VoiceprintDeepfakeDetector<T>
Models & Types · AiDotNet.Safety.Audio
Detects deepfake audio by analyzing speaker voiceprint consistency throughout the recording.
Inheritance: ModelBase<T, Vector<T>, Vector<T>> → SafetyModuleBase<T> → AudioSafetyModuleBase<T> → VoiceprintDeepfakeDetector<T>
For Beginners
Each person’s voice has unique characteristics — like a fingerprint. Real speech keeps these characteristics consistent. Voice cloning often produces subtle inconsistencies that this module detects by comparing “voice fingerprints” from different parts of the recording.
How It Works
Extracts short-term speaker embeddings from overlapping frames of the audio and measures consistency across the recording. Real speech has stable speaker characteristics (pitch contour, formant patterns, spectral envelope) while cloned/synthesized voices often show temporal inconsistencies in these features — either too stable (robotic) or with sudden jumps (spliced segments).
References:
- VoiceRadar: Voice deepfake detection framework (NDSS 2025)
- SafeEar: Privacy-preserving audio deepfake detection (ACM CCS 2024)
- Voice cloning detection via speaker verification mismatch (2024)
Constructors
| Constructor | Summary |
|---|---|
VoiceprintDeepfakeDetector(double threshold, int sampleRate) |
Initializes a new voiceprint deepfake detector. |
Properties
| Property | Summary |
|---|---|
string ModuleName { get; } |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<SafetyFinding> EvaluateAudio(Vector<T> audioSamples, int sampleRate) |