Inspiration

Nobody likes hearing profanities.

What it does

Based on a person's speech patterns, it predicts if the word "duck" is about the be said and then bleeps before it is.

How I built it

I used Google's Cloud Speech API to record what people say, and I created an algorithm that periodically trains an LSTM network with that information. The application repeatedly predicts the next four letters and beeps if they spell "duck". I used Neataptic, a neural network library for javascript, and node.js.

Challenges I ran into

-Ignoring the strange looks every time I said "duck"

-The Cloud Speech API did not work as well as I'd hoped. It also had a time limit, so I had to cut the training short -- which meant it was accurate sometimes, but I also saw a lot of spaces (the error wasn't reduced enough, so the first output rounded up to a 1).

What I learned

Basic concepts of ML

What's next for Bleeper

I should also look for a speech-to-text API without a time limit.

Built With

Share this project:
×

Updates