The year is 2030 Deep-fake have become common nowadays and some of your mischievous friends have deep-faked your face and put it on a porn star and send it to your girlfriend (assuming you have one) and now you have to prove yourself innocent.But as an Electrical Engineer you have trick on your sleeve that is Electrical network frequency to understand first you have to understand how does power system makes it's iconic hum.
The alternating current running through power lines changes direction constantly. A full cycle back and forth, 50 times a second 60 in some of the parts of the world, but we will consider 50 times a second. But that transmission isn't perfectly efficient. There are imperfections and stray electromagnetic fields. And a little bit of that power becomes vibration, which makes sound at that frequency of 50 hertz. Or 100, or other multiples of 50. That's the "mains hum" you can sometimes hear coming from power lines, or maybe from speakers that aren't quite plugged in properly.
Figure a) Transmission line
Also, that hum isn't exactly 50 cycles a second. The grid frequency shifts slightly and constantly over time as the engineers in charge balance supply and demand. It's only a tiny fraction of a shift, but it's there. The whole grid runs at the same, slightly wobbling, almost-50 hertz, no matter where you are.
That mains hum can be really annoying for people recording audio. You can hear it in the background sometimes. But if you were to log what that wobbling grid frequency was, every second of every day, keep all that in a database, and then compare it to recordings with mains hum on them, then in theory, you'd have a perfect watermark to prove the time when anything was recorded.
b) Picture showing how ENF data is analyzed
In more technical terms these method is called Electrical network frequency analysis. It may be used to establish if and where the recording has been edited. Therefore, the ENF criterion provides a powerful method to assess the evidential integrity of audio data that has been recorded onto audio, video, computer and telecommunications equipment.
It's fairly simple,signal processing. You're looking to see if there is a component around 50 hertz or one of the harmonics.
The longer recording you have, the better. As an absolute minimum, about a minute.If it's a noisy recording,or if the ENF isn't particularly clear,then the longer recording you have, the better. - So: find the mains hum frequency in the footage and then write code to match against the millions of possibilities in the National Grid logs.
So how this all will help you ?It's fairly simple, if we edit any kind of video the ENF analysis will reveal that there was a jump in the frequency and it will not match our records.
So there you have it a good solution to defend against deep-fake.
Besides deep-fake there are many other applications of ENF analysis to like catching speed run in a game, timestamp a crime scene.
Gamer's among you must know about speed-run. And many of the players cheat in that ,by doing ENF analysis we can catch these cheaters i.e., if they have edited the video the sequence will not match to that of the gird sequence. Time stamping of a video can also be done using this method and there are other tons of application of this.The Police in the German federal state of Bavaria are keeping track of this since 2010 and the United Kingdom Police since 2005.
For more tech savvy readers i have made a Matlab script that performs this analysis.
the google drive link is below.
Abhishek! Everything is done rightly in this blog. I remember once we were discussing the electric hum; at that time I was a little bit confused, but your blog cleared my confusion.