# Modern deepfake detection uses 'liveness detection' - identifying subtle markers that distinguish human-generated from AI-generated content

**Date:** 2025-12-18  
**Tags:** Security, Deepfakes, AI  
**URL:** https://kelexine.is-a.dev/til/deepfake-detection-liveness

---

TIL: Modern deepfake detection uses 'liveness detection' - identifying subtle markers that distinguish human-generated from AI-generated content. For voice, this includes micro-variations in breath patterns, tonal shifts, and acoustic fingerprints that current synthesis can't perfectly replicate.


```python
def detect_synthetic_audio(audio):
    features = {
        'breath_patterns': analyze_breath(audio),
        'spectral_flux': compute_spectral_variance(audio),
        'micro_pauses': detect_natural_hesitation(audio),
        'artifacts': scan_for_synthesis_markers(audio)
    }
    return liveness_model.predict(features)
```




---

*This content is available at [kelexine.is-a.dev/til/deepfake-detection-liveness](https://kelexine.is-a.dev/til/deepfake-detection-liveness)*