This repo provides a code for estimation of DeepSpeech Distances, new evaluation metrics for neural speech synthesis.

conda env create -f environment.yml
conda activate DSD-TF1

Create "Audio/" directory inside the project and insert all audio files with following structure "Audio/{conversion_name}/{auido_type}/*.wav" then run the Jupyter file.

This fork is to calculate Fréchet DeepSpeech Distance and Kernel DeepSpeech Distance only of a selected data path. Added environment file to run it locally. TODO: Jupyter for Google Colab (the old one from the original repo doesn't work anymore as it's for tensorflow 1.x)

We provide a tensorflow meta graph file for DeepSpeech2 based on the original one available with the checkpoint. The provided file differs from the original only in the lack of map-reduce ops defined by horovod library; therefore the resulting model is equivalent to the original.

This is an 'alpha' version of the API; although fully functional it will be heavily updated and simplified soon.

[1] Mikołaj Bińkowski, Jeff Donahue, Sander Dieleman, Aidan Clark, Erich Elsen, Norman Casagrande, Luis C. Cobo, Karen Simonyan, High Fidelity Speech Synthesis with Adversarial Networks, ICLR 2020.

[2] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, Sepp Hochreiter, GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium, NeurIPS 2017.

[3] Mikołaj Bińkowski, Dougal J. Sutherland, Michael Arbel, Arthur Gretton, Demystifying MMD GANs, ICLR 2018.