Concept Convos

Home Page of Concept Convos
Podcast Viewing Page

Inspiration

Our main inspiration came from watching AI-generated YouTube videos. While endlessly scrolling through the app, we found so many examples of celebrities having their voices cloned using AI tools. We saw endless potential in these technologies, and were especially intrigued by the interactions between these trained computer models. With a team that listens to several podcasts, we were always left wanting our favorite celebrities to be guests on our favourite shows! And so, we created Concept Convos.

What it does

Concept Convos is a web application that creates a short podcast with your two favourite celebrities, using topics that you decide! There are currently over 7 different famous stars/characters to choose from, leading to 50+ combinations of voices to keep you entertained for hours!

How we built it

The web application was built using the Flask microweb framework, and coded using Python, HTML, CSS and JavaScript. This interface allows users to input their desired celebrities, as well as a prompt for the podcast generation. We also implemented a progress bar, as well as a podcast viewing page. Using Flask and Python, we were able to utilize the OpenAI API for generating the script of the podcast, as well as the Play.HT API for generating the voice sound files.

Challenges we ran into

Our main challenges came from learning JavaScript, HTML and CSS. We were all mostly experienced with Python, and there was a learning curve to the frontend development. We also struggled to connect our frontend and backend with HTTP requests and sockets as we had never used these beforehand. Finally, we also struggled to get the voice generations to sound adequate, as some celebrities had more difficult to obtain training data.

Accomplishments that we're proud of

We're proud of the fact that we managed to complete everything that we had planned to do within the given timeslot. We even had time to spare, to add extra features such as a progress bar and face-accurate images.

What we learned

We learned numerous new skills and tools, ranging from JavaScript, Flask, CSS and the OpenAI API.

What's next for Concept Convos

The next steps would be to increase the number of possible celebrities' and characters available on our app. We could also obtain better quality training data for the existing options to make their speech patterns and tone even more accurate to their real-life counterparts.