Voice Liveness Detection for Audio Deepfake Call Authentication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic speaker verification systems are vulnerable to deepfake attacks, which use machine-generated speech to bypass voice biometrics and gain unauthorized access to call center systems, necessitating improved means for detecting fraudulent audio-based deepfake technology and ensuring voice liveness detection.
Innovation Solution
Systems and methods for deepfake detection using speech-phrase repetition detection, background change detection, and passive voice-liveness detection, combined with an automated model update module to adapt to new presentation attacks, and employing machine-learning architectures for temporal feature analysis and Natural Language Processing (NLP) to classify speech as human or machine-generated.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional automatic speaker verification systems are used, then authentication speed is improved, but vulnerability to deepfake attacks increases
Solution Approach 1:
The patent introduces an intermediary deepfake detection system that sits between the traditional ASV system and the authentication decision. This intermediary analyzes audio signals for deepfake characteristics (such as spectral artifacts, temporal inconsistencies, and unnatural speech patterns) before the final verification decision is made, thereby maintaining fast authentication while adding a security layer against deepfake attacks.
Solution Approach 2:
The authentication system is segmented into multiple independent components: traditional speaker verification module, deepfake detection module, and a decision-fusion module. Each component operates independently with its own analysis methods, allowing the system to process authentication requests quickly while simultaneously checking for deepfake indicators without compromising either speed or reliability.
2Measurement precision
If deepfake detection analysis is added to authenticate calls, then detection accuracy is improved, but processing time increases
Solution Approach 1:
The system applies partial analysis by focusing deepfake detection efforts only on suspicious or high-risk calls rather than analyzing every call in detail. For low-risk calls, a lightweight verification is performed, while for high-risk calls, full deepfake analysis is applied. This selective approach maintains high detection accuracy for problematic calls while minimizing processing time for normal calls.
Solution Approach 2:
The system performs preliminary lightweight screening of all calls using fast feature extraction methods, and only proceeds to more time-consuming deepfake analysis when the preliminary screen indicates potential suspicious patterns. This preliminary action filters out the majority of normal calls quickly, reserving detailed analysis only for cases where it is truly needed.
3Reliability
If multiple detection methods are combined for voice liveness detection, then detection reliability is improved, but system complexity increases
Solution Approach 1:
The patent merges multiple detection methods (spectral analysis, temporal pattern recognition, background noise analysis, and speech feature extraction) into a unified deepfake detection system. These methods are integrated at the feature extraction and decision-fusion levels, sharing common computational infrastructure and data processing pipelines, which reduces overall system complexity while maintaining the benefits of multiple detection approaches.
Data Source
AI summary
Disclosed are systems and methods including software processes executed by a server that detect audio-based synthetic speech (“deepfakes”) in a call conversation. The server applies an NLP engine to transcribe call audio and analyze the text for anomalous patterns to detect synthetic speech. Additionally or alternatively, the server executes a voice “liveness” detection system for detecting machine speech, such as synthetic speech or replayed speech. The system performs phrase repetition detection, background change detection, and passive voice liveness detection in call audio signals to detect liveness of a speech utterance. An automated model update module allows the liveness detection model to adapt to new types of presentation attacks, based on the human provided feedback.