Beware of Deepfake audio scams

As ai moves forwards, there will be more and more misuse of technology. Recently, an audio Deepfake of a CEO’s voice was used in a $243,000 scam.

Here is a sample of what audio syncretization can achieve (the example is not related to the scam):

In 2018, The Goolge’s Tacotron team created a text to speech synthesizer that can copy your voice after listening just 5 seconds.



Berkley Deep Unsupervised Learning

CS294-158 Deep Unsupervised Learning Spring 2019

Lectures, papers and assignments for Berkeyls Deep Unsupervised Learning Course are now available here:

The course deals with 2 areas of deep learning, namely Deep Generative Models and Self-supervised Learning.

Topics are:

  • Generative adversarial networks
  • variational autoencoders
  • autoregressive models
  • flow models
  • energy based models
  • compression
  • self-supervised learning
  • semi-supervised learning.

The course is currently ongoing so not all lectures are available yet.