Eavesdrop

Inspiration

Inspired by the video game in the tv show "Silicon Valley" that listens to players and changes game scenes to include advertisements based on their conversation.

What it does

Takes audio input, and input from pre-written material eg. a movie script. To create modified material based on conversations overheard in the audio input and the pre-written material.

How we built it

We modified a script from the google cloud platform to generate input from speech to text. We then used primarily python to interface between the conversational input, the movie script and the gtp-2 algorithm.

Challenges we ran into

Our laptops do not have the computational power to run the GTP-2 model efficiently

Accomplishments that we're proud of

Learning to interface between local programs and google cloud platform APIs

What we learned

How to work with google cloud
How to modify the outputs of the gtp-2 generator to give better outputs

What's next for Eavesdrop

Adding text to speech output to read the modified material back to the users.
Sending output to a animated movie scene generator to dynamically generate upcoming scenes in the movie that include prompts from the watchers conversations during previous scenes.