Wake-Word Speaker Verification Using Clustering Without Enrollment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Voice assistants like BIXBY, SIRI, and ALEXA face issues with user enrollment being tedious, leading to degraded voice wake-up performance and increased invalid wake-ups when enrollment is not required.
Innovation Solution
A system and method for self-wake word speaker verification that automatically identifies speaker vectors from incoming audio, performs embedding vector updates, and uses clustering to determine speaker verification with varying similarity thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional voice recognition systems are used, then voice input is convenient, but security against unauthorized access is insufficient
Solution Approach 1:
The patent introduces a voiceprint verification module as an intermediary between voice recognition and task execution. This module extracts voiceprint features from the recognized voice and compares them against stored voiceprint templates, acting as a security mediator that maintains the convenience of voice input while adding authentication reliability
Solution Approach 2:
The system implements feedback by comparing real-time voiceprint features with pre-stored voiceprint templates. The verification result feeds back into the system to either authorize or reject the voice command, creating a closed-loop security mechanism that ensures only authenticated users can access protected functions
2Reliability
If voiceprint verification is added to enhance security, then authentication reliability improves, but system complexity increases
Solution Approach 1:
The patent merges the voiceprint verification functionality into the existing voice recognition system. The voiceprint extraction and verification modules are integrated with the voice recognition pipeline, sharing common components such as feature extraction mechanisms and processing workflows, thereby reducing overall system complexity while enhancing authentication capabilities
Solution Approach 2:
The voiceprint verification module is designed with multi-functionality, serving both as an authentication mechanism and as a security verification layer. It can be applied across different voice-based applications and scenarios, making the system more versatile without proportionally increasing complexity
3Reliability
If multiple verification steps are implemented, then security reliability improves, but processing time increases
Solution Approach 1:
The system performs preliminary action by pre-extracting and storing voiceprint templates during user registration. During verification, the system only needs to compare the current voiceprint against the pre-prepared template, significantly reducing processing time compared to performing full voiceprint analysis in real-time
Solution Approach 2:
The verification process is segmented into distinct stages: voiceprint extraction, feature comparison, and authentication decision. This segmentation allows each stage to be optimized independently and enables parallel processing where possible, reducing overall verification time while maintaining security reliability
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method includes obtaining audio data and identifying an utterance of a wake word or wake phrase in the audio data. The method also includes generating an embedding vector based on the utterance from the audio data and accessing a set of previously-generated vectors representing previous utterances of the wake word or wake phrase. The method further includes performing clustering on the embedding vector and the set of previously-generated vectors to identify a cluster including the embedding vector, where the identified cluster is associated with a speaker. The method also includes updating a speaker vector associated with the speaker based on the embedding vector and determining, using a speaker verification model, a similarity score between the updated speaker vector and the embedding vector. In addition, the method includes determining, based on the similarity score, whether a speaker providing the utterance matches the speaker associated with the identified cluster.