A speech detection method based on logarithmic graph Fourier transform feature extraction

Abstract

This invention relates to the field of speech verification technology, and more particularly to a speech detection method based on logarithmic graph Fourier transform feature extraction, comprising the following steps: constructing a translation operator for the speech graph, using an exponential function to describe the decay of dependencies between speech samples, generating the Laplacian matrix of the graph, and representing the speech signal as an undirected graph to capture intra-frame and inter-frame structural relationships; mapping the sample values ​​of the speech signal to graph node signals, transforming the speech signal from the time domain to the graph frequency domain, extracting frequency domain features, and forming an enhanced feature representation by synchronously merging intra-frame and inter-frame oscillation analysis and combining time domain features; generating a detection score to determine whether the speech signal is a playback attack or belongs to normal speech. This invention, by introducing logarithmic graph Fourier transform and graph signal processing methods, effectively solves the limitations of existing technologies in playback speech detection, significantly improving the comprehensiveness of feature extraction, discriminative ability, and performance of the detection system.

Description

Technical Field

[0001] This invention relates to the field of speech verification technology, specifically to a speech detection method based on logarithmic graph Fourier transform feature extraction. Background Technology

[0002] In speech signal processing, playback speech detection is a crucial task, aiming to distinguish between real and played-back speech. Playback speech is frequently used by attackers as a means to deceive voice authentication systems, posing a serious threat to their security. Traditional methods in this field largely rely on the spectral characteristics and statistical properties of speech signals to construct detection models. However, these methods often struggle to effectively capture the high-order structural information implicit in speech signals, especially in scenarios involving complex speech data.

[0003] In recent years, feature extraction methods based on spectral analysis, such as Mel-frequency cepstral coefficients and constant-Q cepstral coefficients, have been widely used in playback speech detection tasks. These methods extract low-order features that reflect the time-frequency information of speech by analyzing the spectral characteristics of the speech signal, which are then used to train classification models. However, these low-order features are insufficient to fully describe the complex dynamic structure of the speech signal, thus limiting the performance of the detection system.

[0004] To address this problem, researchers have begun to explore graph signal processing methods to model the structural information of speech signals. Graph signal processing theory offers a new perspective: by modeling speech signals as graph structures, the spatiotemporal dependencies of speech signals can be captured. Among these methods, those based on the Graph Fourier Transform (GFT) can effectively extract implicit features of speech signals through frequency domain analysis. In particular, the log-Graph Fourier Transform (log-GFT), as an improved method, can better capture local and global features in speech signals, exhibiting high discriminative power for playback speech detection.

[0005] Furthermore, detection performance is closely related to the robustness of the selected features and the capabilities of the classification model. In recent years, deep learning models, particularly those based on Convolutional Neural Networks (CNNs) and Graph Convolutional Networks (GCNs), have demonstrated superior capabilities in processing complex speech signals. GCNs, as deep learning models capable of processing graph-structured data, can extract high-order structural information from speech signals layer by layer by aggregating feature information from nodes and their neighbors. This makes it possible to build more robust playback speech detection systems.

[0006] While these techniques have achieved some success in playback speech detection, many challenges remain. For example, existing methods often ignore the dependencies between local frames of the speech signal, thus failing to fully capture the spatiotemporal dynamics of the signal. Furthermore, because the differences between playback speech and real speech can be subtle, the performance of existing systems remains limited when faced with complex environments and diverse speech data.

[0007] In summary, how to extract implicit features from speech signals by combining logarithmic graphical Fourier transform and capture high-order structural information using graph convolutional networks has become a key technical challenge for improving the performance of playback speech detection. This provides a clear direction for future research and is also of great significance for applications in the field of speech security. Summary of the Invention

[0008] The purpose of this invention is to overcome or at least partially solve the above problems by proposing a speech detection method based on logarithmic graph Fourier transform feature extraction.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a speech detection method based on logarithmic graph Fourier transform feature extraction, comprising the following steps:

[0010] S1: Define the graph adjacency matrix based on the dependencies between speech samples, construct the translation operator of the speech graph through forward and backward translation operations, use the exponential function to describe the decay of the dependencies between speech samples, generate the Laplacian matrix of the graph, and represent the speech signal as an undirected graph to capture the structural relationships within and between frames.

[0011] S2: Map the sample values ​​of the speech signal to graph node signals, apply the logarithmic graph Fourier transform using the graph Laplace matrix, transform the speech signal from the time domain to the graph frequency domain, extract frequency domain features, and simultaneously merge intra-frame and inter-frame oscillation analyses, combining time domain features to form an enhanced feature representation;

[0012] S3: Use a pre-trained classification model to process the enhanced feature vector, generate a detection score to determine whether the speech signal is subject to a replay attack, and calculate the performance metrics t-DCF and EER to evaluate the system performance.

[0013] In a preferred embodiment, in step S1, the forward and backward shift operations of the speech signal are utilized, and the speech graph shift operator is expressed as follows: .

[0014] In a preferred embodiment, in step S1, the exponential function describing the decay of the dependency relationship between speech samples is expressed as:

[0015]

[0016] in It is an experimental constant. It is the experimental threshold.

[0017] In a preferred embodiment, in step S1, the graph Laplacian matrix of the intra-frame speech... It can be represented as Graph Laplacian matrix of inter-frame speech It can be represented as - The diagonal matrix is ​​specified as and The undirected graph representation of the speech graph signal is as follows: ,in Let be the node vectors of the graph.

[0018] In a preferred embodiment, in step S2, the intra-frame Laplacian matrix is ​​used. For the Frame Voice The Fourier projection of the graph is as follows: ,in express The left singular eigenvalue matrix is ​​obtained by using singular value decomposition to analyze the Laplacian matrix of the intra-frame graph. To decompose, that is:

[0019]

[0020] in Represents a diagonal matrix. This represents the right singular characteristic matrix.

[0021] In a preferred embodiment, in step S2, synchronous merging is performed. and The analysis of intra-frame and inter-frame oscillations can be represented as follows:

[0022]

[0023] in express The left singular characteristic matrix, i.e. , Represents a diagonal matrix. Describes the right singular eigenvalue matrix. The inverse logarithmic joint graph Fourier transform is .

[0024] In a preferred embodiment, in step S2, dynamic features of the speech signal, including frequency and intensity variation patterns of the speech, are further extracted by synchronously combining intra-frame and inter-frame oscillation analyses.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. This invention models the speech signal as a graph structure and uses logarithmic graph Fourier transform to perform frequency domain analysis on the speech signal, which can effectively extract the high-order structural information hidden in the speech signal. Compared with traditional spectral feature extraction methods, this invention can better describe the complex dynamic structure of the speech signal, thereby improving the discriminative power of playback speech detection.

[0027] 2. This invention constructs intra-frame and inter-frame graph Laplacian matrices and uses graph Fourier transform to project the speech signal, which can simultaneously capture the intra-frame and inter-frame dependencies of the speech signal. This method overcomes the shortcomings of the prior art in ignoring local inter-frame dependencies, fully extracts the spatiotemporal dynamic characteristics of the speech signal, and enhances the comprehensiveness and robustness of feature representation.

[0028] 3. This invention decomposes the graph Laplacian matrix through singular value decomposition and uses the left singular feature matrix to perform graph Fourier projection on the speech signal, which can extract more discriminative frequency domain features. At the same time, the combination of inverse logarithmic joint graph Fourier transform further enhances the accuracy and stability of feature representation, providing high-quality feature input for subsequent classification models.

[0029] In summary, by introducing logarithmic graph Fourier transform and graph signal processing methods, this invention effectively solves the limitations of existing technologies in playback speech detection, significantly improves the comprehensiveness of feature extraction, discrimination ability, and performance of the detection system, and provides a new technical direction for the field of speech security, which has important theoretical value and practical application significance. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of feature extraction in an embodiment of the present invention. Detailed Implementation

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] Please see Figure 1 This invention provides a technical solution: a speech detection method based on logarithmic graph Fourier transform feature extraction, characterized by including a graph representation construction stage S1, a feature extraction stage S2, and a detection stage S3. The graph representation construction stage S1 includes the following steps:

[0033] S1.1: Define the graph adjacency matrix using the dependencies between speech samples, and construct translation operators for the speech graph through forward and backward translation operations, which capture the relationships between speech samples.

[0034] S1.2: Use an exponential function to describe the decay of dependencies between speech samples, and combine intra-frame and inter-frame adjacency matrices to generate the Laplacian matrix of the graph.

[0035] S1.3: Based on the generated graph Laplacian matrix, the speech signal is represented as an undirected graph, capturing the structural relationships within and between frames.

[0036] Feature extraction stage S2 includes the following steps:

[0037] S2.1: Based on the constructed graph representation, each sample value of the speech signal is mapped to a graph node signal, and the node signal represents the time series features of the speech.

[0038] S2.2: The speech signal is transformed using the graph Laplacian matrix obtained in step S1.2. The logarithmic graph Fourier transform is applied to transform the speech signal from the time domain to the graph frequency domain, and the frequency domain features of the speech samples are extracted.

[0039] S2.3: Capture high-order structural information in speech signals in the graph frequency domain, especially implicit features related to speaker identity, including frequency and intensity variation patterns of speech.

[0040] S2.4: Combining frequency domain features and time domain features, enhanced feature representations are formed by feature concatenation. These features more comprehensively express the phonetic attributes and contextual information of the speech signal.

[0041] S2.5: Standardize or quantify the generated features so that they can be input into subsequent anonymization systems or attacker systems for speech processing and analysis.

[0042] Detection phase S3 includes the following steps:

[0043] S3.1: Using a pre-trained classification model, input the enhanced feature vector and extract the hidden layer representation of the classifier to further capture fine-grained speech signal features.

[0044] S3.2: Use the hidden layer output of the classifier to generate a detection score, which measures whether the input speech signal is subject to a replay attack or is normal speech. Calculate the performance metrics t-DCF (Tandem Detection Cost Function) and EER (Equal Error Rate) to evaluate the robustness and accuracy of the detection system.

[0045] In specific implementation, in step S1.1, the forward and backward shift operations of the speech signal are used, and the speech graph shift operator is represented as follows: .

[0046] In specific implementation, in step S1.2, the exponential function describing the decay of the dependency relationship between speech samples is expressed as:

[0047]

[0048] in It is an experimental constant. It is the experimental threshold.

[0049] In specific implementation, in step S1.3, the graph Laplacian matrix of the intra-frame speech... It can be represented as Graph Laplacian matrix of inter-frame speech It can be represented as - The diagonal matrix is ​​specified as and Utilizing the design and The undirected graph representation of the speech graph signal is as follows: ,in Let be the node vectors of the graph.

[0050] In specific implementation, in step S2.2, the intra-frame Laplacian matrix is ​​used. For the Frame Voice The Fourier projection of the graph is as follows: ,in express The left singular eigenvalue matrix is ​​obtained by using singular value decomposition to analyze the Laplacian matrix of the intra-frame graph. To decompose, that is:

[0051]

[0052] in Represents a diagonal matrix. This represents the right singular characteristic matrix.

[0053] In specific implementation, in step S2.4, synchronous merging is performed. and The analysis of intra-frame and inter-frame oscillations can be represented as follows:

[0054]

[0055] in express The left singular characteristic matrix, i.e. , Represents a diagonal matrix. Describes the right singular eigenvalue matrix. The inverse logarithmic joint graph Fourier transform is .

[0056] This invention provides a speech detection method based on logarithmic graph Fourier transform feature extraction. By constructing a speech graph representation, applying logarithmic graph Fourier transform to extract frequency domain features, and combining this with a pre-trained classification model for detection, it achieves efficient analysis and detection of speech signals. The detailed implementation describes the steps of the graph representation construction stage, feature extraction stage, and detection stage, including the definition of the graph adjacency matrix, the generation of the graph Laplacian matrix, the application of the logarithmic graph Fourier transform, and the detection process of the classification model. This invention can capture high-order structural information of speech signals, especially implicit features related to speaker identity, and forms enhanced feature representations by simultaneously merging intra-frame and inter-frame oscillation analysis, significantly improving the accuracy and robustness of speech detection.

[0057] The technical solution of this invention has a wide range of applications, such as voice anti-spoofing detection, playback attack detection, and speaker recognition. Compared with traditional methods, this invention achieves comprehensive analysis and efficient detection of voice signals by combining graph signal processing technology and machine learning models.

[0058] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A speech detection method based on logarithmic graph Fourier transform feature extraction, characterized in that, Includes the following steps: S1: Define the graph adjacency matrix based on the dependencies between speech samples, construct the translation operator of the speech graph, use the exponential function to describe the decay of the dependencies between speech samples, generate the Laplacian matrix of the graph, and represent the speech signal as an undirected graph to capture the structural relationships within and between frames; S2: Map the sample values ​​of the speech signal to graph node signals, apply the logarithmic graph Fourier transform using the graph Laplace matrix, transform the speech signal from the time domain to the graph frequency domain, extract frequency domain features, and simultaneously merge intra-frame and inter-frame oscillation analyses, combining time domain features to form an enhanced feature representation; S3: Use a pre-trained classification model to process the enhanced feature vector, generate a detection score to determine whether the speech signal is subject to a replay attack, and calculate the performance metrics t-DCF and EER to evaluate the system performance. In step S1, the speech graph translation operator is constructed through forward and backward shift operations of the speech signal. The speech graph shift operator is expressed as: ； In step S1, the exponential function describing the decay of dependencies between speech samples is expressed as: ； in It is an experimental constant. It is the experimental threshold; In step S1, the graph Laplacian matrix of intra-frame speech It can be represented as Graph Laplacian matrix of inter-frame speech It can be represented as - The diagonal matrix is ​​specified as and The undirected graph representation of the speech graph signal is as follows: ,in Let be the node vectors of the graph; In step S2, the intra-frame Laplacian matrix is ​​used. For the Frame Voice The Fourier projection of the graph is as follows: ,in express The left singular eigenvalue matrix is ​​obtained by using singular value decomposition to analyze the Laplacian matrix of the intra-frame graph. To decompose, that is: ； in Represents a diagonal matrix. Represents the right singular characteristic matrix; In step S2, through synchronous merging and The analysis of intra-frame and inter-frame oscillations can be represented as follows: ； in express The left singular characteristic matrix, i.e. , Represents a diagonal matrix. Describes the right singular eigenvalue matrix. The inverse logarithmic joint graph Fourier transform is ; In step S2, dynamic features of the speech signal, including frequency and intensity variation patterns, are further extracted by synchronously combining intra-frame and inter-frame oscillation analysis.

Images