This week, Vitalis had no shortage of assignments and research to do on the subject. Rigorous research, deadlines regarding our product design documentation, preparing for the career fair, and a challenging start to midterm season for our team. Nevertheless, Vitalis powered through and has made some big steps this week!

During the prior weekend, Daniel got the ball rolling for what will be the foundation for our Automatic Speech Recognition (ASR) engine. Utilizing Open AI’s Whisper, Daniel created a small program that takes in audio data, converts it to a wav file, and runs it through a pre-selected Whisper model.

Eagerly, Vitalis began conducting small tests on the effectiveness on three different Whisper models by measuring word error rate (WER). We benchmarked the models in settings of Marston ambiance and isolated silence to gauge the efficacy of the specific model and microphone in real world environments.

Additionally, we began creating flow diagrams to itemize our software components to figure out how many AI models we will need in addition to statically-coded components. This aims to make it easier to allocate responsibilities during development and to make our design modular so components can be easily removed and added.

Lastly, our liaisons and coach, during this weeks meeting, were very impressed with our progress and we are perpetually excited to keep this momentum going forward into the stages of designing and prototyping.