This is the TensorFlow source code for the "Semi-supervised Audio Classification with Consistency-Based Regularization" paper.

The environment can be found in dockerhub:

docker pull loklu/mt_tensorflow:tf1.2.1_py35_lib3

To prepare the dataset, please download the urban sound data and save it under the main folder("./SS_Audio_Classification/"). Then run the following script:

sh prepare_data.sh

To train the model, run:

cd SS_Audio_Classification

sh experiments/script.sh

The code is built on the official code of paper "Mean teachers are better role models".