SoapyMouth

Transcription

Inspiration

One of our team members is a DJ that tirelessly spent hours upon hours removing profanity from words to make them age appropriate for events. We thought, there has to be a better way, where we can detect these words in real-time and catch them before they make it out through the speaker. That's how we came up with SoapyMouth.

What it does

Captures Audio:
Continuously records audio from your microphone in small chunks.
Transcribes Speech:
Uses a real-time local speech-to-text engine (faster-whisper) to transcribe each chunk of audio as you speak.
Detects Swear Words:
Checks each transcribed chunk for the presence of swear words using a customizable profanity filter.
Censors Audio:
If a swear word is detected, after a short delay, the corresponding audio chunk is muted and a beep is played through your speakers, ensuring that offensive language is not broadcasted.

How we built it

First we experimented with multiple transcription engines to get the fastest performing one. Then we settled on a local model of whisper. We fed that into a word list detection algorithm while also routing the microphone input to the app. The app detects if the word is in the blacklist, and censors the corresponding audio output and outputs the rest through the device speakers.