An AI voice deepfake detection service that accepts audio (MP3, WAV, M4A, WebM) or video files and returns the probability that the speech is machine-generated, with model attribution when identifiable (ElevenLabs, Resemble, PlayHT, OpenAI, and others). It is delivered as a web detector, an open-source Chrome extension (Manifest V3) that detects audio on WhatsApp Web, YouTube, and podcasts, and a REST API with JSON in/out and webhook batch processing. Each verdict returns a permanent citation URL in APA style and a one-way audio fingerprint; raw audio is deleted within 24 hours rather than stored. The design is deliberately calibrated—when probability lands near 50%, it states that the result is genuinely uncertain rather than rounding toward a confident answer—and every verdict carries a methodology version label (e.g., v3.2) with monthly model retraining. A free tier is available without an account, with paid plans required beyond roughly 100 verdicts per month; API access is included from the Starter tier upward. The service is registered in England and Wales, with a London registered office address, though the operating company name, team, and funding are not disclosed.
Pros & Cons:
Pros:
Returns model attribution alongside a probability, which is more useful than a bare “AI or not” score for newsroom and legal verification; every verdict includes a permanent, APA-style citation URL designed for editorial and evidentiary use; raw audio is deleted within 24 hours and only a one-way fingerprint is retained, a better privacy posture than services that store uploads; explicitly calibrated to admit uncertainty near 50% instead of inflating confidence; open-source Chrome extension (Manifest V3) lowers integration friction for in-browser verification; REST API with webhooks and batch processing suits production pipelines.
Cons: Independent third-party validation is absent—no academic benchmarks or independent reviews surfaced, unlike Pangram (NYT/University of Chicago) or Originality.ai (Epoch AI), so claimed accuracy is unverified; the operating company is not disclosed—only a London registered-office address, with no team, funding, or named customers; the Chrome extension appears in two store listings with inconsistent ratings (3.7 from 6 reviews versus 4.7 from 380), which makes the user-feedback signal unreliable; free-tier limits beyond “about 100 verdicts per month” are not detailed on the API page and require checking the pricing page; verdicts are probabilistic and the vendor itself states it does not guarantee any particular outcome.
Who is it for?:Newsrooms and fact-checking teams verifying whether audio clips are synthetic; legal and compliance teams that need a citable probability and model attribution; platforms and developers embedding voice authenticity checks into upload or moderation pipelines; security teams investigating voice-cloning fraud such as grandparent and vishing scams.