Speaker Recognition Using Low-Dimensional P-Vector Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice biometric verification systems, particularly those using the i-vector approach, face challenges with high computational requirements and poor performance on short utterances, as they require complex multi-iterative procedures and have low dimensional dimensions, limiting their accuracy and efficiency.
Innovation Solution
The proposed solution combines the strengths of i-vector and GMM-MAP methods by simulating voice using GMM-MAP, transforming the GMM model into a low-dimensional 'p-vector' through dimension reduction, and employing a channel-compensation module based on discriminant analysis, which improves speaker modeling and recognition accuracy, and reduces hardware costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If i-vector approach is used for speaker verification, then computational power requirements increase due to complex multi-iterative factor analysis procedures, but recognition accuracy on short utterances deteriorates due to low dimensional dimensions
Solution Approach 1:
The patent transforms the high-dimensional GMM speaker model into a low-dimensional p-vector by projecting onto a subspace defined by discriminant eigenvectors. This parameter transformation reduces computational complexity while maintaining recognition accuracy by retaining only the most discriminative dimensions for speaker identification.
Solution Approach 2:
The patent extracts the essential speaker information from the high-dimensional GMM model by identifying and retaining only the most discriminative features through discriminant analysis. The p-vector captures the critical speaker characteristics while discarding redundant information, thereby reducing computational requirements.
2Reliability
If i-vector approach is used for speaker verification, then computational power requirements increase, but performance on short utterances deteriorates
Solution Approach 1:
The patent changes the dimensional parameter of the speaker model from high-dimensional i-vectors to low-dimensional p-vectors. This parameter change reduces the computational power required for verification while improving reliability on short utterances by focusing on the most discriminative features.
Solution Approach 2:
The patent performs dimensionality reduction by transforming the speaker model from high-dimensional space to low-dimensional space through projection onto discriminant eigenvectors. This dimensional change maintains verification performance while significantly reducing computational power consumption.
3Measurement precision
If GMM-MAP approach is used for speaker simulation, then recognition accuracy improves, but hardware costs increase due to complex modeling requirements
Solution Approach 1:
The patent extracts the essential speaker information from the complex GMM-MAP model into a compact p-vector representation. This extraction maintains recognition accuracy while reducing hardware complexity by storing and processing only the essential low-dimensional features rather than the full complex model.
Solution Approach 2:
The patent creates a simplified copy of the speaker model in the form of a p-vector that captures the essential characteristics without requiring the full GMM-MAP complexity. This copy enables accurate recognition while significantly reducing hardware requirements for storage and processing.
Data Source
AI summary
A computer-implemented method for verifying identity of a speaker is proposed. A low dimensional p-vector based on a speech of the speaker is extracted from the generated high dimensional speaker model and is then compared with the stored specific speaker's p-vector obtained previously during the enrollment process. The resulting biometric score is then used to determine whether to verify the speaker, or not.