Voice Authentication Segmentation Against Replay and Synthesis Attacks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing voice biometric authentication systems are vulnerable to replay and synthesis attacks, where imposters use recorded or synthesized speech to deceive the system, leading to potential unauthorized access.

Innovation Solution

The system performs independent voice biometric authentication on specific portions of the audio signal, separating static and dynamic components, and incorporates replay and synthesis detection to ensure the speech is live and genuine, using a combination of voice biometric analysis, replay detection, and synthesis detection on each segment.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If voice biometric authentication is performed on the entire audio signal, then authentication speed is improved, but vulnerability to replay and synthesis attacks increases

Engineering Contradiction:
Improveauthentication speedVSAvoidauthentication security
Core Design Contradiction:
SpeedVSReliability

Solution Approach 1:

The audio signal is divided into multiple segments, with specific portions (such as challenge words or specific time intervals) selected for voice biometric authentication. This segmentation allows the system to authenticate users faster by focusing on key segments while maintaining security through strategic selection of authentication portions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Different portions of the audio signal are treated differently - some segments are used for replay detection and synthesis detection, while other segments are used for voice biometric authentication. This local differentiation allows the system to optimize both speed and security by applying appropriate verification methods to specific segments.

Inventive Principle:
Principle #3Local quality

2Reliability

If independent voice biometric authentication is performed on multiple segments, then authentication robustness is improved, but system complexity increases

Engineering Contradiction:
Improveauthentication robustnessVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The authentication process is segmented into distinct stages: replay detection, synthesis detection, and voice biometric authentication. Each segment handles a specific aspect of verification, making the overall complex system manageable through modular design and clear separation of functions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs multiple authentication functions (replay detection, synthesis detection, voice biometric verification) within a unified framework. By integrating these functions into a single authentication flow, the system manages complexity while maintaining robust multi-layered verification.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Reliability

If replay and synthesis detection are incorporated, then authentication security is improved, but processing time increases

Engineering Contradiction:
Improveauthentication securityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

Replay detection and synthesis detection are performed on specific segments of the audio signal before the main voice biometric authentication. By conducting these checks preliminarily on selected portions, the system prevents fraudulent authentication attempts early in the process without requiring extensive processing of the entire audio signal.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system applies replay and synthesis detection to only the necessary portions of the audio signal (such as challenge words or specific time intervals) rather than the entire audio stream. This partial application of detection methods reduces processing time while maintaining adequate security coverage.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4170526B1An authentication system and method
Publication Date: 2026.02.11 VALIDSOFT LTD
  • EP4170526B1 patent drawingFigure 1(a)
  • EP4170526B1 patent drawingFigure 1(b)
  • EP4170526B1 patent drawingFigure 2(a)

AI summary

A computer implemented method, comprising: receiving a first audio signal; identifying one or more portions of the first audio signal as corresponding to one or more pre-determined text sequences; identifying one or more portions of the first audio signal as corresponding to one or more new text sequences; performing a voice authentication on a first portion of the first audio signal identified as corresponding to a first pre-determined text sequence and performing a separate voice authentication on a second portion of the first audio signal identified as corresponding to a new text sequence.