We were inspired to visualize alternate history. We were very interested in working with Deep Generative Adversarial Networks for generative material, and transformers for NLP functions.
What it does
Generates audio of historic speeches given by the speaker them self. Dive into alternative history as you listen to past presidents deliver written but never delivered speeches in some of the most critical moments of history.
How we built it
We built this primarily using cutting edge deep learning models in python, PyTorch, and TensorFlow. A culmination of several difference machine learning architectures, and a vast database cultivated to give the most authentic experience.
Challenges we ran into
One of the most significant challenges we ran into was connecting the 3 models and 2 API's into one cohesive UI/UX. Early Sunday morning most of our data and previous work was wiped and we were forced to change our approaches. We had to quickly adopt our deep learning architectures to this challenge, and had to give up on some of our favorite features.
Accomplishments that we're proud of
We are super proud of what we were able to accomplish in these 24 hours. We battled a great deal of adversity in our short time, as much of our data and programs were deleted early Sunday morning and we were still able to complete our project.
What we learned
For some of us, we learned our first coding language during this time. For the others, we dived deeper than ever before into python. Gained valuable experience working with and parsing large data
What's next for Program
We intend to implement real time video and higher quality audio such that we can generate video that is a realistic depiction of alternative histories.