Audio deepfake detection · Built solo by Alex Picon
Would you trust this voice?
A convincing voice is no longer proof of a person. Test your ears against real and synthetic recordings of the same speaker, then put a recording through HEARSAY and inspect the evidence, not just a verdict.
Trust your ears? Try it.
Same voice. Which recording is fake?
Two recordings attributed to the same person: one genuine, one synthetic. Listen to both, then choose. No access code needed.
Loading the listening game…
Recording B
Not played yetLoading the listening game…
What does this game measure?
Your score is for this session and these demonstration clips only. Each pair contains exactly one synthetic recording, so random guessing averages 50%. That setup is different from screening an unknown recording. Speakers and clips may overlap training data; this is not a scientific test of you or HEARSAY.
The reveal uses dataset labels, not a HEARSAY prediction. Audio and labels are public; the game is for learning, not a secure exam.
Recording sources and creditsA familiar voice is no longer enough
When a voice sounds right, a false story can travel.
A fabricated recording can impersonate a relative asking for money, a manager requesting a transfer, or a public figure saying something they never said. Voice synthesis also supports accessibility and creative work. The risk is deception: passing generated speech off as a real person's words.
Listen critically. Verify independently.
If a caller makes an urgent request, pause and contact the person using a number you already trust. FTC guidance on fake emergencies.
Reported imposter-scam losses
Reported to the FTC for 2025. This includes impersonation across many channels; it is not a voice-cloning loss estimate.
FTC · June 2026Fake speech spotted in one study
A 2023 study with 529 listeners tested English and Mandarin synthetic speech. This is a study result, not a universal human detection rate.
UCL / PLOS ONE · August 2023From a gut feeling to a checkable result
What HEARSAY adds
- Listen or upload. Choose a library clip or your own recording.
- Run a detector. HEARSAY analyzes learned audio features in an isolated worker.
- Read the result in context. Compare modes, check coverage and verify provenance before acting.
Speed
- Warm run
- 0.89 s
- First load
- 8.64 s
- Coverage
- First 12 seconds
Accuracy
- Warm run
- 2.28 s
- First load
- 14.12 s
- Coverage
- Three spaced 4.0375-second windows
Deployment smoke checks on September 27, 2026: warm timings use synthetic-2.flac; first-load timings use synthetic-1.flac. CPU server: 6 vCPUs, 16 GiB RAM; workers limited to 2 CPUs. These are individual observations on short samples, not latency guarantees or an accuracy benchmark. Accuracy mode's name is not a promise that it wins on every recording.
Reading a decision score
0.5 cutoff
leans genuineleans synthetic
0.91 decision score, not a probability; above the 0.5 cutoff, so it leans synthetic.
What a score can and cannot tell you
Higher scores lean synthetic. A score of 0.9 does not mean a 90% chance of being fake: these are decision scores, not calibrated probabilities. New generators, languages, background noise and encoding changes can shift performance.
Why even a good detector needs context
Illustrative example: screen 1,000 recordings with a hypothetical detector that catches 90% of fakes and falsely flags 5% of genuine audio.
Teaching assumptions, not HEARSAY performance measurements.
Inspect the evidence
Analyze audio
1. Pick audio
Download four test clips (ZIP)
Demonstration samples may have training overlap; they are not a fair accuracy benchmark.
2. Run it
Comparisons
Jobs run sequentially. Results refresh every two seconds. Interpretations use a 0.5 decision cutoff, not calibrated confidence.
No comparisons yet. Upload a recording to begin.
Isolated by design
Offline workers · One detector at a time · No writable host mounts · Bounded analysis time.
Not a verdict
Scores are for the data they were measured on; they do not transfer to every recording.
Credit where due
Model sources and usage limits · Recording sources · Portrait credits