Identity Authentication Using Scenario-Mapped Voiceprint Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional voiceprint recognition technologies face challenges in maintaining recognition accuracy due to scenario mismatches, such as variations in speaking status, manner, noise interference, and device differences, leading to a mismatch between registration and recognition scenarios, which are not fully addressed by existing methods.
Innovation Solution
An identity authentication method that utilizes separate probability distribution models and parameters for registration and recognition scenarios, with a mapping relationship to correct likelihood scores, compensating for scenario mismatches by mapping features between these scenarios using a trained linear Gaussian distribution and machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional voiceprint recognition methods are used, then the system is simple to implement, but recognition accuracy deteriorates due to scenario mismatches between registration and recognition phases
Solution Approach 1:
The patent segments the voiceprint recognition system into distinct components: a registration module that creates voiceprint models during registration, and a recognition module that performs authentication during recognition. Crucially, it introduces separate probability distribution models for each scenario (registration scenario model and recognition scenario model) to handle the mismatches that occur between these two phases, thereby improving recognition accuracy without overwhelming system complexity.
Solution Approach 2:
The patent changes the parameters by introducing scenario-specific probability distribution models that adapt to different conditions. Instead of using a single uniform model, it employs separate models that account for variations in speaking status, noise levels, and device characteristics between registration and recognition scenarios, thereby compensating for scenario mismatches and improving measurement precision.
2Measurement precision
If separate probability distribution models are introduced for registration and recognition scenarios, then recognition accuracy is improved, but device complexity increases
Solution Approach 1:
The patent introduces probability distribution models as intermediary components that mediate between the raw voice features and the authentication decision. These models act as translators that adapt features from different scenarios (registration and recognition) into a common framework, allowing accurate comparison while managing complexity through structured intermediate representations rather than direct complex comparisons.
Solution Approach 2:
The patent creates universal probability distribution models that can handle multiple scenarios. The registration scenario model and recognition scenario model are designed to be multi-functional, accommodating various conditions such as different speaking statuses, noise environments, and device types within a unified framework, thereby improving accuracy without proportionally increasing complexity.
3Ease of operation
If voiceprint features are extracted from small quantity of reserved voice messages, then registration process is quick and convenient, but robustness of voiceprint feature decreases
Solution Approach 1:
The patent performs preliminary action by establishing scenario-specific probability distribution models during the registration phase. These models are trained in advance on the available voice messages (even if limited in quantity) to capture the characteristics of the registration scenario. This preliminary modeling compensates for the limited data by encoding scenario-specific patterns that will be useful during recognition, thereby improving robustness without requiring extensive registration data.
Solution Approach 2:
The patent implements feedback mechanisms where the probability distribution models learn from the registration data and provide feedback about scenario characteristics. The models adjust their parameters based on the available voice messages, capturing robust features even from small datasets. This feedback loop allows the system to maximize the information extracted from limited registration data, improving reliability while maintaining registration convenience.
Data Source
AI summary
An identity authentication method is provided, which relates to the field of artificial intelligence. The method includes: obtaining a biological feature extracted from a first registration object in a registration scenario; obtaining a biological feature extracted from a first recognition object in a recognition scenario; calculating a likelihood score based on the biological feature of the first registration object and the biological feature of the first recognition object; and determining, based on the likelihood score, whether an identity of the first recognition object is consistent with an identity of the first registration object. The likelihood score includes a posterior probability part and a marginal probability part.