VoxSim Smart VR Speech and Debate Training System

The Project

There are three primary components to my program: the user interface on unity software, AI powered feedback analysis, and the C# coding scripts used to tie everything together. To start, this was my first time modeling with Unity Software, so learning how to use the new tools was one of many firsts. In addition to adapting to the software tools, I had trouble finding suitable 3D assets to help decorate and construct the virtual environment users reside within. Higher quality assets are not free and are expensive. While I was on the prototyping stage, balancing between having a minimum viable product and satisfying the aesthetic outlook was difficult. I had to deconstruct many assets for specific elements like a podium or TV display. I wanted helpful feedback that was customizable for the setting of the talking point, so I delved deep into prompt engineering for specific outputs. This involved adjusting the temperature of the response and minimizing variance. After every user input, I assigned specific areas for the AI to address, such as brevity and strength of message as well as a general letter grade. My current prototype is focused specifically on MUN opening speak for a principle committee. The AI is acting in the role of a dais member as the assess the speaker. The C# scripts underneath the program bring all of the elements together. For instance, timing of the message duration and the act of saving the recording are steps required before voice to text conversion is possible. In addition, my C# scripts were responsible for accessing the OpenAI API keys and uploading the text version of the user input. I am quite proud of my project because I got to piece together individual elements from the Unity software to an AI-powered response all the while achieving a working model. I was very pleased when showing the model to my MUN peers as a tool.

View this project on YouTube