TL;DR: We used a 1-nn classifier to extract flags from spectrograms of audio files.

Original writeup (https://secgroup.github.io/2017/05/29/faustctf2017-writeup-alexa/).