Speaker Verification Model Evaluation Using Privacy-Preserving Event Logs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for evaluating speaker recognition performance are cumbersome, costly, and raise privacy and security concerns due to the need for manual input and collection of user data, making it difficult to assess the performance of verification systems deployed on computing devices.

Innovation Solution

A method and system that utilize verification data generated during the verification process, such as event logs, to evaluate the performance of speaker recognition systems without sharing personal or biometric data, allowing for automated assessment of verification model performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional manual evaluation methods are used to assess speaker recognition performance, then evaluation accuracy can be maintained, but the process becomes cumbersome, costly, and raises privacy and security concerns

Engineering Contradiction:
Improveevaluation efficiencyVSAvoidevaluation process complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The verification system automatically evaluates its own performance using verification data generated during normal operation. The system self-monitors performance metrics such as false reject rate and false accept rate without requiring external manual evaluation, thereby improving efficiency while maintaining accuracy through automated assessment of verification results

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system implements continuous feedback loops where verification results are automatically collected, analyzed, and used to assess model performance. Performance metrics are fed back to automatically trigger model replacements when thresholds are exceeded, creating a self-improving system that eliminates manual intervention while maintaining high evaluation accuracy

Inventive Principle:
Principle #23Feedback

2Measurement precision

If user data is collected for evaluation purposes, then performance assessment can be conducted, but privacy and security concerns arise

Engineering Contradiction:
Improveperformance metric accuracyVSAvoidprivacy and security risks
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The system extracts only the necessary verification outcomes and performance metrics from the verification process, separating these essential evaluation data points from personal and biometric information. By taking out only what is needed for performance assessment while leaving sensitive user data behind, the system achieves accurate measurement without compromising privacy or security

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system introduces an intermediary evaluation layer that processes verification results without exposing personal data. This intermediary mechanism enables performance assessment by working with anonymized verification outcomes rather than raw user data, thereby maintaining measurement precision while eliminating privacy and security risks

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4233050B1Assessing speaker recognition performance
Publication Date: 2025.12.03 GOOGLE LLC
  • EP4233050B1 patent drawingFigure 1A
  • EP4233050B1 patent drawingFigure 1B
  • EP4233050B1 patent drawingFigure 1C

AI summary

A method (300) for evaluating a verification model (146) includes receiving first and second sets of verification results (148) where each verification result indicates whether a primary model or an alternative model verifies an identity of a user (10) as a registered user. The method further includes identifying each verification result in the first and second sets that includes a performance metric (212), determining a first score (222) of the primary model based on a number of the verification results identified in the first set that includes the performance metric, and determining a second score of the alternative model based on a number of the verification results identified in the second set that includes the performance metric. The method further includes determining whether a verification capability of the alternative model is better than a verification capability of the primary model based on the first score and the second score.