cmrfrd / detectingdeepfakes_blackhat2019

Detect audio deep fakes with bispectral analysis

Detecting Deep Fakes Blackhat 2019

This is the associated repo for the talk Detecting Deep Fakes with Mice

Overview

As machines have gotten better at producing synthetic audio / video that look presumably real there is an ever present need to increase our ability to discern between what is real versus synthetic. The goal of this repo is to present a simple tool to detect real versus fake speech via the use of bispectral analysis

How to run this repo

Just run make

Then go to localhost:8888 in your browser and view the notebooks for further instruction.

detectingdeepfakes_blackhat2019's Issues

All samples are judged as human

Hi，
I run the "DetectingAudioDeepFakes_Sklearn.ipynb"，the model is extracted from it. But all audio in folder"samples" are judged as human. Could you help me solve this problem？

thanks,
ZJH

Embedded model issues

Hello,

I attempted to run the embedded model on the sample data provided, but all files were marked as a robot. I'm not sure if I'm doing something wrong, or if perhaps the embedded model was not the intended one?

Thank you!

Unable to find 'marked' cell to load embedded SVM model

Hi,
I am able to find marked cell to load the embedded SVM model in jupyter notebook. Should I run "DetectingAudioDeepFakes_TF2.ipynb" first followed by "DetectingAudioDeepFakes_Sklearn.ipynb". Can you please share the exact steps to run these files.

Please share your thoughts on this.

Thanks,
Kinjal