This is a implementation of SpecAugment that speech data augmentation method which directly process the spectrogram with Tensorflow & Pytorch, introduced by Google Brain[1]. This is currently under the Apache 2.0, Please feel free to use for your project. Enjoy!

Note: Since the original repository is not actively maintained and has lots of bugs, I have to edit some to make it work.

First, you need to have python 3 installed along with Tensorflow or PyTorch

Next, you need to install some audio libraries work properly. To install the requirement packages, run the following command:

pip3 install -r requirements.txt

And then, run the specAugment.py program. It modifies the spectrogram by warping it in the time direction, masking blocks of consecutive frequency channels, and masking blocks of utterances in time.

Learn more examples about how to do specific tasks in SpecAugment at the test code.

git clone https://github.com/thanhtvt/SpecAugment.git
cd ./SpecAugment/tests
# Uncomment the following line to run with Tensorflow
# python3 spec_augment_test_TF.py
python3 spec_augment_test_pytorch.py

In test code, we using one of the LibriSpeech dataset.

1. https://arxiv.org/pdf/1904.08779.pdf
2. https://github.com/DemisEom/SpecAugment (Original repo)