Speaker Recognition Audio Segmentation for Unauthorized Access Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speaker recognition systems are vulnerable to unauthorized access attacks where a third party speaks after the enrolled user, as the system may not detect a speaker change quickly enough, allowing unauthorized speech to be processed as if spoken by the enrolled user.
Innovation Solution
A method involving biometric processes on audio signals to detect speaker changes by analyzing successive sections of the audio signal, determining if the speaker recognition scores exceed a threshold, and using a combination of text-dependent and text-independent biometric processes to verify the speaker's identity, ensuring only the enrolled user's speech is processed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If speaker recognition is performed on the entire audio signal at once, then the system can verify the enrolled user's identity, but it cannot detect unauthorized speech that occurs after the enrolled user speaks
Solution Approach 1:
The audio signal is divided into multiple sections, with the first section processed for initial speaker verification and subsequent sections processed for speaker change detection. This segmentation allows the system to verify the enrolled user's identity while simultaneously monitoring for unauthorized speech occurrences throughout the audio signal.
2Measurement precision
If the speaker recognition process waits for the complete speech to be processed, then it can make an accurate determination, but it responds too slowly to prevent unauthorized access
Solution Approach 1:
The system performs preliminary speaker verification on the first section of the audio signal to establish the enrolled user's identity before processing the complete speech. This preliminary action enables the system to make an initial determination quickly while still allowing for subsequent verification of the entire speech if needed.
Solution Approach 2:
The system continuously monitors subsequent sections of the audio signal for speaker changes and provides feedback to the speaker recognition process. If a speaker change is detected, the system can immediately terminate processing and declare unauthorized access, eliminating the need to wait for complete speech processing.
3Speed
If the system processes short speech segments quickly, then it can detect speaker changes rapidly, but it may not have enough data to accurately verify the enrolled user's identity
Solution Approach 1:
The audio signal is divided into multiple sections, with the first section containing sufficient duration for accurate biometric verification of the enrolled user's identity. Subsequent sections are then processed for speaker change detection, ensuring that both verification accuracy and change detection speed are optimized.
Data Source
AI summary
A method of speaker recognition comprises: receiving an audio signal comprising speech; performing a biometric process on a first part of the audio signal, wherein the first part of the audio signal extends over a first time period; obtaining a speaker recognition score from the biometric process for the first part of the audio signal; performing a biometric process on a plurality of second parts of the audio signal, wherein the second parts of the audio signal are successive sections of the first part of the audio signal, and wherein each second part of the audio signal extends over a second time period and the second time period is shorter than the first time period; obtaining a respective speaker recognition score from the biometric process for each second part of the audio signal; and determining whether there has been a speaker change based on the respective speaker recognition scores for successive second parts of the audio signal.