ENF is Electrical Network Frequency Analysis. It's a method of working out when an audio recording was made by listening for mains hum interference on the audio.
Wait, what? - that seems like magic
All electrical grids use a form of electricity called alternating current (AC) which oscillates between a maximum and minimum voltage either 50 or 60 times a second. This osciallation can create interference from electrical appliances like phone-chargers, kettles or microwaves. If someone makes an audio recording near an electrical appliance interference from the appliance will often appear on the recording as an audible hum which oscillates at the same frequency as the grid. The frequency of the grid is almost exactly either 50 or 60 Hertz, but it varies by a few percentage points every second due to changes in demand and when the grid frequency changes, the frequency of the hum on the recording changes too. Because electricity grids usually cover a wide geographical area the changes in frequency will be roughly the same wherever on the grid the audio was recorded, so an audio recording made on the Central European Grid will have the same changes in intereference frequency whether it was made in Lisbon or Istanbul.
Some grid operators publish the second-by-second changes in frequency. By comparing the public records of frequency changes to mains interference on recordings it's possible to work out exactly when the recording was made, sometimes down to the second.
Tom Scott made a nice video which goes into more detail:
What's Open ENF?
Electrical Frequency Analysis tools are currently only available to law enforcement agencies. Open ENF is an attempt to create a public ENF system that anyone can use
Will it work with any audio or video?
Nope. Many recordings - especially ones made outdoors - don't contain any mains interference. For other recordings the signal is too weak to make a reasonable match. OpenENF is improving all the time, but it currently works best when:
The recording is made indoors
The audio is between 5 and 20 minutes long
You have a rough idea of when the recording was made. For example, you might know that the recording was made in August 2020, but you're not sure exactly which day.
How quickly can Open ENF find a match?
That depends on the quality of the recording and the time span over which you're searching. A 5 minute recording with a strong hum and a search range of a couple of months is typically completed in under a minute. The maximum timespan you can currently seach over is 3 years and currently searches that take more than 10 minutes to complete will timeout. This usually happens if you're submitting more than 20 minutes of audio and searching over a couple of years. We're working to improve performance but bear in mind that Open ENF is in its beta-testing phase and is currently free to use. If you want to search over a longer time period email search@openenf.io to see if it's possible with your recording.
How accurate are the timestamps made by Open ENF?
Again, that depends on the quality of the hum signal in the recording. A strong hum signal on a recording of 10+ minutes is likely to return a more accurate result than a weak, 30-second recording. However, even in the most ideal case is very difficult to establish the time of recording with 100% accuracy, since no recording perfectly matches the data. We're working to provide DNA-test-like accuracy so we'd be able to say There's a 98% chance this recording was made at 12:54PM on 1st August 2019, but for now we classify scores as either strong, medium or weak.
Does it work anywhere in the world?
Sadly not. At the time of writing only 4 grids publish consistent up-to-date grid frequency data. These are the Great British grid (covering Great Britain but not Northern Ireland), the Irish grid (which covers the entire island of Ireland), the Nordic grid (covering Norway, Sweden and Finland) and the Central European grid (which covers most countries in mainland Europe plus all of Turkey, Morocco, Algeria and Tunisia.)
OpenENF currently works with the Great British and Central European Grids, and we hope to have the Nordic and Irish grids searchable in early 2023. We're also exploring ways of making grid data searchable for other grids.
Is the code open source?
Not yet, but we hope to be publishing the frequency analysis and search code to a public repository very soon.