⌨👂 audio keylogger

So I heard this could be done with a phone on a desk using the accelerometer, but what if we tried using keyboard + mic data first?

Wat

1. Record audio, create JSON keypress outputs
2. Train a model that can reconstruct keypresses from the audio.

$ pip install -r requirements.txt
$ ./record.py

Data goes into the ./recordings dir. When you’ve made a few, save the data:

$ ./save.sh

Wehn you’re done, upload to archive.org:

$ ./publish.sh

Future / todo

• Train whisper or similar with the outputs + evaluate with different sample rates etc.
• Create a Dockerized web app that uploads to archive.org, for mass data collection.
• Train something to separate keypress audio from background noise.