In the 2010 film Inception by Christopher Nolan, he introduced the dream within a dream theme where the main characters had the ability to create a world representation that is so close to the reality that they couldn't tell the difference. We want to utilize the current advances in GAN training and GRUs to build a pipeline for realistic. Recurrent Adversarial Network
What it does
Our pipeline takes in parameters, like a company website or service description link, web scrapes for desired information geared toward informing the user. That data is converted to audio then we use our CRAN GAN to generate realistic video. With this approach, we demonstrate a new technique for interactive storytelling and also provide a great help in combating fake news videos media with our model pipeline.
How we built it
We used google colaboratory which allows us to utilize google cloud platform GPU, TPU resources, VidTIMIT Audio-Video Dataset, python, Keras, Tensor-flow, Google's speech to text API, Colab
Challenges we ran into
Setting up speech to text capability Creating our Gated recurrent unit to take in our in Finding an appropriate loss function implementation to train our discriminator and encoders.
Accomplishments that we're proud of
We built a complete pipeline architecture of a CRAN GAN
What we learned
How to apply the research state of the art research paper techniques to our machine learning.
What's next for Inception
Improving the GRU so that we can train the model.