Hybrid Speaker Verification Triage for Multilingual Accuracy and Speed
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speaker verification systems face challenges in handling multiple languages and dialects, requiring separate models for each, which is computationally intensive and time-consuming, especially for low-resource languages, and incur significant human and computational effort for maintenance and updates.
Innovation Solution
A hybrid multilingual text-dependent and text-independent speaker verification system that combines lightweight text-dependent models on user devices with more computationally intensive text-independent models on remote servers, using a triage stage to invoke the text-independent model only when text-dependent confidence scores fail to satisfy a threshold, optimizing performance and reducing computational burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If text-independent speaker verification is used, then convenience is improved, but accuracy deteriorates due to background noise and channel effects
Solution Approach 1:
The verification process is segmented into two distinct stages: text-dependent verification (high accuracy, requires reading text) and text-independent verification (convenient, no text required). The system divides the overall authentication flow into these separate functional segments that can be executed independently or in sequence, allowing each segment to optimize for its specific strength.
Solution Approach 2:
The patent merges text-dependent and text-independent speaker verification systems into a unified hybrid framework. The evidence from both verification methods is combined through evidence combination rules to produce a final authentication decision, leveraging the accuracy of text-dependent verification and the convenience of text-independent verification.
2Measurement precision
If text-dependent speaker verification is used, then accuracy is improved, but productivity deteriorates due to requiring users to read text
Solution Approach 1:
The system dynamically adapts the verification process based on conditions. It can switch between text-dependent and text-independent modes, or combine both, depending on security requirements and user needs. The verification protocol is flexible and can be adjusted in real-time, making the process neither strictly sequential nor fixed, thereby optimizing both accuracy and speed based on contextual factors.
3Reliability
If hybrid verification combining both methods is used, then reliability is improved, but device complexity increases
Solution Approach 1:
The hybrid verification system is designed with universal components that handle both text-dependent and text-independent verification through common processing pipelines. Shared modules include feature extraction, evidence combination logic, and decision-making frameworks, allowing the system to perform multiple verification functions without requiring entirely separate processing paths for each method.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A speaker verification method (400) includes receiving audio data (120) corresponding to an utterance (119), processing a. first portion (121) of the audio data, that characterizes a predetermined hotword to generate a text-dependent evaluation vector (214), and generating one or more text-dependent confidence scores (215). When one of the text-dependent confidence scores satisfies a threshold, the operations include identifying a speaker of the utterance as a respective enrolled user associated with the text-dependent confidence score that satisfies the threshold and initiating performance of an action without performing speaker verification. When none of the text-dependent confidence scores satisfy the threshold, the operations include processing a second portion (122) of the audio data that characterizes a query to generate a text-independent evaluation vector (224), generating one or more text-independent confidence scores (225), and determining whether the identity of the speaker of the utterance includes any of the enrolled users.