Spoofing Detection Using CQCC Features and Deep Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional voice-based security systems are vulnerable to spoofing attacks such as voice conversion, speech synthesis, and replay attacks, which are difficult to distinguish from genuine access, especially due to high-quality recording and playback devices making replay attacks nearly indistinguishable.

Innovation Solution

The implementation of a deep-learning framework that combines convolutional neural networks (CNNs) and recurrent neural networks (RNNs) for feature extraction and classification, utilizing Constant-Q Cepstral Coefficients (CQCCs) and other acoustic features to effectively differentiate between genuine and spoofed speech, along with score-level fusion for improved robustness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional CQCC features are used for spoofing detection, then detection capability is improved, but memory usage and processing time increase significantly

Engineering Contradiction:
Improvespoofing detection capabilityVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent extracts only the essential acoustic features needed for spoofing detection rather than computing the complete CQCC feature set. By selecting and extracting specific discriminative features from the audio signal, the system achieves effective spoofing detection while avoiding the high computational cost of full CQCC computation.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the feature extraction process into multiple stages: first extracting basic acoustic features, then selectively processing only those features that are most discriminative for spoofing detection. This segmented approach reduces the overall processing burden while maintaining detection effectiveness.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If conventional CQCC features are used for spoofing detection, then detection capability is improved, but memory usage increases significantly

Engineering Contradiction:
Improvespoofing detection capabilityVSAvoidmemory usage
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential acoustic features needed for spoofing detection rather than computing the complete CQCC feature set. By selecting and extracting specific discriminative features from the audio signal, the system achieves effective spoofing detection while avoiding the high computational cost of full CQCC computation.

Inventive Principle:
Principle #2Taking out (Extraction)

3Reliability

If conventional spoofing detection systems are used, then basic detection is possible, but they cannot distinguish high-quality replay attacks from genuine speech

Engineering Contradiction:
Improvedetection accuracyVSAvoidvulnerability to replay attacks
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent converts the harmful effect of replay attacks into a beneficial detection opportunity by analyzing artifacts introduced during playback. Instead of treating replayed speech as indistinguishable from genuine speech, the system identifies characteristic patterns in the acoustic features that reveal the speech has been played back, thereby transforming the spoofing attempt into a detectable anomaly.

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

Solution Approach 2:

The patent introduces acoustic feature analysis as an intermediary layer between the audio input and the spoofing detection decision. By computing and analyzing specific acoustic features that are sensitive to playback artifacts, the system creates an intermediate representation that reveals the true nature of the input signal, enabling reliable distinction between genuine and replayed speech.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Measurement precision

If deep-learning frameworks with CNN and RNN are implemented, then spoofing detection performance is improved, but system complexity increases

Engineering Contradiction:
Improvespoofing detection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the deep learning system into two specialized components: a CNN module for extracting spatial/acoustic features and an RNN module for capturing temporal dependencies. This segmentation allows each component to be optimized for its specific function, improving overall detection accuracy while making the complex system more manageable and interpretable.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent merges the strengths of CNN and RNN architectures into a unified spoofing detection system. The CNN component processes acoustic features to identify patterns, while the RNN component analyzes temporal sequences to detect spoofing characteristics. By combining these complementary approaches, the system achieves superior detection performance that leverages both spatial and temporal information.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3590113B1Method and apparatus for detecting spoofing conditions
Publication Date: 2024.05.29 PINDROP SECURITY INC
  • EP3590113B1 patent drawingFigure 1~2
  • EP3590113B1 patent drawingFigure 3A
  • EP3590113B1 patent drawingFigure 3B

AI summary

An automated speaker verification (ASV) system incorporates a first deep neural network to extract deep acoustic features, such as deep CQCC features, from a received voice sample. The deep acoustic features are processed by a second deep neural network that classifies the deep acoustic features according to a determined likelihood of including a spoofing condition. A binary classifier then classifies the voice sample as being genuine or spoofed.