Meow Censor

Screenshot of site

Inspiration

We are tired of those f-[CENSORED]-ing audio bleeps! They're annoying, grating, and sound really unnatural! So we came up with a way to meow-censor your audio, with an simple and cute tool!

What it does

Takes in audio (upload or record), outputs the audio with the foul language censored with a REAL meow of the appropriate duration.

How we built it

Building the Meow Library

In the prepare-library.py script, we used the freesound-API to scrape & download Creative Commons meow sounds, and process them to create: 1) folder of processed meow sounds, with blank space trimmed 2) a .csv with these cols: filename,path,duration_seconds,average_pitch_hz

How It Works

The application operates as a three-stage pipeline: 1) Transcriber Agent: openai-whisper transcribes the source audio, generating word-level timestamps.

2) Censor Agent: The transcript is sent to a Google Gemini model via the Agent Development Kit (google-adk) to identify the timestamps of profane words.

3) Audio Surgeon Agent: pydub slices the original audio, finds the best-fitting meow from a pre-analyzed library, normalizes its volume, time-stretches it, and splices it into the final audio track.