This repository contains the code of differentiable short-time Fourier transform (DSTFT). It allows automatic tuning of spectrogram parameters by optimizing them by gradient descent. DSTFT is a time-frequency neural network layer whose weights are its parameters (i.e. window and hop lengths at the moment).
git clone https://github.com/maxime-leiber/dstft
cd dstft
conda env create -f environment.yml
conda env update -f environment.yml
pip install -e .