VerbalFlow | Devpost

Landing page
Speech recording view

Inspiration

As undergrad and high school students, we have to go through various interview processes for internships and CO-OPs. We would also all benefit greatly from polishing our public speaking and communication skills. So we built a tool to help you become a better speaker.

What it does

Verbalflow gives its users tailored feedback on their communication skills using various APIs. It collects data about the user's face using vgg-face and facenet APIs, and it also records the user's word choice and tone. It then feeds all of the data it collected into groq's API to create it's feedback.

How we built it

We built Verbalflow over 36 hours using various languages and APIs such as Python and Flask for the back-end, and HTML and CSS for the front-end. We used Flask to host a web server and our files on the frontend, and HTTPS requests to send the data that was gathered from the frontend to the backend. After receiving the data in the backend, Groq analyzes the data and curates feedback based on the information.

Challenges we ran into

While building the speech recognition part of our project we ran into audio corruption problems; every time we tried to receive audio data from the frontend it would always be unusable. We eventually, did some digging and found out that Google didn't support the file type that we were trying to send. But when it did work, we started getting memory and pointer-related issues on top of more audio corruption.

Accomplishments that we're proud of

In a short demo, we were able to have it successfully gather data on my speech, tone and facial language, and provide feedback.

What we learned

We learned many new languages since most of us had little experience with the tech stack we worked with such as Flask and Javascript. On top of that, we learned that sometimes bugs can be things, you would never have suspected to root of the problem, but with a lot of time and research, it's possible to find any bugs.

What's next for VerbalFlow

In the future, we are thinking of implementing other methods of data collection to make the feedback it gives even better, such as retina tracking, gesture, and posture tracking. We are also going to improve the feedback it gives and make it feel more personalized and human.