Inspiration

I've had trouble in the past with interviews, and speaking with confidence. It can especially be difficult to know your strengths and weaknesses while speaking, as nobody hears themselves in the same way other people hear them. Dicere solves this problem by being a quick and snappy way to get real-time feedback on your speech patterns.

What it does

Dicere provides feedback on speech patterns by analyzing and grading speeches and recommending ways to improve how we talk.

How we built it

Python/FastAPI backend (ported to typescript for better compatability with cf worker) using assemblyai API to process audio in real-time, the frontend is ReactJS

Challenges we ran into

Picking the right model: It was difficult to determine the approach to take for speech analysis. I considered OpenAI's whisper, but decided against this as is filters out filler words, which is the exact opposite of what you want for a project like this. Making the frontend: It took a lot of work to make the frontend look good, I'm just not that good at frontends :(

Accomplishments that we're proud of

Speed: Dicere process audio very quickly, and provides accurate feedback Aesthetics: I put in a lot of work to make the user interface look nice and make it easy to use Impact: Dicere packs a punch, it only takes a few minutes to use it and it provides quality feedback, especially considering how simple and accessible it is

What we learned

The nuance of language: I learned that speech patterns are incredibly unique. What works for a casual conversation may not work for a technical interview, which makes it difficult to define what is "good" and "bad." Processing audio data: Processing the audio data quickly and accurately could have been very difficult, but assemblyai simplified things substantially.

What's next for Dicere - Talk With Confidence

In the future I plan to allow users to track and save speech progress, as well as to select different prompt styles such as behavioral and technical interview prompts, creative prompts, or even unprompted speech analysis.

Built With

Share this project:

Updates