Human's ability to share knowledge has been vital in our species' success. As humans evolved and technology improved, so have the ways in which we share knowledge. A common form of content delivery in today's world is a powerpoint slide presentation. We wanted to create a tool that would make it easier to create Google Slides presentation without even lifting a finger.
What it does
AutoDeck automatically generates a Google Slides presentation as you speak. Incoming audio from the presenter is streamed through various machine learning models and converted to text to be tokenized and interpreted as commands / content.
How we built it
AutoDeck's speech to text engine was build on Chrome's Web Speech API. The text was then sent to Google Cloud platform natural language API for analysis. The NLP performs steps such as entity detection and sentiment analysis. This information is then sent to our parser. By using the salience of each entity, we are able to determine its importance and take the appropriate steps to construct the slide.
Challenges we ran into
Integrating all the moving parts into one cohesive application. The data pipeline was tricky to setup and have running continuously.
Accomplishments that we're proud of
What's next for AutoDeck