Contents
An audio file rarely carries a GPS tag the way a photo does, so there is usually no location field to delete, but audio does carry a location signal of its own, the electric network frequency, a faint mains-hum trace of which power grid the recording was made on, and that rides in the waveform where a metadata strip cannot reach it.
Audio usually has no GPS tag to delete
With a photo, removing location starts with the GPS tag, a pair of coordinates written straight into the EXIF block. Most audio files have no such field. A voice memo or a music file carries title, artist, timestamp and device fields, but not a latitude and longitude, so the “delete the location tag” step that dominates remove GPS and location from a photo is usually empty for audio.
There are exceptions. Some field recorders, body-worn devices and phone apps do write a location field into the container, a custom atom or a Broadcast WAV coding-history note, and if yours does, clear it: ExifTool or ffmpeg removes it along with the rest of the tag block. But that is the easy half, and for audio it is often already empty. The location problem for audio is not in the tag. It is in the sound.
The real location signal is the mains hum
Mains electricity hums at a nominal 50 or 60 hertz, and that frequency drifts slightly from moment to moment, in a way shared uniformly across a whole synchronous power grid. Any recorder running on or near mains power picks that hum up, and it is captured into the recording as the electric network frequency, or ENF. Grigoras (2005) set out the founding ENF criterion at the IAFPA meeting: because the whole interconnected grid carries the same fluctuating frequency, the captured hum is a shared stamp that can be checked against a reference log.
That stamp carries place. Hajj-Ahmad, Garg and Wu (IEEE TIFS 2015) showed that ENF “is a signature of power distribution networks that can be captured by multimedia signals recorded near electrical activities,” and that its statistics can infer which grid a recording was made on, continental Europe, the United Kingdom, North America, rather than the street. The resolution is improving. Chowdhury and Sarkar (2019) find that ENF variations “are inherently located in a multimedia signal which is recorded close to the grid,” and that a recording’s location “can be localized by analyzing the ENF sequences of that signal” against grid references around the world. Garg, Hajj-Ahmad and Wu (2021) go further in a feasibility study, using correlation against known anchor points to move from continent-scale toward city-scale placement. The trace is in the waveform, not the tags, so a metadata strip does not touch it.
The microphone fingerprint is a second waveform trace worth knowing about here, since it survives a strip for the same reason the ENF hum does. Buchholz, Kraetzer and Dittmann (IH 2009) identified microphones from their recordings at up to 93.5 percent accuracy, a device signature that a tag strip leaves in place just as it leaves the ENF stamp. Both are covered in depth, with how far each can be pushed down at a cost, in strip ENF and microphone fingerprint from audio.
Why you cannot cleanly strip the ENF
The obvious idea is to notch out the 50 or 60 hertz line. The trouble is that the nominal frequency is stable and narrow, so the notch has to be narrow too, which leaves the higher harmonics of the hum behind, and it replaces the natural mains band with an unnaturally clean gap that is itself a flag. Chuang, Garg and Wu (IEEE TIFS 2013) studied exactly this, describing methods that “remove and alter the ENF signal while trying to preserve the host signal, and devises detection methods targeting these operations.” In other words, the research on removing ENF comes packaged with research on detecting the removal. The tag is the label; the mains hum is the evidence, and the evidence does not peel off.
Where to go next
For audio, location is mostly not a tag you can delete, it is a signal you can at best degrade. Clear any container location field if your recorder writes one, then be honest that the ENF grid stamp and the device signature remain in the sound. The general rule for audio metadata, container off easily and signal not, is in remove metadata from an audio or MP3 file, and the depth treatment of both waveform traces is in strip ENF and microphone fingerprint from audio. The photo version of the location problem, where the GPS tag and the visible scene are the two halves, is in remove GPS and location from a photo. And if you need to establish where or when an unknown recording was made, see what forensics can learn from a recording.
Sources
- Grigoras (2005). Digital Audio Recording Analysis: The Electric Network Frequency Criterion. IAFPA.
- Hajj-Ahmad, Garg, Wu (2015). ENF-Based Region-of-Recording Identification for Media Signals. IEEE TIFS.
- Chowdhury, Sarkar (2019). Location Forensics Analysis Using ENF Sequences.
- Garg, Hajj-Ahmad, Wu (2021). Feasibility Study on Intra-Grid Location Estimation.
- Chuang, Garg, Wu (2013). Anti-Forensics and Countermeasures of Electrical Network Frequency Analysis. IEEE TIFS.
- Buchholz, Kraetzer, Dittmann (2009). Microphone Classification Using Fourier Coefficients. IH 2009.