A method for detecting counterfeit audio based on multi-dimensional data analysis

Abstract

The application discloses a kind of based on multi-dimensional data analysis's counterfeit audio detection method, it is related to audio analysis technical field, step S1, obtains identification audio and carries out identification data extraction, the identification data includes audio basic data and audio related data;Step S2, audio basic data is analyzed and exports audio influence coefficient, according to audio influence coefficient, analysis is carried out, and the generation of audio basic mark is carried out to identification audio;Step S3, audio related data is handled and analyzed, and exports relevant influence coefficient;Step S4, relevant influence coefficient and audio influence coefficient are analyzed by weight ratio, and the generation of environmental influence degree mark is carried out to identification audio;Step S5, audio basic mark and environmental influence degree mark in identification audio are integrated and handled, and generate fake result target;Through the analysis of identification equipment and real-time climate, the identification error caused by temperature factor of identification equipment is excluded, and the accuracy of identification is improved.

Description

Technical Field

[0001] This invention relates to the field of audio analysis, and more specifically to a method for detecting fake audio based on multidimensional data analysis. Background Technology

[0002] In the audio authentication methods for cash, most methods use models that analyze basic audio data to determine the authenticity of the audio.

[0003] A search revealed a priori document with publication number CN108806718A, which proposes an audio identification method based on the analysis of the ENF phase spectrum and instantaneous frequency spectrum. The document analyzes the phase spectrum and instantaneous frequency spectrum of the ENF signal, which are sensitive to signal truncation, extracts effective feature sets, and processes these features. Feature-level fusion technology is used to process the feature data, reducing the feature dimensionality while improving the recognition gap. Deep learning methods are applied for model training, significantly improving the accuracy of passive tampering detection in digital audio. The method demonstrates high stability and robustness against recordings in complex environments and noisy speech. This provides a widely applicable algorithm for improving the accuracy and automation of passive tampering detection in digital audio.

[0004] Based on the prior art documents, the following shortcomings were found in the existing technology:

[0005] 1. In most cases, the data processing logic for audio identification in existing technologies is too simplistic and has flaws in identification.

[0006] 2. Most existing technologies perform noise reduction analysis on environmental factors, but never consider whether the identification equipment itself has identification faults, which may lead to errors in the identification results.

[0007] To address the aforementioned problems, a method for detecting fake audio based on multidimensional data analysis is proposed. Summary of the Invention

[0008] The purpose of this invention is to provide a method for detecting fake audio based on multidimensional data analysis, so as to overcome the shortcomings of the prior art.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] The method for detecting fake audio based on multidimensional data analysis includes the following steps:

[0011] Step S1: Obtain the identification audio and extract the identification data, which includes basic audio data and audio-related data;

[0012] Step S2: Analyze the basic audio data and output the audio influence coefficient. Based on the audio influence coefficient, analyze the audio data and generate basic audio tags for the identification audio.

[0013] Step S3: Process and analyze the audio-related data, and output the relevant influence coefficients;

[0014] Step S4: Perform a weighted ratio analysis on the relevant influence coefficient and the audio influence coefficient, and generate an environmental impact degree marker for the identified audio.

[0015] Step S5: Integrate the basic audio markers and environmental influence markers in the identified audio to generate a forged result target.

[0016] In a preferred embodiment, the audio fundamental data includes spectral amplitude, short-time energy, and spectral smoothness. The spectral amplitude represents the intensity of different frequency components in the audio. A frequency component with a larger spectral amplitude means that the audio signal intensity at that frequency is higher, and the likelihood of it being a genuine audio recording is greater. The short-time energy refers to the energy value of the audio signal in each time period, which can be used to reflect the strength changes of the audio signal in the time domain. A larger short-time energy indicates that the amplitude of the audio signal is larger in that time period, and the higher the authenticity of the audio. The spectral smoothness is used to measure the smoothness of the changes in the audio signal spectrum in the frequency direction. A larger spectral smoothness indicates that the spectral changes are more gradual. The greater the spectral smoothness, the smoother the changes in the audio spectrum, and the greater the likelihood of it being a genuine recording.

[0017] In a preferred embodiment, the audio-related data includes the temperature of the identification device, the real-time climate temperature, and the real-time climate wind speed. The temperature of the identification device represents the real-time temperature at which the audio is being identified. The higher the temperature, the greater the impact on the identification device during the identification process, leading to inaccuracies in the identification results. The real-time climate temperature represents the ambient temperature at the time the audio is being identified. The higher the temperature, the greater the temperature impact on the identification device. The real-time climate wind speed is the ambient wind speed at the time the audio is being identified. The stronger the wind, the lower the surface temperature.

[0018] In a preferred embodiment, the audio influence coefficient analysis output process is as follows:

[0019] The spectral amplitude value h, short-time energy k, and spectral smoothness i of the identification audio are obtained. These values ​​are then integrated and analyzed. The sum of the short-time energy k and the spectral smoothness i is multiplied by the spectral amplitude value h, and the cube root of the product is calculated. An audio analysis error constant m is then added to obtain the audio influence coefficient Hik. The audio influence coefficient Hik is always greater than 0 and less than 1. Hik reflects the basic quality of the identification audio itself; a higher Hik indicates higher quality.

[0020] In a preferred embodiment, the audio base tags include low-fidelity audio tags, mid-fidelity audio tags, and high-fidelity audio tags, and the steps for generating the audio base tags are as follows:

[0021] Set reference values ​​H1 and H2 for audio impact coefficients, where H1 is less than 1 and greater than H2.

[0022] The audio influence coefficient Hik is substituted into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than 0 and less than the audio influence coefficient reference value H2, a low-fidelity audio label is generated for the identified audio.

[0023] Substitute the audio influence coefficient Hik into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than the audio influence coefficient reference value H2 and less than the audio influence coefficient reference value H1, generate a fidelity mark in the audio for the identified audio.

[0024] Substitute the audio influence coefficient Hik into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than the audio influence coefficient reference value H1 and less than 1, generate an audio high-fidelity mark for the identified audio.

[0025] Audio high-fidelity markers identify audio with higher audio quality and greater authenticity compared to audio medium-fidelity markers, and so on.

[0026] In a preferred embodiment, the output process for the relevant influence coefficient is as follows:

[0027] The device temperature t1, real-time climate temperature t2, and real-time climate wind force value f are obtained. The difference between the real-time climate temperature t2 and the device temperature t1 is calculated and multiplied by the temperature error value q to obtain the first product. The correlation influence coefficient Tf is obtained by dividing the cube of the climate wind force value f by the absolute value of the square root of the first product. The correlation influence coefficient Tf is greater than 0. The larger the correlation influence coefficient Tf, the smaller the influence of the identification device on the identification audio.

[0028] In a preferred embodiment, the environmental impact level marker includes a high environmental impact marker and a low environmental impact marker, and the steps for generating the environmental impact level marker are as follows:

[0029] Weighted analysis was performed on the correlation coefficient Tf and the audio influence coefficient Hik to obtain the coefficient comparison parameter Δ. The weight parameters corresponding to the correlation coefficient Tf and the audio influence coefficient Hik are G1 and G2, respectively, where G1+G2=1.5624.

[0030] The larger the coefficient comparison parameter Δ, the greater the authenticity of the audio. Set the coefficient comparison threshold Wh, which is greater than 0.

[0031] The coefficient comparison parameter Δ is substituted into the coefficient comparison threshold Wh for comparison analysis. When the coefficient comparison parameter Δ is greater than 0 and less than the coefficient comparison threshold Wh, it is marked as having a high degree of influence on the identification audio generation environment.

[0032] The coefficient comparison parameter Δ is substituted into the coefficient comparison threshold Wh for comparison analysis. When the coefficient comparison parameter Δ is greater than the coefficient comparison threshold Wh, the influence of the identification audio generation environment is marked as low.

[0033] Compared to markers with high environmental impact, markers with low environmental impact have less environmental impact in the audio samples being identified, and the identification equipment also has less influence on the identification process.

[0034] In a preferred embodiment, the forged result target includes a real target, a suspicious target, and a forged target, and the steps for generating the forged result target are as follows:

[0035] The audio base markers and environmental impact level markers in the identification audio are integrated and analyzed. If the identification audio has both high-fidelity audio markers and low-level environmental impact markers, a real target is generated for the identification interval.

[0036] If the identified audio simultaneously possesses a high-fidelity audio marker and a high-level environmental influence marker, a medium-fidelity audio marker and a high-level environmental influence marker, or a medium-fidelity audio marker and a low-level environmental influence marker, a suspicious target is generated for that identification interval.

[0037] If the identified audio simultaneously contains both low-fidelity audio markers and low-level environmental influence markers, or both low-fidelity audio markers and high-level environmental influence markers, a forgery target is generated for that identification interval.

[0038] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0039] 1. The quality of the identification audio is analyzed by superimposing multiple data such as spectral amplitude, short-time energy and spectral smoothness, so as to obtain the basic audio quality of the identification audio more accurately, and make a preliminary audio authenticity estimate based on its basic audio quality.

[0040] 2. By analyzing the identification equipment and real-time climate data, the identification errors caused by temperature and other factors during the identification process are eliminated, thereby improving the accuracy of the identification.

[0041] 3. By performing a second authenticity assessment on the audio after excluding the influence of identification equipment and environmental factors, the accuracy of identification is further improved through repeated authenticity assessments. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0043] Figure 1 This is a flowchart of a method for detecting fake audio based on multidimensional data analysis according to the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Please see Figure 1 As shown in this embodiment, a method for detecting fake audio based on multidimensional data analysis is described. Therefore, a method for detecting fake audio based on multidimensional data analysis includes the following steps:

[0046] Step S1: Obtain the identification audio and extract the identification data, which includes basic audio data and audio-related data;

[0047] Step S2: Analyze the basic audio data and output the audio influence coefficient. Based on the audio influence coefficient, analyze the audio data and generate basic audio tags for the identification audio.

[0048] Step S3: Process and analyze the audio-related data, and output the relevant influence coefficients;

[0049] Step S4: Perform a weighted ratio analysis on the relevant influence coefficient and the audio influence coefficient, and generate an environmental impact degree marker for the identified audio.

[0050] Step S5: Integrate the basic audio markers and environmental influence markers in the identified audio to generate a forged result target.

[0051] The basic audio data includes spectral amplitude, short-time energy, and spectral smoothness. The spectral amplitude represents the intensity of different frequency components in the audio. A larger spectral amplitude indicates a higher intensity of the audio signal at that frequency, making it more likely to be a genuine recording. The short-time energy refers to the energy value of the audio signal within each time period, reflecting the strength changes of the audio signal in the time domain. A larger short-time energy indicates a larger amplitude of the audio signal within that time period, resulting in a higher degree of audio authenticity. The spectral smoothness measures the smoothness of the audio signal's spectrum changes in the frequency direction. A larger spectral smoothness indicates a more gradual spectral change, and the greater the spectral smoothness, the more likely it is to be a genuine recording.

[0052] It should be noted that these features can be extracted using digital signal processing algorithms. Commonly used algorithms include:

[0053] Spectral amplitude value: The spectral amplitude value can be calculated using the Fast Fourier Transform (FFT), which can convert an audio signal in the time domain into a spectrum in the frequency domain;

[0054] Short-time energy: Short-time energy can be obtained by calculating the sum of squares of the audio signal in each time interval;

[0055] Spectral smoothness: Spectral smoothness can be obtained by calculating the first or second derivative of the spectrum, or by using other smoothing methods to obtain a smooth spectrum;

[0056] These features can be extracted using various audio processing libraries and software, such as the Librosa library in Python and the Signal Processing Toolbox in MATLAB.

[0057] The audio-related data includes the temperature of the identification device, the real-time climate temperature, and the real-time climate wind speed. The temperature of the identification device represents the real-time temperature at which the audio is being identified. The higher the temperature, the greater the impact on the identification device during the identification process, leading to inaccurate identification results. The real-time climate temperature represents the ambient temperature at the time the audio is being identified. The higher the temperature, the greater the temperature impact on the identification device. The real-time climate wind speed is the ambient wind speed at the time the audio is being identified. The stronger the wind, the lower the surface temperature.

[0058] It should be noted that the methods for obtaining the temperature of the identification equipment, real-time climate temperature, and real-time climate wind speed values ​​are as follows:

[0059] Temperature of the testing equipment: The testing equipment has a built-in temperature sensor, and the sensor data can be read directly through the interface or driver provided by the operating system.

[0060] Real-time climate temperature: Real-time temperature data can be obtained from the Internet by calling the API of a weather website or using a weather data query service;

[0061] Real-time wind speed values: Real-time wind speed data can be obtained from the Internet by calling the API of a weather website or using a weather data query service.

[0062] The audio impact coefficient analysis output process is as follows:

[0063] The spectral amplitude value h, short-time energy k, and spectral smoothness i in the identified audio are obtained. These values ​​are then integrated and analyzed. The sum of the short-time energy k and the spectral smoothness i is multiplied by the spectral amplitude value h, and the cube root of the product is calculated. This product is then added to the audio analysis error constant m to obtain the audio influence coefficient Hik. The formula for calculating the audio influence coefficient Hik is as follows:

[0064]

[0065] The audio influence coefficient Hik is always greater than 0 and less than 1. The audio influence coefficient Hik reflects the basic quality of the audio itself. The higher the audio influence coefficient Hik, the higher the quality of the audio being identified.

[0066] The audio base tags include low-fidelity audio tags, medium-fidelity audio tags, and high-fidelity audio tags. The steps for generating the audio base tags are as follows:

[0067] Set reference values ​​H1 and H2 for audio impact coefficients, where H1 is less than 1 and greater than H2.

[0068] The audio influence coefficient Hik is substituted into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than 0 and less than the audio influence coefficient reference value H2, a low-fidelity audio label is generated for the identified audio.

[0069] Substitute the audio influence coefficient Hik into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than the audio influence coefficient reference value H2 and less than the audio influence coefficient reference value H1, generate a fidelity mark in the audio for the identified audio.

[0070] Substitute the audio influence coefficient Hik into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than the audio influence coefficient reference value H1 and less than 1, generate an audio high-fidelity mark for the identified audio.

[0071] Audio high-fidelity markers identify audio with higher audio quality and greater authenticity compared to audio medium-fidelity markers, and so on.

[0072] It should be noted that the audio base tag generation specifically employs a classification model training algorithm, which can be implemented using a support vector machine (SVM) or a convolutional neural network (CNN).

[0073] The steps for building this classification model are as follows:

[0074] Data collection: Collect a batch of sample data containing different audio, each audio with a corresponding audio influence coefficient Hik;

[0075] Feature extraction: Extract features from each audio file, which may include spectral amplitude values, short-time energy, and spectral smoothness;

[0076] Training set partitioning: The dataset is divided into a training set and a test set for model training and validation;

[0077] Build a classification model: Establish a classification model, such as a support vector machine (SVM) or a neural network (such as a convolutional neural network, CNN), to associate audio impact coefficients and features with audio quality levels (low fidelity, medium fidelity, high fidelity);

[0078] Model training: The classification model is trained using the training set so that it can accurately map audio impact coefficients and features to the corresponding audio quality levels;

[0079] Model validation: Use a test set to validate the model and evaluate its performance and accuracy;

[0080] Tag generation: Based on the model's prediction results, generate corresponding audio quality tags (low fidelity, medium fidelity, high fidelity) for each identified audio.

[0081] Audio quality assessment: The generated audio quality markers are compared with the audio influence coefficient reference values ​​H1 and H2 to determine the authenticity and quality of the audio.

[0082] Output results: Based on the identification results, the audio is labeled as low fidelity, medium fidelity, or high fidelity to indicate its authenticity and quality level.

[0083] The output process for the relevant influence coefficient is as follows:

[0084] Obtain the equipment temperature t1, real-time climate temperature t2, and real-time climate wind force value f. Calculate the difference between the real-time climate temperature t2 and the equipment temperature t1 and multiply it by the temperature error value q to obtain the first product. Then, obtain the correlation influence coefficient Tf by dividing the cube of the climate wind force value f by the absolute value of the square root of the first product. The formula for obtaining the correlation influence coefficient Tf is as follows:

[0085]

[0086] The correlation coefficient Tf is greater than 0. The larger the correlation coefficient Tf is, the smaller the influence of the identification device on the identification audio.

[0087] The environmental impact level markers include high-level environmental impact markers and low-level environmental impact markers. The steps for generating the environmental impact level markers are as follows:

[0088] Weighted analysis was performed on the correlation coefficient Tf and the audio influence coefficient Hik to obtain the coefficient comparison parameter Δ. The weight parameters corresponding to the correlation coefficient Tf and the audio influence coefficient Hik are G1 and G2, respectively, where G1+G2=1.5624.

[0089] The specific formula for calculating the coefficient comparison parameter Δ is as follows:

[0090] Δ = G1*Tf + G2*Hik (Δ > 0);

[0091] The larger the coefficient comparison parameter Δ, the greater the authenticity of the audio. Set the coefficient comparison threshold Wh, which is greater than 0.

[0092] The coefficient comparison parameter Δ is substituted into the coefficient comparison threshold Wh for comparison analysis. When the coefficient comparison parameter Δ is greater than 0 and less than the coefficient comparison threshold Wh, it is marked as having a high degree of influence on the identification audio generation environment.

[0093] The coefficient comparison parameter Δ is substituted into the coefficient comparison threshold Wh for comparison analysis. When the coefficient comparison parameter Δ is greater than the coefficient comparison threshold Wh, the influence of the identification audio generation environment is marked as low.

[0094] Compared to markers with high environmental impact, markers with low environmental impact have less environmental impact in the audio samples being identified, and the identification equipment also has less influence on the identification process.

[0095] It should be noted that the step of generating the environmental impact degree label also involves a classification model. The classification algorithm used can be logistic regression, which is suitable for binary classification problems. It can associate the relevant impact coefficient comparison parameter Δ with the degree of environmental impact and generate labels for high and low environmental impact.

[0096] The specific training steps are as follows:

[0097] Collect labeled audio datasets, including labels of environmental impact levels and values ​​of the relevant impact coefficient comparison parameter Δ;

[0098] The audio data is preprocessed, and the relevant influence coefficient comparison parameter Δ is extracted as a feature.

[0099] Divide the dataset into training and testing sets;

[0100] Use logistic regression or other classification algorithms to train the model on the training set and establish a model for identifying the degree of environmental impact;

[0101] To evaluate model performance on the test set, metrics such as accuracy, recall, and precision can be used.

[0102] Adjustments and optimizations are made based on the model's evaluation results until satisfactory performance is achieved;

[0103] The final model can be used to determine the degree of environmental impact of new audio samples to be identified, and generate labels indicating high or low environmental impact.

[0104] The forged result targets include real targets, suspicious targets, and forged targets. The steps for generating the forged result targets are as follows:

[0105] The basic audio markers and environmental impact level markers in the identification audio are integrated and analyzed. If the identification audio simultaneously has both high-fidelity audio markers and low-level environmental impact markers, a real target is generated for the identification interval.

[0106] If the identified audio simultaneously possesses a high-fidelity audio marker and a high-level environmental influence marker, a medium-fidelity audio marker and a high-level environmental influence marker, or a medium-fidelity audio marker and a low-level environmental influence marker, a suspicious target is generated for that identification interval.

[0107] If the identified audio simultaneously contains both low-fidelity audio markers and low-level environmental influence markers, or both low-fidelity audio markers and high-level environmental influence markers, a forgery target is generated for that identification interval.

[0108] The invention employs a multi-data overlay processing method, combining spectral amplitude, short-time energy, and spectral smoothness, to analyze the quality of the identification audio. This allows for a more accurate acquisition of the basic audio quality, which is then used to make a preliminary estimation of audio authenticity. Furthermore, by analyzing the identification equipment and real-time climate data, errors caused by temperature and other factors during the identification process are eliminated, thus improving the accuracy of the identification. Finally, after eliminating the influence of identification equipment and environmental factors, a second authenticity estimation is performed on the audio. Through repeated authenticity estimation, the accuracy of the identification is further improved.

[0109] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0110] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0111] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0112] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0113] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0114] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0115] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0116] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting fake audio based on multidimensional data analysis, characterized in that, The method includes the following steps: Step S1: Obtain the identification audio and extract the identification data. The identification data includes basic audio data and audio-related data. The basic audio data includes spectral amplitude value, short-time energy and spectral smoothness. The audio-related data includes identification device temperature, real-time climate temperature and real-time climate wind force value. Step S2: Analyze the basic audio data and output the audio influence coefficient. Based on the audio influence coefficient, analyze the audio data and generate basic audio tags for the identification audio. Step S3: Process and analyze the audio-related data, and output the relevant influence coefficients; Step S4: Perform a weighted ratio analysis on the relevant influence coefficient and the audio influence coefficient, and generate an environmental impact degree marker for the identified audio. Step S5: Integrate the basic audio markers and environmental influence markers in the identified audio to generate a forged result target.

2. The method for detecting fake audio based on multidimensional data analysis according to claim 1, characterized in that, The spectral amplitude value represents the intensity of different frequency components in the audio. A frequency component with a larger spectral amplitude value means that the audio signal intensity at that frequency is higher, and the greater the likelihood that it is an authentic audio recording. The short-time energy refers to the energy value of the audio signal in each time period, which can be used to reflect the intensity change of the audio signal in the time domain. A larger short-time energy indicates that the amplitude of the audio signal is larger in that time period, and the higher the authenticity of the audio. The spectral smoothness is used to measure the smoothness of the change of the audio signal spectrum in the frequency direction. A larger spectral smoothness indicates that the spectral change is more gradual. The greater the spectral smoothness, the smoother the change of the audio spectrum, and the greater the likelihood that it is an authentic recording.

3. The method for detecting fake audio based on multidimensional data analysis according to claim 2, characterized in that, The temperature of the identification device represents the real-time temperature at which the audio is being identified. The higher the temperature, the greater the impact on the identification device during the identification process, leading to inaccurate identification results. The real-time climate temperature represents the ambient temperature at the time the audio is being identified. The higher the temperature, the greater the impact of temperature on the identification device. The real-time climate wind force value represents the ambient wind force at the time the audio is being identified. The stronger the wind, the lower the surface temperature.

4. The method for detecting fake audio based on multidimensional data analysis according to claim 3, characterized in that, The audio impact coefficient analysis output process is as follows: The spectral amplitude value h, short-time energy k, and spectral smoothness i of the identification audio are obtained. These values ​​are then integrated and analyzed. The sum of the short-time energy k and the spectral smoothness i is multiplied by the spectral amplitude value h, and the cube root of the product is calculated. An audio analysis error constant m is then added to obtain the audio influence coefficient Hik. The audio influence coefficient Hik is always greater than 0 and less than 1. Hik reflects the basic quality of the identification audio itself; a higher Hik indicates higher quality.

5. The method for detecting fake audio based on multidimensional data analysis according to claim 4, characterized in that, The audio base tags include low-fidelity audio tags, medium-fidelity audio tags, and high-fidelity audio tags. The steps for generating the audio base tags are as follows: Set reference values ​​H1 and H2 for the audio impact coefficient, where H1 is less than 1 and greater than H2. The audio influence coefficient Hik is substituted into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than 0 and less than the audio influence coefficient reference value H2, a low-fidelity audio label is generated for the identified audio. Substitute the audio influence coefficient Hik into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than the audio influence coefficient reference value H2 and less than the audio influence coefficient reference value H1, generate a fidelity mark in the audio for the identified audio. Substitute the audio influence coefficient Hik into the audio influence coefficient reference values ​​H1 and H2 for comparison and analysis. When the audio influence coefficient Hik is greater than the audio influence coefficient reference value H1 and less than 1, generate an audio high-fidelity mark for the identified audio. Audio high-fidelity markers identify audio with higher audio quality and greater authenticity compared to audio medium-fidelity markers, and so on.

6. The method for detecting fake audio based on multidimensional data analysis according to claim 5, characterized in that, The output process for the relevant influence coefficient is as follows: The device temperature t1, real-time climate temperature t2, and real-time climate wind force value f are obtained. The difference between the real-time climate temperature t2 and the device temperature t1 is calculated and multiplied by the temperature error value q to obtain the first product. The correlation influence coefficient Tf is obtained by dividing the cube of the climate wind force value f by the absolute value of the square root of the first product. The correlation influence coefficient Tf is greater than 0. The larger the correlation influence coefficient Tf, the smaller the influence of the identification device on the identification audio.

7. The method for detecting fake audio based on multidimensional data analysis according to claim 6, characterized in that, The environmental impact level markers include high-level environmental impact markers and low-level environmental impact markers. The steps for generating the environmental impact level markers are as follows: Weighted analysis was performed on the correlation coefficient Tf and the audio influence coefficient Hik to obtain the coefficient comparison parameter Δ. The weight parameters corresponding to the correlation coefficient Tf and the audio influence coefficient Hik are G1 and G2, respectively, where G1+G2=1.5624. The larger the coefficient comparison parameter Δ, the greater the authenticity of the audio. Set the coefficient comparison threshold Wh, which is greater than 0. The coefficient comparison parameter Δ is substituted into the coefficient comparison threshold Wh for comparison analysis. When the coefficient comparison parameter Δ is greater than 0 and less than the coefficient comparison threshold Wh, it is marked as having a high degree of influence on the identification audio generation environment. The coefficient comparison parameter Δ is substituted into the coefficient comparison threshold Wh for comparison analysis. When the coefficient comparison parameter Δ is greater than the coefficient comparison threshold Wh, the influence of the identification audio generation environment is marked as low. Compared to markers with high environmental impact, markers with low environmental impact have less environmental impact in the audio samples being identified, and the identification equipment also has less influence on the identification process.

8. The method for detecting fake audio based on multidimensional data analysis according to claim 7, characterized in that, The forged result targets include real targets, suspicious targets, and forged targets. The steps for generating the forged result targets are as follows: The basic audio markers and environmental impact level markers in the identification audio are integrated and analyzed. If the identification audio has both high-fidelity audio markers and low-level environmental impact markers, a real target is generated for the identification audio. If the identified audio simultaneously possesses high-fidelity audio markers and high-level environmental influence markers, medium-fidelity audio markers and high-level environmental influence markers, or medium-fidelity audio markers and low-level environmental influence markers, a suspicious target is generated for the identified audio. If the identified audio simultaneously contains both low-fidelity audio markers and low-level environmental influence markers, or both low-fidelity audio markers and high-level environmental influence markers, a forgery target will be generated for that identified audio.

Images