This repository contains code for the project of a deep learning enabled semantic communcation system for speech signals, named DeepSC-S. The paper can be found here. This project is trained with TensorFlow 2.4.1.

The dataset adopted in this project can be found in Edinburgh DataShare. Note that the original dataset version utilized in this project has been upgraded, and the orginal simulation environment is TensorFlow 1.14.0. We train the upgraded dataset with some modified hyperparameters in TensorFlow 2.4.1 instead of TensorFlow 1.14.0, which may results in a slight difference of the training loss.

Email: [email protected]