Voice Liveness Detection via Spectral Difference Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice replay detection algorithms are ineffective in distinguishing live and replayed voice biometric data, leading to false positives and resource-intensive computations, which compromises the accuracy and efficiency of biometric verification transactions.
Innovation Solution
A method that calculates the spectral property difference between captured and recorded voice biometric data using a computing device, inputting this difference into a machine learning algorithm to determine user liveness by generating an output score that satisfies a threshold, thereby differentiating between live and replayed voice data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If known voice replay detection algorithms are used, then liveness detection is attempted, but accuracy is poor with high false positive rates and insufficient replay detection
Solution Approach 1:
The patent transforms the voice signal from time-domain to frequency-domain representation using spectrograms, fundamentally changing the parameter space for analysis. This transformation reveals spectral characteristics that are invisible in the original signal, enabling more accurate differentiation between live and replayed voices through machine learning algorithms that operate on these transformed parameters.
Solution Approach 2:
The patent introduces an intermediary processing layer consisting of spectral analysis and machine learning models that act as mediators between the raw voice signal and the liveness determination. This intermediary layer extracts meaningful features and transforms them into a format that can be reliably classified, improving both accuracy and confidence in verification results.
2Reliability
If known voice replay detection algorithms are used, then liveness detection is performed, but substantial computer system memory and computing resources are consumed
Solution Approach 1:
The patent extracts only the essential spectral features from the voice signal using spectrogram analysis, rather than processing the entire raw signal with computationally intensive algorithms. This extraction of key features reduces the data volume and computational complexity while maintaining the ability to detect replays with high reliability.
Solution Approach 2:
The patent replaces traditional signal processing approaches with machine learning-based spectral analysis. This substitution enables more efficient computation by leveraging patterns learned from training data, reducing the computational resources and memory required for real-time liveness detection compared to conventional algorithms.
3Measurement precision
If known voice replay detection algorithms are used, then liveness detection is attempted, but the time required for determination increases
Solution Approach 1:
The patent performs preliminary spectral analysis and feature extraction before the actual liveness classification occurs. By pre-processing the voice signal into spectrogram representations and extracting key spectral features in advance, the system reduces the computational burden during the critical verification moment, thereby reducing overall verification time while maintaining high detection accuracy.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A method for determining user liveness is provided that includes calculating, by a computing device, a spectral property difference between voice biometric data captured from a user and user record voice biometric data. The user and the computing device constitute a user-computing device pair, and the voice biometric data is captured by the computing device during a verification transaction. Moreover, the method includes inputting the spectral property difference into a machine learning algorithm, calculating an output score with the machine learning algorithm, and determining the voice biometric data was captured from a live user when the output score satisfies a threshold score.