Voice Biomarker Authentication for Synthetic Speech Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for distinguishing between natural and synthetic voices are inadequate, particularly in critical communication contexts, leading to security and integrity risks.
Innovation Solution
A voice authentication method using biomarkers such as pitch, intensity, formants, Mel Cepstral Distance, and Sound of Silence to differentiate between natural and synthetic voices, employing statistical analysis and normalization to calculate an authentication index.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If signal theory-based analysis methods are used to discriminate between natural and synthesised voices, then the discrimination capability is provided, but the reliability and accuracy are insufficient for critical security contexts
Solution Approach 1:
The voice authentication method segments the voice signal analysis into multiple independent biomarker dimensions (pitch, intensity, formants, Mel Cepstral Distance, Sound of Silence). Each biomarker is extracted and analyzed separately, then combined to form a comprehensive authentication index. This segmentation allows for more precise measurement in each dimension while maintaining overall reliability through the multi-faceted approach.
Solution Approach 2:
The invention transforms the voice authentication problem from traditional signal theory parameters to specific biomarker parameters (pitch, intensity, formants, Mel Cepstral Distance, Sound of Silence). By changing the parameter space and using statistical analysis on these specific biomarkers, the method achieves both high measurement precision for each parameter and high reliability through the combination of multiple parameters.
2Adaptability or versatility
If traditional voice analysis methods are used, then the process is simple, but the ability to detect synthetic voices in critical communication networks is insufficient
Solution Approach 1:
The invention adds new dimensions to voice analysis by introducing five specific biomarkers (pitch, intensity, formants, Mel Cepstral Distance, Sound of Silence) that go beyond traditional signal theory. This dimensional expansion enables the system to detect synthetic voices that traditional methods miss, while the modular structure of extracting and analyzing each biomarker separately keeps the implementation manageable.
Solution Approach 2:
The authentication system is designed with multi-functionality by incorporating multiple biomarkers that can detect various types of voice manipulation and synthesis. The same framework can authenticate different types of voices and detect different synthesis methods, making the system universally applicable to critical communication security without requiring separate specialized systems.
3Reliability
If voice authentication is implemented to ensure security, then communication integrity is protected, but false authentication risks may occur
Solution Approach 1:
The authentication method uses feedback through the calculation of an authentication index that combines multiple biomarkers. The statistical analysis of each biomarker provides feedback on the likelihood of authenticity, and the combined index provides a comprehensive verification. This multi-layered feedback mechanism reduces false authentication by cross-validating across multiple independent biomarker dimensions.
Solution Approach 2:
The authentication system uses a composite approach by combining multiple different biomarkers (pitch, intensity, formants, Mel Cepstral Distance, Sound of Silence) into a unified authentication index. This composite methodology leverages the strengths of each individual biomarker while compensating for their individual weaknesses, thereby reducing false authentication risks through diversified verification.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Method (100) for authenticating voice files comprising: defining (101) a plurality of voice biomarkers (BM1 -BMM ) chosen from among the following types: pitch, intensity, formant, Mel Cepstral Distance, Sound of Silence; providing (102) a voice file to be authenticated (FAUX ); extracting (103) at least two voice biomarkers (BM1 -BMM) from the voice file to be authenticated (FAUX). In addition, the method comprises calculating (104) from said at least two voice biometrics (BM1 -BMM) an authentication index (INDX) such as to take values within an authentication interval (INT) having one end indicative of a situation in which the voice file to be authenticated (FAUNRV ) has a non-natural origin and another extreme indicative of another situation in which the voice file to be authenticated (FAUNV ) has a natural origin.