Spoofprint Embeddings for Unknown Voice Deepfake Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional voice biometric systems are vulnerable to spoofing attacks from advanced speech synthesis technologies, lacking the ability to generalize and detect unknown spoofing techniques effectively.

Innovation Solution

A neural network architecture is employed for spoof detection, utilizing embedding extractors to differentiate between voiceprint and spoofprint features, with a large margin cosine loss function to maximize variance between genuine and spoofed classes, and minimize intra-class variance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional voice biometric systems are used for speaker recognition, then the system is simple to operate and implement, but the system becomes vulnerable to spoofing attacks from advanced speech synthesis technologies

Engineering Contradiction:
Improvespoofing detection capabilityVSAvoidneural network architecture complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system segments the voice biometric verification process into two independent pathways: a voiceprint extraction pathway for speaker identification and a spoofprint extraction pathway for spoofing detection. This segmentation allows the system to maintain simple speaker recognition functionality while adding specialized spoofing detection capabilities through separate neural network branches that process different acoustic features.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system introduces an intermediary spoofprint embedding layer that acts as a mediator between the raw audio input and the final verification decision. This intermediary component extracts spoofing-specific features (such as spectral artifacts and temporal inconsistencies) that are distinct from voiceprint features, enabling the system to detect synthesized speech without compromising the original voice recognition functionality.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If conventional voice matching approaches are used, then the system operates efficiently with simple processing, but the system becomes obsolete due to improvements in speech synthesis tools capable of fooling these systems

Engineering Contradiction:
Improvedetection of unknown spoofing techniquesVSAvoidprocessing time for dual embedding extraction
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system merges the voiceprint extraction and spoofprint extraction operations into a single unified neural network architecture that processes the audio input simultaneously through both pathways. This combined approach allows the system to extract both speaker identity features and spoofing detection features in one pass, improving adaptability to various spoofing techniques while minimizing additional processing time compared to sequential processing.

Inventive Principle:
Principle #5Merging (Combining)

3Measurement precision

If the neural network extracts features for both voiceprints and spoofprints, then the system can distinguish genuine from spoofed voice patterns, but the system complexity increases

Engineering Contradiction:
Improvediscrimination between genuine and spoofed voiceVSAvoidneural network layer structure
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system applies local quality by designing specialized embedding extraction layers with different feature extraction capabilities: the voiceprint embedding layer focuses on speaker-specific acoustic characteristics (pitch, timbre, formants), while the spoofprint embedding layer focuses on spoofing-specific artifacts (spectral inconsistencies, temporal patterns, synthesis artifacts). This localized specialization enables precise discrimination between genuine and spoofed voice patterns without requiring the entire network to be uniformly complex.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4097717B1Robust spoofing detection system using deep residual neural networks
Publication Date: 2026.03.18 PINDROP SECURITY INC
  • EP4097717B1 patent drawingFigure 1
  • EP4097717B1 patent drawingFigure 2
  • EP4097717B1 patent drawingFigure 3

AI summary

Embodiments described herein provide for systems and methods for implementing a neural network architecture for spoof detection in audio signals. The neural network architecture contains a layers defining embedding extractors that extract embeddings from input audio signals. Spoofprint embeddings are generated for particular system enrollees to detect attempts to spoof the enrollee's voice. Optionally, voiceprint embeddings are generated for the system enrollees to recognize the enrollee's voice. The voiceprints are extracted using features related to the enrollee's voice. The spoofprints are extracted using features related to features of how the enrollee speaks and other artifacts. The spoofprints facilitate detection of efforts to fool voice biometrics using synthesized speech (e.g., deepfakes) that spoof and emulate the enrollee's voice.