The voice is not a fingerprint, and that is precisely why method matters: specialist training in Forensic Speaker Identification (CEFAC), a hybrid perceptual-acoustic examination and probabilistic conclusions built to withstand adversarial scrutiny.
The market confuses identification, comparison, verification and speaker profiling. Choosing the right examination decides the value of the evidence; each one answers a different question.
The central task: the examination compares the questioned sample with the known sample of the person being compared and concludes in terms of probabilistic support, on the 9-point scale (from -4 to +4, with zero as inconclusive), assessing similarity and typicality: how alike the features are and how common they are in the population.
An umbrella term: placing an unknown speaker inside or outside a set of candidates. In a closed-set scenario, the speaker is among them; in an open-set scenario, he may not be, and the expert report states that difference before concluding.
Who is who within the same dialogue: wiretaps, group audio messages and recorded meetings with multiple voices have each utterance attributed to the corresponding speaker.
One-to-one authentication against an enrolled model: access logic, not evidentiary logic. Commercial biometrics is not to be confused with forensic examination, and the expert report explains why.
When there is no suspect, the speech describes the speaker: sex, age range, regional origin, level of education and possible pathologies, on a sociophonetic basis.
Faithful, standardised transcription: pauses, hesitations, overlaps and unintelligible passages marked, with phonetic hypotheses documented and a degree of reliability declared for each passage. Nothing to do with commercial or automatic transcription.
Cuts, insertions and edits leave traces: spectral discontinuity, jumps in background noise, recompression, codec signature. The path runs both ways: proving the editing or documenting the absence of traces, thereby supporting integrity.
Examination with declared methodology: vocoder artefacts, anomalous prosody, absence of the physiological micro-variations of the human voice (natural jitter and shimmer) and digital traces in the file itself.
Intelligibility enhancement that never creates content; analysis of non-vocal sound sources (gunshots, impacts, sequence of events) and reconstruction of the audibility of a scene; collection of voice exemplars under a directed protocol; rebuttal of another expert's report, drafting of questions to the expert and party-appointed expert support.
The examination combines the trained ear and the instrument, following the practice recommended by the IAFPA and by the ENFSI Best Practice Manual, conducted with specialist training in Forensic Speaker Identification (CEFAC).
When the material allows, perceptual-acoustic analysis is complemented by automatic speaker comparison systems, always under human supervision and interpretation: the machine computes, the expert concludes. And every examination goes through quality control: double-checking of findings, documentation that allows reproduction by another expert and declared attention to the mitigation of contextual bias.
The conclusion is expressed as a likelihood ratio: the expert report does not declare "it is the person"; it quantifies how far the evidence supports the same-speaker hypothesis against the different-speaker hypothesis, on a 9-point verbal scale. That is the way of concluding that withstands adversarial scrutiny.
The workflow in phasesBefore the voice examination, the forensic extraction of the original audio straight from the device (iOS and Android), before the application recompresses it, with hashing, binding to the device and documented chain of custody. The examination begins at the source of the evidence, not at the copy that survived being forwarded.
Methodological grounding: IAFPA · ENFSI Best Practice Manual for Forensic Speaker Comparison.
A declared limitation is a strength, not a weakness: it is what separates science from guesswork, and it is where fragile reports collapse.
Integrity of wiretaps and attribution of voices (who is who), covert recordings, threats sent by audio message and exclusion of a suspect: negative proof is also a result, and it has already cleared innocent people of accusation.
Harassment and offensive remarks in messaging groups, authenticity of a meeting recording, denial of authorship of audio attributed to an employee or employer.
Audio in custody disputes and allegations of parental alienation, threats between former spouses, verification of tampering in home recordings.
CEO fraud using a cloned voice, authenticity of meeting recordings, internal investigations and validation of evidence from the whistleblowing channel.
Comparative phonetic analysis concluded against identification, excluding the person as the author of the questioned speech.
Video with questioned audio? Video authenticity, deepfake detection, photogrammetry and CCTV complete the examination of the recording.
Chain of custodyThe foundation that supports everything: extraction of the original from the device, hashing and preservation. Voice, image and data examined under a single signature: complete multimedia evidence.
Before deciding, it is worth seeing what has already come through this laboratory: cases described without identifying the parties, in the format of challenge, method and result.
The intruder's control channel was written into smart contracts. The examination decoded what he had deleted and handed the authorities concrete routes to identification.
See the case Negative proofThe official examination had concluded that he took part. The re-examination showed the links were false positives, and the accused person was cleared.
See the case