Check out https://www.sinewavespeech.com/ to try out sine wave speech in an interactive way.
The code for the website lives under frontend/.
Sine wave speech is a method of converting speech to sounds that are utterly unintelligible on the first listen, but with a little practice – hearing a few before-and-after examples – become easy to follow.
This repo also contains a Python package for converting audio recordings to sine wave speech. For an example of how to use the package, see sine_wave_speech_example.ipynb.
In short, using fancy signal processing methods (LPC), we determine the four most important frequencies at every point in time and only keep those.
You can see on the spectrogram of the modified audio that only four frequencies are ever active at a time:
Listen to sentence-sine-wave.wav.
Check out this page for a detailed explanation.