Aletheia · Audio deepfake detection

#1 in commercial latency.

24 ms per clip, with 97.3% of deepfakes detected. Results from the Podonos benchmark across 23 systems.

Evaluate AletheiaView source
Latency per clip
24ms#1 among commercial systems
Processing speed
2.3×the next-fastest commercial system
Deepfakes detected
97.3%of synthetic audio in the test
Files scored
4,524Every file. None declined.
Detection accuracy
94.47%#9 of 23 systems

Speed and detection accuracy

Higher is more accurate. Further right is faster.

Detectif.aiOther commercial systemsAccurate and fast: 90%+ accuracy, 10×+ real time
70%80%90%100%0×25×50×75×100×125×Accurate and fastDetectif.ai: 94.47% accuracy, 119.0× real-time speed, 24 ms per clip.Detectif.aiNII Synthetiq Audio v0.8-Beta: 89.57% accuracy, 41.7× real-time speed, 91 ms per clip.Corsound AI: 87.79% accuracy, 28.6× real-time speed, 180 ms per clip.Corsound AIPindrop: 95.05% accuracy, 13.2× real-time speed, 282 ms per clip.Hive: 83.53% accuracy, 2.9× real-time speed, 881 ms per clip.HiveWhispeak: 97.7% accuracy, 2.6× real-time speed, 1.1 s per clip.Reality Defender: 71.27% accuracy, 0.7× real-time speed, 5.7 s per clip.Reality Defender

Processing speed · 1× = real time

94.47% accuracyat 119× real-time speed

Commercial latency ranking

Average time per audio clip. Lower is faster.

  1. 1Detectif.ai24 ms
  2. 2Pella Research57 ms
  3. 3NII Synthetiq Audio91 ms
  4. 4Corsound AI180 ms
  5. 5Pindrop282 ms
  6. 6Resemble DETECT-World399 ms
  7. 7Fennura553 ms
  8. 8Hive881 ms
  9. 9Whispeak1.1 s
  10. 10Resemble AI1.2 s
  11. 11Reality Defender5.7 s

Reported timings · bars use a log scale

Accuracy scored by Podonos. Detectif.ai timings are vendor-reported; hardware and service conditions vary. The latency ranking includes the 11 commercial systems with published timing.

Podonos source · September 2026

How it was tested

4,524 audio clips, split evenly between genuine and synthetic speech. Podonos scores submitted predictions against private labels.

These are the main benchmark results. The 8 kHz and 16 kHz telephone-audio tracks are separate evaluations.

Detection results

Deepfakes missed
2.7%
Genuine audio flagged
8.4%

The 94.47% accuracy result includes both genuine and synthetic audio.

Benchmark results by system
22 systems, ordered by accuracy.
SystemAccuracyDeepfakes detectedLatency per clipSpeed vs real time
Resemble DETECT-World99.47%99.6%399 ms8.3×
Fennura98.63%98.5%553 ms7.1×
Aurigin AI98.21%98.9%Not reportedNot reported
Resemble AI98.05%98.6%1.2 s2.5×
Whispeak97.70%98.3%1.1 s2.6×
Pella Research95.82%97.2%57 ms47.6×
Pindrop95.05%96.3%282 ms13.2×
Detectif.ai94.47%97.3%24 ms119.0×
NII Synthetiq Audio v0.8-Beta89.57%85.8%91 ms41.7×
Corsound AI87.79%76.9%180 ms28.6×
Hive83.53%69.5%881 ms2.9×
Reality Defender71.27%96.4%5.7 s0.7×
NII AntiDeepfake70.47%41.4%91 ms41.7×
Wav2Vec2 (2019 LA)62.89%39.2%622 ms7.1×
AST (ASVspoof 5)56.83%82.6%5 ms588.2×
Wav2Vec2 (2024 mix)55.55%44.2%219 ms17.9×
Deepfake-V2 (W2V2-base)53.03%9.1%94 ms37.0×
AST (VoxCelebSpoof)50.99%2.5%8 ms333.3×
RawNet2 (2019 LA)50.66%37.3%94 ms28.6×
LCNN-LFCC (2019 LA)50.00%100.0%23 ms178.6×
AASIST (2019 LA)48.17%48.9%322 ms9.1×
AASIST3 (ASVspoof 5)47.63%1.6%363 ms7.7×

Accuracy is calculated on the files each system accepts. Corsound AI and Reality Defender rejected some files, so their results cover a smaller sample. Aurigin AI did not submit timing. Processing speed is the reciprocal of the reported real-time factor; systems use different hardware, analysis windows and service conditions.

Evaluate it on your calls.

Test Aletheia on your channels, languages and voice generators.

Request an evaluationRead the research