A forged speech detection method using dual-dimensional graph attention

Abstract

The present invention relates to a forged speech detection method using a two-dimensional graph attention method. The method pre-emphasizes the original speech and extracts the linear frequency cepstral coefficient features of each frame. According to the linear frequency cepstral coefficient features, a BP algorithm is used to perform GMM modeling on the real speech and the forged speech. An expectation maximization algorithm is used to iteratively operate on the training data until convergence. The number of speech frames is counted and the LGP of the speech features of each frame in each Gaussian component is calculated respectively. Then, a constant term is subtracted and a normalization process is performed. The present invention uses a bidirectional subdivision method (BP) to model the Gaussian mixture model so that the difference in the mean vectors of adjacent Gaussian components is small. This not only enables the two-dimensional convolution to effectively capture the dependency relationship between the Gaussian components, but also provides a clearer and closer feature relationship for the graph attention module, and enhances the representation ability of the attention mechanism for global features. The method optimizes the feature space structure from local to global, thereby improving the performance of forged speech detection.

Description

Technical Field

[0001] The present invention relates to a forged speech detection method using dual-dimensional graph attention, belonging to the technical field of forged speech detection. Background Art

[0002] Spoofed speech detection technology is used to determine the authenticity of speech to be tested and prevent malicious attacks. It has important applications in fields such as Automatic Speaker Verification (ASV) and forensic examinations. Existing methods use the log-Gaussian probability (LGP) of the forged speech as a feature for detection based on Gaussian mixture model (GMM) components. However, such methods only perform one-dimensional convolution in the time dimension and fail to fully exploit the relationships between Gaussian components.

[0003] The present invention uses bidirectional partitioning (BP) to model GMM, so that the mean vector differences of adjacent Gaussian components are small. The dependency between Gaussian components is captured through two-dimensional convolution processing, and a graph attention module is introduced to aggregate information in the time dimension and Gaussian component dimension, thereby improving the accuracy of forged speech detection. Summary of the Invention

[0004] In order to overcome the shortcomings of existing forged speech detection methods that only perform one-dimensional convolution on the time dimension and fail to fully explore the relationship between Gaussian components, the present invention provides a forged speech detection method using two-dimensional graph attention, which can more deeply capture the correlation information between the Gaussian component dimension and the time dimension, enabling the model to more comprehensively analyze the structural characteristics of the input data, thereby improving the model's judgment ability.

[0005] A method for detecting forged speech using dual-dimensional graph attention, comprising the following steps:

[0006] Step 1: Pre-emphasize the original speech to enhance the details of the high-frequency waveform, then perform windowing and framing to extract the Linear Frequency Cepstrum Coefficients (LFCC) features of each frame.

[0007] Step 2: Based on the LFCC features, the BP algorithm is used to construct GMM models for both the real and fake speech. The Expectation Maximization (EM) algorithm is used to iteratively calculate the training data until convergence. The number of speech frames is counted and the LGP of each frame's speech features in each Gaussian component is calculated. The constant term is then subtracted and the model is normalized.

[0008] Step 3: Input the LGP feature matrix into six residual block modules. Each residual block consists of two 2D convolutional layers, a SeLU activation function, and a batch normalization layer. A 2×2 max pooling layer is added at the end of the first four residual blocks to reduce complexity and improve generalization ability.

[0009] Step 4: The deep features generated by the residual block are input into the temporal attention module and the Gaussian component attention module respectively to obtain graph nodes of the same dimension but different numbers. Both attention modules contain a graph attention network (GAT) and a graph pooling module (GPM). The graph nodes obtained by the two networks are then concatenated and input into the Gaussian component-temporal joint graph attention module to obtain the final graph node.

[0010] Step 5: The input features of the fake and real speech are processed in parallel. The nodes on both sides are flattened and concatenated as the input of the fully connected layer to generate confidence scores for the real and fake categories. The confidence scores are converted into probabilities through the softmax function. The weighted cross entropy loss function (WCE) is used to calculate the error between the actual label and the predicted label. The loss is minimized to optimize the classification performance of the model and improve the accuracy of the judgment of the authenticity of the speech.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] The present invention combines the deep features of the Gaussian component and time dimensions, processed separately by independent graph attention modules, with a concatenation operation to fuse them. Joint feature extraction is then performed using a Gaussian component-time joint graph attention module. This joint feature extraction approach captures the correlation between the Gaussian component and time dimensions more deeply, enabling the model to more comprehensively analyze the structural characteristics of the input data, thereby improving the model's decision-making capabilities.

[0013] This paper uses a bidirectional segmentation (BP) method to model the Gaussian mixture model, minimizing the difference in the mean vectors of adjacent Gaussian components. This strategy not only enables two-dimensional convolution to more effectively capture the dependencies between Gaussian components, but also provides the graph attention module with clearer and tighter feature relationships, significantly enhancing the attention mechanism's ability to represent global features. This method optimizes the feature space structure from a local to a global perspective, significantly improving the performance of forged speech detection.

[0014] This method builds Gaussian mixture models (GMMs) for both real and fake speech, extracts their features, and then feeds them into parallel models with the same structure but different parameters. The features of the real and fake speech are then fused at the fully connected layer, fully leveraging the relationship between them and further improving the model's accuracy in distinguishing real from fake speech. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 This is a flowchart and principle block diagram of the forged speech detection method using dual-dimensional graph attention of the present invention;

[0017] Figure 2 This is a schematic diagram of the GMM model generation and training principle of the forged speech detection method using dual-dimensional graph attention in the present invention;

[0018] Figure 3 This is a structural diagram of the residual block module of the forged speech detection method using dual-dimensional graph attention in the present invention;

[0019] Figure 4 This is the process and principle block diagram of the time attention module and Gaussian component attention module of the forged speech detection method using dual-dimensional graph attention in the present invention. DETAILED DESCRIPTION

[0020] 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.

[0021] Reference Figure 1, a forged speech detection method using dual-dimensional graph attention, comprising the following steps:

[0022] Step 1: Pre-emphasize the original speech to enhance the details of the high-frequency waveform, then perform windowing and framing to extract the Linear Frequency Cepstrum Coefficients (LFCC) features of each frame.

[0023] Step 1 specifically includes:

[0024] The input speech signal is first pre-emphasized using a first-order high-pass filter to boost high-frequency components and compensate for the natural attenuation of high-frequency energy. The signal is then framed into 20-30 millisecond frames, with a frame shift of 50% of the frame length to achieve overlap and preserve inter-frame correlation. Each frame is then windowed (e.g., using a Hamming window) to reduce edge discontinuities.

[0025] A Fast Fourier Transform (FFT) is performed on each frame of speech, converting the time-domain signal into the frequency domain and obtaining spectral information. The spectrum is then filtered using a triangular filter bank with linear frequency spacing. Each filter covers a specific frequency band, and the energy of the filter output is calculated.

[0026] The natural logarithm of the filter output energy is taken to make the data more consistent with the human ear's perception of loudness, and multiplication operations are converted to addition operations. Next, the resulting logarithmic energy sequence is subjected to a discrete cosine transform (DCT) to remove correlations between features and compress the data dimension, resulting in linear frequency scale cepstral coefficient (LFCC) features, where each coefficient represents the spectral characteristics of the speech signal.

[0027] In order to capture the dynamic changes of the speech signal, the first-order and second-order differential coefficients (i.e., Δ and ΔΔ coefficients) are further calculated. These differential coefficients reflect the changing trend of the features over time and improve the model's ability to model time series information.

[0028] Step 2: Based on the LFCC features, the BP algorithm is used to construct GMM models for both the real and fake speech. The Expectation Maximization (EM) algorithm is used to iteratively calculate the training data until convergence. The number of speech frames is counted and the LGP of each frame's speech features in each Gaussian component is calculated. The constant term is then subtracted and the model is normalized.

[0029] Step 2 specifically includes:

[0030] In this invention, a BP algorithm combined with GMM is used for training, gradually evolving a single-component Gaussian distribution into a multi-component Gaussian model. This method first initializes the GMM of a single Gaussian component using the global mean and variance, then gradually increases the number of Gaussian components to accommodate the complexity of the data. GMMs are used to model both real and fake speech, with λ representing the parameter set consisting of the mean vector, covariance matrix, and mixing weights of each Gaussian component, i.e.:

[0031] λ={ω i ,μ i ,∑ i},i=1,2,…,512 (1)

[0032] Among them, ω i is a mixed weight that satisfies μ i and ∑ i are the mean vector and covariance matrix of the i-th Gaussian component respectively. The model parameter set λ is obtained by the EM algorithm.

[0033] In each split of the BP algorithm, each Gaussian component is divided into two. This split is achieved by adding or subtracting a small perturbation to the mean vector of the original component. Specifically, given the mean vector μ and covariance matrix Σ of a Gaussian component, the mean vectors of the two newly generated Gaussian components are μ+ò and μ-ò, respectively, where ò is a small perturbation vector whose value is δ×eigenvector(Σ). δ is a preset constant, and eigenvector(Σ) represents the main eigenvector of the covariance matrix Σ, indicating the direction of maximum data variation, as shown in the following formula:

[0034] μ new1 = μ+δ×eigenvector(Σ) (2)

[0035] μ new2 = μ-δ×eigenvector(Σ) (3)

[0036] μ new1 and μ new2is the mean vector of the two newly generated Gaussian components, the covariance matrix Σ remains unchanged, and the initial weights of the two new Gaussian components are set to half of the original component weights to ensure that the sum remains consistent. After each split, the new GMM model contains more components, and then an appropriate number of EM iterations are performed to estimate the new GMM parameters to ensure that the model parameters can fully fit the data after the introduction of new components. When the last split is completed, the EM iteration process continues until the change in model parameters is lower than a certain threshold, or the preset number of iterations is reached, and GMM models that fit real speech and fake speech are obtained respectively. The principles of generating the GMM model and the training process are as follows Figure 2 shown.

[0037] After the GMM models of the real speech and the fake speech are established, the LFCC features of the speech are input into the two GMM models respectively to obtain the LGP features of each frame of speech features in the GMM. Assume x t Represents the LFCC feature vector of the t-th frame speech signal, and finds x t The probability density of each component is then logarithmized to obtain:

[0038]

[0039] where d is the vector x t The dimension of x is , T represents the transposition operator. In order to focus on the discriminative features, the features that directly reflect x are retained. t The probability of the items under different components. Therefore, only the items related to x in formula (4) are retained. t Related items, we get x t The logarithmic probability of the i-th Gaussian component is:

[0040]

[0041] After calculating the LGP of all speech frames in each Gaussian component t,i Afterwards, the data is normalized along the Gaussian component dimension to a distribution with a mean of 0 and a variance of 1, thereby obtaining the LGP feature matrices under the real speech GMM and the fake speech GMM respectively.

[0042] Step 3: Input the LGP feature matrix into six residual block modules. Each residual block consists of two 2D convolutional layers, a SeLU activation function, and a batch normalization layer. A 2×2 max pooling layer is added at the end of the first four residual blocks to reduce complexity and improve generalization ability.

[0043] The network architecture proposed in this paper is based on a deep residual learning framework and includes two stages of residual block processing. Before the LGP features are input into the residual network, they are first expanded or truncated and unified to 400 frames in the time dimension. In the first stage, the network consists of four residual blocks, each of which contains two two-dimensional convolutional layers with a convolution kernel size of (3,3) and 32 output channels. Batch normalization and SeLU activation functions are set between convolutional layers to enhance the nonlinearity and stability of the model. A 2×2 maximum pooling layer is also added after each residual block to reduce the feature dimension and reduce the subsequent computational burden.

[0044] In the second phase, unlike the first, the network includes two residual blocks, the number of output channels of the convolutional layer is increased to 64, and no pooling layer is added after each block. After six residual blocks, batch normalization and SeLU activation functions are applied again to further stabilize the feature distribution and enhance nonlinear expression capabilities, providing more discriminative feature inputs for the subsequent graph attention module.

[0045] This design effectively improves the expressiveness of features by gradually deepening the network depth and increasing the number of output channels. At the same time, the pooling layer is used to reduce the number of parameters and the amount of calculation, thereby improving the generalization ability of the model. The entire network aims to achieve efficient processing of complex data by balancing depth and width. The residual block module structure is as follows Figure 3 shown.

[0046] Step 4: The deep features generated by the residual block are input into the temporal attention module and the Gaussian component attention module respectively to obtain graph nodes of the same dimension but different numbers. Both attention modules contain a graph attention network (GAT) and a graph pooling module (GPM). The graph nodes obtained by the two networks are then concatenated and input into the Gaussian component-temporal joint graph attention module to obtain the final graph node.

[0047] Step 4 specifically includes:

[0048] In the graph attention module, the input feature Q has a dimension of (64, 25, 32), where 64 represents the number of feature channels, and 25 and 32 represent the number of time and Gaussian component dimensions, respectively. First, the input data is processed by the Gaussian component attention module to capture features along the Gaussian component dimension. A maximum pooling operation is performed along the time dimension to obtain feature representations along the Gaussian component dimension, resulting in a feature map of size (64, 1, 32). This feature map is then transformed to a representation of size (64, 32) and fed into the graph attention network (GAT) to generate attention-weighted graph features of size (32, 32). Subsequently, the graph features are pooled using the graph pooling module (GPM), selecting the most representative nodes to obtain graph features of size (32, 18).

[0049] The structure of the temporal attention module remains unchanged, and it is mainly used to capture feature information in the temporal dimension. First, the input features are max-pooled in the Gaussian component dimension to obtain a temporal feature representation with a dimension of (64, 25, 1). Next, a dimension conversion is performed to obtain a feature representation of (64, 25), which is input into the GAT to generate a graph feature with a dimension of (32, 25). Subsequently, the GPM is used to pool the graph nodes and select key nodes to obtain a representation with a dimension of (32, 14), providing graph features in the temporal dimension for the subsequent joint attention module.

[0050] In the Gaussian component-temporal joint attention module, the Gaussian component and graph features on the time dimension are concatenated as input and processed by the module to capture the joint information between the two. The concatenated feature dimension is (64, 32), which is input into the joint graph attention network for further feature extraction, and the output dimension is (16, 32). After that, the dimension is reduced to (16, 16) through GPM. Next, the feature of size (16, 16) is flattened into a 256-dimensional feature vector. The feature vectors of the real speech and the forged speech are concatenated into a 512-dimensional vector and input into the fully connected layer (FC layer) to generate confidence scores for the two categories of real / fake. Figure 4 This is the process and principle block diagram of the graph attention module.

[0051] Then, the two scores are mapped to the range of (0, 1) through softmax, and the sum of the two values ​​is 1, representing the probability of belonging to each category. Finally, the cross-entropy loss function is used to calculate the error between the actual label and the predicted label. By minimizing the cross-entropy loss, the classification performance of the model is optimized and the accuracy of the judgment of the authenticity of the input speech is improved.

[0052] The Graph Attention Network (GAT) includes:

[0053] In GAT, the core operation is to aggregate the information of a node and its neighboring nodes through the self-attention mechanism to capture the relationship between nodes and generate more discriminative features. The input of the GAT layer is a graph G = (N, E, H), where N represents the node set, E represents the edge set, and H is the feature representation of each node. First, for each node n, traverse its neighboring node set M(n), and use w map Calculate the attention coefficient a between node n and its neighbor node u u,n :

[0054]

[0055] Among them, h n and h u Represent the original features of itself and neighbor nodes respectively, ⊙ represents the multiplication of the corresponding elements of the features, w map It is a weight vector with the same dimension as the node feature, which is used to weight the elements one by one to obtain the attention score that measures the strength of the relationship between nodes.

[0056] Then, the aggregated feature m of the central node n is obtained by weighted summation n :

[0057]

[0058] Then introduce the learnable mapping matrix W att and W res , used to convert m n and h n Projected to the same target dimension d' (32 in this invention), the new feature representation o is obtained after addition. n :

[0059] o n =SeLU(BN(W att m n +W res h n )) (8)

[0060] Among them, SeLU represents the activation function and BN represents batch normalization. This structure is achieved by fusing the aggregated features m n and its own characteristics h n , so that o n It also contains local neighborhood information and its own inherent information, further improving the discriminative ability of node representation, and combines SeLU and BN to enhance the stability and expression ability of the model.

[0061] The graph pooling module (GPM) specifically includes:

[0062] The main function of GPM is to select the most representative nodes to improve the discriminative power of the model. Its implementation process is as follows:

[0063] When the input graph contains N nodes, the node feature matrix is ​​H∈R N×d , where d is the dimension of node features. GPM first passes a learnable projection vector p∈R d , calculate the projection score s for each node i :

[0064]

[0065] h i is the feature representation of node i, p T Represents the transpose of the projection vector. Then according to the projection score s i Sort all nodes, select the first k nodes with the largest projection scores, and get the corresponding node index set idh. The node feature matrix is ​​updated to H'∈R k×d :

[0066] H'=H idh ×sigmoid(s idh ) (10)

[0067] Among them, H idh and s idh They represent the features of the selected nodes and the corresponding projection scores, respectively. Unselected nodes will be discarded. GPM reduces computational complexity while retaining important structural information, which helps improve the classification performance of the model.

[0068] Step 5: The input features of the fake and real speech are processed in parallel. The nodes on both sides are flattened and concatenated as the input of the fully connected layer to generate confidence scores for the real and fake categories. The confidence scores are converted into probabilities through the softmax function. The weighted cross entropy loss function (WCE) is used to calculate the error between the actual label and the predicted label. The loss is minimized to optimize the classification performance of the model and improve the accuracy of the judgment of the authenticity of the speech.

[0069] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It is apparent to those skilled in the art that various changes, modifications, substitutions, and variations to these embodiments may be made without departing from the principles and spirit of the present invention, and these changes and modifications still fall within the scope of protection of the present invention.

Claims

1. A forged speech detection method using dual-dimensional graph attention, characterized by: The following steps are involved: Step 1: First, pre-emphasize the original speech to enhance the details of the high-frequency waveform, then perform windowing and framing to extract the linear frequency cepstral coefficient features of each frame; Step 2: Based on the linear frequency cepstral coefficient features, the BP algorithm is used to perform GMM modeling on the real and fake speech respectively. The expectation maximization algorithm is used to iteratively operate on the training data until convergence. The number of speech frames is counted and the LGP of each frame's speech features in each Gaussian component is calculated respectively. The constant term is then subtracted and normalized. Step 3: Input the LGP feature matrix into six residual block modules. Each residual block consists of two 2D convolutional layers, a SeLU activation function, and a batch normalization layer. A 2×2 max pooling layer is added at the end of the first four residual blocks to reduce complexity and improve generalization ability. Step 4: The deep features generated by the residual block are input into the temporal attention module and the Gaussian component attention module respectively to obtain graph nodes of the same dimension but different numbers. Both attention modules contain a graph attention network and a graph pooling module. The graph nodes obtained by the two networks are then concatenated and input into the Gaussian component-temporal joint graph attention module to obtain the final graph node. Step 5: The input features of the fake and real speech are processed in parallel. The nodes on both sides are flattened and concatenated as the input of the fully connected layer to generate confidence scores for the real / fake categories. The confidence scores are converted into probabilities through the softmax function. The weighted cross-entropy loss function is used to calculate the error between the actual label and the predicted label. The loss is minimized to optimize the classification performance of the model and improve the accuracy of the judgment of the authenticity of the speech.

2. The method for detecting forged speech using dual-dimensional graph attention according to claim 1, wherein: The pre-emphasis processing and extraction of linear frequency cepstral coefficient features in step 1 specifically include: The input speech signal is first pre-emphasized, using a first-order high-pass filter to enhance high-frequency components and compensate for the natural attenuation of high-frequency energy. The signal is then framed into 20-30 millisecond frames, with a frame shift of 50% of the frame length to achieve overlap between adjacent frames and preserve correlation between signal frames. Each frame of the speech signal is windowed to reduce edge discontinuities. Perform fast Fourier transform on each frame of speech to convert the time domain signal into a frequency domain signal to obtain spectrum information. Then use a triangular filter bank with linear frequency intervals to filter the spectrum. The natural logarithm of the energy output by the filter is taken to make the data more consistent with the human ear's perception of loudness. The multiplication operation is converted into an addition operation, and the obtained logarithmic energy sequence is subjected to discrete cosine transform to remove the correlation between features and compress the data dimension to obtain the linear frequency scale cepstral coefficient feature, where each coefficient represents the spectral feature of the speech signal.

3. The method for detecting forged speech using dual-dimensional graph attention according to claim 1, wherein: The GMM modeling and LGP feature extraction in step 2 specifically include: The BP algorithm is combined with GMM for training, and the Gaussian distribution of a single component is gradually evolved into a Gaussian model with multiple components. The global mean and variance are first used to initialize the GMM of a single Gaussian component, and then the number of Gaussian components is gradually increased to adapt to the complexity of the data. GMM is used to model real and fake speech respectively, and λ is used to represent the parameter set composed of the mean vector, covariance matrix and mixing weight of each Gaussian component, that is: λ={ω i ,m i ,∑ i },i=1,2,…,512 (1) Among them, ω i is a mixed weight that satisfies μ i and ∑ i They represent the mean vector and covariance matrix of the i-th Gaussian component respectively, and the model parameter set λ is obtained by the EM algorithm; In each split of the BP algorithm, each Gaussian component is divided into two. This split is achieved by adding or subtracting a small perturbation to the mean vector of the original component. Given the mean vector μ and covariance matrix Σ of a Gaussian component, the mean vectors of the two newly generated Gaussian components are μ+ò and μ-ò, respectively. ò is a small perturbation vector whose value is δ×eigenvector(Σ), δ is a preset constant, and eigenvector(Σ) represents the main eigenvector of the covariance matrix Σ, indicating the direction of maximum data variation. The formula is as follows: m new1 = μ+δ×eigenvector(Σ) (2) m new2 = μ-δ×eigenvector(Σ) (3) μ new1 and μ new2 The mean vector of the two newly generated Gaussian components is used, and the covariance matrix Σ remains unchanged. The initial weights of the two new Gaussian components are set to half of the original component weights to ensure that the sum remains consistent. After each split, the new GMM model contains more components, and an appropriate number of EM iterations are performed to estimate the new GMM parameters to ensure that the model parameters can fully fit the data after the introduction of new components. When the last split is completed, the EM iteration process continues until the change in model parameters is lower than a certain threshold or the preset number of iterations is reached, and the GMM models that fit the real speech and the fake speech are obtained respectively; After the GMM models of real speech and fake speech are established, the LFCC features of the speech are input into the two GMM models respectively to obtain the LGP features of each frame of speech features in the GMM. Assuming x t Represents the LFCC feature vector of the t-th frame speech signal, and finds x t The probability density of each component is then logarithmized to obtain: where d is the vector x t The dimension of x is , T represents the transposition operator, in order to focus on the discriminative features, retain the direct reflection of x t The probability of the items under different components is only retained in formula (4) t Related items, we get x t The logarithmic probability of the i-th Gaussian component is: After calculating the LGP of all speech frames in each Gaussian component t,i Afterwards, the data is standardized along the Gaussian component dimension to a distribution with a mean of 0 and a variance of 1, and the LGP feature matrices under the real speech GMM and the fake speech GMM are obtained respectively.

4. The method for detecting forged speech using dual-dimensional graph attention according to claim 1, wherein: The step 4 specifically includes: In the module, the dimension of the input feature Q is (64, 25, 32), where 64 represents the number of feature channels, 25 and 32 are the number of time dimensions and the number of Gaussian component dimensions, respectively. The input data is processed by the Gaussian component attention module to capture the features on the Gaussian component dimension. The maximum pooling operation is performed along the time dimension to obtain the feature representation on the Gaussian component dimension, and a feature map of size (64, 1, 32) is obtained. The feature map is dimensionally transformed to obtain a representation of dimension (64, 32) and input into the graph attention network to generate attention-weighted graph features of dimension (32, 32). The graph features are pooled using the graph pooling module (GPM) to select the most representative nodes to obtain graph features of dimension (32, 18).

5. The method for detecting forged speech using dual-dimensional graph attention according to claim 4, wherein: The graph attention network in step 4 specifically includes: In the graph attention network, the information of nodes and their neighboring nodes is aggregated through the self-attention mechanism to capture the relationship between nodes and generate more discriminative features. The input of the graph attention network layer is the graph G = (N, E, H), where N represents the node set, E represents the edge set, and H is the feature representation of each node. For each node n, its neighboring node set M(n) is traversed and the neighboring nodes are obtained through w map Calculate the attention coefficient a between node n and its neighbor node u u,n : Among them, h n and h u Represent the original features of itself and neighbor nodes respectively, ⊙ represents the multiplication of the corresponding elements of the features, w map It is a weight vector with the same dimension as the node feature, which is used to weight the elements one by one to obtain the attention score that measures the strength of the relationship between nodes; The aggregated feature m of the central node n is obtained by weighted summation n : Then the learnable mapping matrix W is introduced att and W res , used to convert m n and h n Projected to the same target dimension d', the new feature representation o is obtained after addition. n : o n =SeLU(BN(W att m n +W res h n )) (8) Where SeLU represents the activation function and BN represents batch normalization.

6. The method for detecting forged speech using dual-dimensional graph attention according to claim 4, wherein: The image pooling module in step 4 specifically includes: When the input graph contains N nodes, the node feature matrix is ​​H∈R N×d , where d is the dimension of node features, GPM first passes a learnable projection vector p∈R d , calculate the projection score s for each node i : s i =p T h i (9) h i is the feature representation of node i, p T Represents the transpose of the projection vector, and then according to the projection score s i Sort all nodes, select the first k nodes with the largest projection scores, get the corresponding node index set idh, and update the node feature matrix to H'∈R k×d : H'=H idh ×sigmoid(s idh ) (10) Among them, H idh and s idh They represent the features of the selected nodes and the corresponding projection scores respectively. The unselected nodes will be discarded.

Images