IEEE/ACM Transactions on Audio, Speech, and Language Processing},
  volume={25},
  number={5},
  pages={1154--1164},
  year={2017},
  publisher={IEEE}
}

Developing

As a developer, you can test the package using pytest:

pip install pytest

Then run tests using

pytest tifresi

You can also use the source code checker flake8:

pip install flake8

Then run tests using

flake8 .

Main files

• utils.py: Utility to load, preprocess, downsample the signal
• stft.py: core objects
  1. TFGauss: to compute and invert the STFT with a full Gaussian window
  2. GaussTruncTF: to compute and invert the STFT with a truncated Gaussian window (faster)
• metrics.py: compute metrics to evaluate phase and spectrogram quality
• transforms.py: useful simple transform function for spectrograms
• hparams.py: default parameters
• phase: folder containing source code for phase reconstruction using pghi
  • TODO : create the folder and move files
• tests: folder containing all tests

TODO

• Put the package on pypi
• Improve doc
• Put the documentation on readthedoc or somthing similar