Continuous Voice Biometrics for Passive Multi-Speaker Enrollment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice biometrics systems face challenges in efficiently enrolling new speakers and maintaining accurate speaker profiles, particularly in dynamic environments with multiple speakers, leading to issues like staleness and false authentication.
Innovation Solution
A passive and continuous voice biometrics system that employs machine-learning models to extract embeddings from audio signals, enabling flexible enrollment and profile management without explicit user prompts, using adaptive thresholding and unsupervised clustering to maintain up-to-date speaker profiles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If active enrollment is used, then speaker verification accuracy is improved, but user experience deteriorates due to time-consuming enrollment process
Solution Approach 1:
The system performs preliminary enrollment actions automatically in the background without user prompts. When a user first interacts with the device, the system captures audio samples and creates speaker profiles automatically, so that when the user needs to use voice verification later, the enrollment is already complete and accurate.
Solution Approach 2:
The system serves itself by automatically performing enrollment tasks without requiring user participation. The voice biometrics system captures audio, extracts features, and creates/update speaker profiles autonomously based on audio samples collected during device usage, eliminating the need for users to manually complete enrollment forms or repeat phrases.
2Device complexity
If static enrollment is used, then system simplicity is maintained, but speaker profile accuracy deteriorates over time due to staleness
Solution Approach 1:
The system transitions from static to dynamic enrollment by continuously updating speaker profiles in the background. The voice biometrics system monitors audio samples over time and automatically updates speaker profiles to reflect current voice characteristics, adapting to voice changes without requiring system complexity changes or user intervention.
Solution Approach 2:
The system uses feedback from continuous audio sampling to update speaker profiles. By monitoring voice samples over time and comparing them against existing profiles, the system identifies when updates are needed and automatically incorporates new audio data to maintain accurate, up-to-date speaker representations.
3Adaptability or versatility
If multiple speakers are handled simultaneously, then system versatility is improved, but speaker differentiation difficulty increases
Solution Approach 1:
The system segments the audio signal into distinct speaker portions using speaker diarization technology. By dividing the mixed audio into separate segments associated with different speakers, the system can individually process and identify each speaker's characteristics, making differentiation manageable even when multiple speakers are present simultaneously.
Solution Approach 2:
The system applies local quality analysis by examining specific acoustic characteristics of each speaker segment. Instead of treating all speakers uniformly, the system identifies and analyzes distinctive local features of each speaker's voice pattern, enabling accurate differentiation based on unique local acoustic signatures.
Data Source
AI summary
Embodiments described herein provide for a voice biometrics system execute machine-learning architectures capable of passive, active, continuous, or static operations, or a combination thereof. Systems passively and/or continuously, in some cases in addition to actively and/or statically, enrolling speakers as the speakers speak into or around an edge device (e.g., car, television, radio, phone). The system identifies users on the fly without requiring a new speaker to mirror prompted utterances for reconfiguring operations. The system manages speaker profiles as speakers provide utterances to the system. Machine-learning architectures implement a passive and continuous voice biometrics system, possibly without knowledge of speaker identities. The system creates identities in an unsupervised manner, sometimes passively enrolling and recognizing known or unknown speakers. The system offers personalization and security across a wide range of applications, including media content for over-the-top services and IoT devices (e.g., personal assistants, vehicles), and call centers.