Autoencoder Anti-Spoofing for Real-Time Edge Audio
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anti-spoofing methods are inefficient and unsuitable for real-time or edge device applications, as they require multiple passes through detection algorithms and manual setting of window sizes to identify partially spoofed signals.
Innovation Solution
A method using a trained autoencoder with a spoof encoder and vector quantised codebook to output an n-dimensional array parameterising the authenticity of sub-periods of an audio or video signal, allowing for immediate identification of real and fake parts of the signal with a single inference pass.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the signal is cut into multiple slices for individual testing to detect partially spoofed signals, then the authenticity of specific parts can be identified, but the computational efficiency deteriorates and multiple passes through the detection algorithm are required
Solution Approach 1:
The patent segments the signal into multiple sub-periods (e.g., 100 sub-periods of 10ms each for a 1-second signal) and processes each segment independently through the autoencoder. This segmentation enables fine-grained authenticity detection while maintaining computational efficiency by processing segments in parallel during a single forward pass, eliminating the need for multiple sequential passes through the detection algorithm.
Solution Approach 2:
The patent transforms the signal from a single temporal dimension into a multi-dimensional representation by dividing it into sub-periods and encoding each in a latent space. The autoencoder maps each sub-period to a latent representation that can be independently analyzed, enabling simultaneous detection of authenticity across all segments in one computational pass.
2Measurement precision
If manual window size setting is used to identify fake sections, then detection accuracy can be adjusted, but the ease of operation deteriorates and real-time processing becomes difficult
Solution Approach 1:
The patent implements self-service by automatically dividing the signal into sub-periods and processing them through the autoencoder without requiring manual intervention. The system autonomously determines the temporal resolution and processing parameters, eliminating the need for operators to manually set window sizes while maintaining detection accuracy and enabling real-time processing.
3Measurement precision
If multiple passes through the detection algorithm are used to test each window slice, then the authenticity of different sections can be determined, but the time required for processing increases
Solution Approach 1:
The patent merges the authentication of multiple signal segments into a single computational pass by processing all sub-periods simultaneously through the autoencoder. This consolidation eliminates the need for sequential multiple passes, reducing processing time while maintaining comprehensive authenticity detection across the entire signal.
Solution Approach 2:
The patent ensures continuous processing by performing the autoencoding operation in a single forward pass that simultaneously analyzes all sub-periods. This continuous action eliminates idle time between sequential passes and maintains steady computational throughput, significantly reducing total processing time compared to iterative approaches.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
A method of anti-spoofing for identifying the authenticity of sub-periods of an audio or video signal, the method comprising: receiving an audio or video signal having a time period comprising a plurality of sub-periods; applying a trained encoder, the trained encoder configured to output an n-dimensional array of continuous values parameterising the authenticity of each sub-period, wherein the encoder is trained as part of an autoencoder, the autoencoder having at least one encoder, at least one vector quantised codebook, and at least one decoder; and outputting an indication of the authenticity of at least one sub-period based on the n-dimensional array and the at least one vector quantised codebook; wherein the autoencoder is trained using training data comprising a set of training audio or video signals, sub-periods of the training audio or video signals being associated with respective labels of whether said sub-periods are real or fake.