SpeakFlow

Inspiration

Many of us have experience with using language-learning apps and while they teach vocabulary, many don't properly analyze the pronunciation. Our goal is to make people's speech sound more native.

What it does

SpeakFlow records your speech, analyzes your pronunciation, tone, and grammar using AI to make your speech flow better.

How we built it

The frontend is built using HTML, CSS, JS, and React to create a feature-rich but clear interface. The backend is programmed using Node.JS. The speech analysis works by feeding an audio recording of the user's speech to Microsoft Azure Speech AI to obtain the character spoken as well as the phonemes. Then, this information is fed into Claude Haiku in order to evaluate the accuracy of the user's speech.

Challenges we ran into

In our original design, we had many features that we implemented, but the UI that we built became too cluttered. So, we had to overhaul the UI to make it more user-friendly. Additionally, our original Claude Haiku prompt was too vague, so the responses felt too robotic. So, we have to engineer a prompt that felt more "human" and gave more constructive feedback about the user's speaking.

Accomplishments that we're proud of

Frontend is feature-rich, but is user-friendly
The AI model properly recognizes native vs. improper pronunciation ## What we learned
Better experience with frontend development
Learned how to use AI APIs