Contents
A recording can reveal when it was made, sometimes to the second, the electrical grid region it was made in, the class of room and microphone behind it, and whether it was spliced or copied. The strongest of these, dating a clip from the faint mains hum, is court-accepted. Most of the rest name a class rather than an identity, and all of them fade as the file is re-encoded. This is the recording-specific companion to what forensics can learn from a file: the audio traces an examiner works through, job by job.
When: the mains hum is the hardest trace
The richest time signal in a recording is the Electric Network Frequency, the faint 50 or 60 Hz hum a microphone picks up from the power grid, which fluctuates identically across a whole synchronous area. Matched against a continuous frequency archive it can pin a recording’s time at court-accepted strength: Cooper (2009) matched a 70-minute Glasgow recording to a London archive 676 km away, and a 2-minute extract across a 36-day archive. It is the only audio trace with its own statutory best-practice standard (ENFSI, 2009). The limits are physical, not statistical: it needs mains coupling at capture and roughly 10 minutes or more of audio, and it is often simply absent.
Where: a grid region, not an address
Location resolves to a class, never a street. From ENF fluctuation statistics alone, Hajj-Ahmad, Garg and Wu (2015) classify the grid of origin, continental Europe against the United Kingdom against US-East, at about 81 to 88 percent, but because one synchronous grid carries the same frequency at every outlet, the trace names the grid, not the city. Acoustics add the room: Moore, Brookes and Naylor (2013) identified the room class at 96.1 percent across 22 rooms, yet identical floor plans confused about 30 percent of trials, and Baum and colleagues (2022) estimate a blind roomprint from speech alone at up to 93.6 percent on test recordings, dropping sharply once the microphone position changes. The ceiling is a grid region plus a room class plus a setting category, never an address.
What device, and edited where
Device questions answer at model or class level, closed-set, and only against an enrolled reference. Garcia-Romero and Espy-Wilson (2010) reached 99.0 percent identification on longer microphone segments and 93.2 percent on 3-second handset clips, but open-set naming of an arbitrary unknown device is unsolved, and two units of the same model are near-indistinguishable (how accurate is microphone fingerprinting). Editing is read as a continuity problem: a sample-domain splice seam reaches a true-accept rate of 0.965 at zero false acceptances on uncompressed audio (Korycki, 2013), an ENF phase break locates a single edit at about 6 percent equal-error (Nicolalde-Rodriguez, Apolinário and Biscainho, 2010), and an environmental-signature check localizes a splice even after MP3 re-encoding (Zhao, Chen, Wang and Malik, 2017). Every one of these is supporting, not decisive, and the governing standard states it plainly: “evidence of discontinuities is not evidence of editing” (ENFSI).
What has happened to it since
A recording carries its own processing history, recoverable only up the quality gradient. Whether a file was ever lossily compressed is detectable at about 98.6 percent, independent of the codec (Hennequin, Royo-Letelier and Moussallam, 2017), but a terminal re-encode overwrites the earlier chain, so a redistributed file reveals far less than an original. Re-recording is the reset event: playing audio back and capturing it again destroys the upstream device, codec and ENF traces while acquiring a fresh, internally consistent chain, which is why detecting it decides whether every other read describes the original or a copy. Removal is itself a finding, because scrubbing the mains hum leaves a detectable spectral hole and inter-harmonic inconsistency (Chuang, Garg and Wu, 2013).
How far to trust the read
No audio-forensic method outputs “authentic.” Each weighs whether one property of the file is consistent with an untouched original, reported as a strength of support for a proposition rather than a verdict (ENFSI, 2015), and the field has no catch-all test of authenticity. The report reads like a stack: the recording is consistent with a capture in this grid region, in this room class, through this device class, with no detected continuity break, after this last encoding stage. Anything stronger needs reference data, archive coverage and a preserved original. A finding earns weight only when independent traces agree, the same bounded, corroborated standard that governs the whole field. The metadata spoke goes deeper at can audio metadata be faked.
Sources
- Cooper, A. J. (2009). The Electric Network Frequency (ENF) as an Aid to Authenticating Forensic Digital Audio Recordings: An Automated Approach. AES 33rd International Conference on Audio Forensics.
- European Network of Forensic Science Institutes (2009). Best Practice Guidelines for ENF Analysis in Forensic Authentication of Digital Evidence (FSAAWG-BPM-ENF-001).
- Hajj-Ahmad, Garg, Wu (2015). ENF-Based Region-of-Recording Identification for Media Signals. IEEE Transactions on Information Forensics and Security 10(6):1125-1136. DOI: 10.1109/TIFS.2015.2398367
- Moore, Brookes, Naylor (2013). Roomprints for Forensic Audio Applications. IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA) 2013.
- Baum, Cuccovillo, Yaroshchuk, Aichroth (2022). Environment Classification via Blind Roomprints Estimation. arXiv:2209.07196.
- Garcia-Romero, Espy-Wilson (2010). Automatic Acquisition Device Identification from Speech Recordings. IEEE ICASSP 2010:1806-1809.
- Korycki, R. (2013). Time and Spectral Analysis Methods with Machine Learning for the Authentication of Digital Audio Recordings. Forensic Science International 230(1):117-126. DOI: 10.1016/j.forsciint.2013.02.020
- Nicolalde-Rodriguez, Apolinário, Biscainho (2010). Audio Authenticity: Detecting ENF Discontinuity with High Precision Phase Analysis. IEEE Transactions on Information Forensics and Security 5(3):534-543. DOI: 10.1109/TIFS.2010.2051270
- Zhao, Chen, Wang, Malik (2017). Audio Splicing Detection and Localization Using Environmental Signature. Multimedia Tools and Applications 76(12):13897-13927.
- Hennequin, Royo-Letelier, Moussallam (2017). Codec Independent Lossy Audio Compression Detection. IEEE ICASSP 2017.
- Chuang, Garg, Wu (2013). Anti-Forensics and Countermeasures of Electrical Network Frequency Analysis. IEEE Transactions on Information Forensics and Security 8(12):2073-2088. DOI: 10.1109/TIFS.2013.2285515
- European Network of Forensic Science Institutes (2015). ENFSI Guideline for Evaluative Reporting in Forensic Science (STEOFRAE).