Fake audio detection method for noisy scene

Abstract

The invention relates to the technical field of audio detection, and particularly discloses a forged audio detection method for a noisy scene, which comprises the following steps of: 1, carrying out detection analysis on an audio to obtain detection data; step 2, based on the detection data, calculating a forgery coefficient; step 3, based on the forgery coefficient, evaluating whether the audio is a forgery audio, and generating an audio evaluation signal; 4, analyzing the influence of the noise on the evaluation of the forged audio based on the audio forging early warning signal, and generating a correlation signal; 5, re-evaluating whether the audio is a forged audio or not based on the associated signal, and generating an audio secondary evaluation signal; according to the invention, the accuracy of detection is improved, and the stability and reliability of audio forging detection in different noise environments are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of audio detection, and in particular to a method for detecting forged audio in noisy scenes. Background Art

[0002] Audio is a technical term that describes the devices and functions related to sound within the audio range. All sounds that humans can hear can be called audio, which may include noise, etc. Audio can also refer to files that store sound content, or vibrations that act as filters.

[0003] However, with the continuous development of audio processing technology, the generation and dissemination of forged audio has become easier and easier, posing serious security risks to society. Traditional forged audio detection methods mainly rely on the analysis of audio waveform features, but in noisy scenarios, noise often interferes with audio waveform features, thereby affecting the accuracy of detection. In addition, when evaluating audio forgery, existing technologies cannot perform multi-dimensional evaluation based on the degree of abnormality of each sub-curve in the audio waveform, as well as the persistence and stability of these abnormalities in the entire audio, resulting in reduced accuracy of audio forgery. Summary of the invention

[0004] The purpose of the present invention is to provide a method for detecting forged audio in noisy scenes to solve the technical problems in the above background.

[0005] The purpose of the present invention can be achieved through the following technical solutions: In a first aspect, the present invention provides a method for detecting forged audio in a noisy scene, comprising the following steps: Step 1: Detect and analyze the audio to obtain detection data; Among them, the detection data includes: abnormal degree value and continuous abnormal value; Step 2: Calculate the counterfeiting coefficient based on the detection data; Step 3: Based on the forgery coefficient, evaluate whether the audio is forged audio and generate an audio evaluation signal; The audio evaluation signal includes: an audio forgery warning signal and an audio authenticity signal; Step 4: Based on the audio forgery warning signal, analyze the impact of noise on the forged audio evaluation and generate a correlation signal; Among them, the associated signals include: associated signals and unassociated signals; Step 5: Based on the relevant signal, re-evaluate whether the audio is forged audio and generate an audio secondary evaluation signal; The audio secondary evaluation signal includes: an audio forged signal and an audio real signal.

[0006] As a further solution of the present invention: the process of acquiring the detection data is as follows: Analyze the audio and obtain the audio waveform curve based on a two-dimensional coordinate system with time as the X-axis and amplitude as the Y-axis; Evenly divide the audio waveform curve to obtain a number of waveform sub-curves; The abnormal characterization value of each waveform sub-curve is calculated respectively, and based on the abnormal characterization value of each waveform sub-curve, the abnormal degree value and the continuous abnormal value are calculated respectively.

[0007] As a further solution of the present invention: the process of obtaining the abnormal characterization value is: Based on any waveform sub-curve, respectively obtain the amplitude sub-slope and the amplitude sub-mean of the waveform sub-curve; At the same time, the total amplitude slope and the total amplitude mean of the audio waveform curve are obtained respectively; The amplitude sub-slope and the amplitude total slope are calculated by difference, and the absolute value is taken to obtain the sub-slope difference, and then the sub-slope difference and the amplitude total slope are calculated by ratio to obtain the sub-slope difference ratio; The sub-mean of the amplitude is calculated with the total mean of the amplitude, and the absolute value is taken to obtain the sub-amplitude difference, and then the sub-amplitude difference is calculated with the total mean of the amplitude to obtain the sub-amplitude difference ratio; The sub-slope difference ratio and the sub-amplitude difference ratio are summed and calculated to obtain the abnormal characterization value.

[0008] As a further solution of the present invention: the process of obtaining the abnormality degree value is: Preset an abnormal characterization threshold, and compare and analyze the abnormal characterization values ​​of all waveform sub-curves with the abnormal characterization threshold; If the abnormal characterization value is less than or equal to the abnormal characterization threshold, a normal sub-curve signal is generated; If the abnormal characterization value is greater than the abnormal characterization threshold, an abnormal sub-curve signal is generated, and the waveform sub-curve is marked as an abnormal sub-curve; The abnormal characterization values ​​of all abnormal sub-curves are extracted, and the mean of all abnormal characterization values ​​is calculated to obtain the abnormal degree value.

[0009] As a further solution of the present invention: the process of obtaining the continuous abnormal value is: The abnormal characterization values ​​of all waveform sub-curves are extracted, and the variance of all abnormal characterization values ​​is calculated to obtain the continuous abnormal value.

[0010] As a further solution of the present invention: the process of obtaining the forged coefficient is: By formula: , calculate the forgery coefficient WZ, where a1 and a2 are preset proportional factors, both greater than 0, CD is the abnormality degree value, and CX is the continuous abnormality value.

[0011] As a further solution of the present invention: the process of generating the audio evaluation signal is: Preset a forgery coefficient threshold, and compare and analyze the forgery coefficient with the forgery coefficient threshold; If the forged coefficient is less than or equal to the forged coefficient threshold, a true audio signal is generated; If the forgery coefficient is greater than the forgery coefficient threshold, an audio forgery warning signal is generated.

[0012] As a further solution of the present invention: the process of generating the correlation signal is: Based on the audio forgery warning signal, the audio noise is detected and analyzed to obtain the abnormal noise period; Compare and analyze the abnormal noise period with the audio pre-forgery period, extract the overlapping period between the abnormal noise period and the audio pre-forgery period, and mark it as the overlapping period; The audio pre-forgery period is a portion of the audio period corresponding to all abnormal sub-curves; Calculate the difference between the audio pre-forgery period and the overlap period to obtain a correlation difference, and then calculate the ratio of the correlation difference to the audio pre-forgery period to obtain a correlation difference ratio; Preset a correlation difference ratio threshold, and compare and analyze the correlation difference ratio with the correlation difference ratio threshold; If the correlation difference ratio ≤ the correlation difference ratio threshold, a no-correlation signal is generated; If the correlation difference ratio>the correlation difference ratio threshold, a correlation signal is generated; The process of obtaining the abnormal noise period is as follows: Based on a two-dimensional coordinate system with time as the X-axis and noise decibels as the Y-axis, draw an audio noise curve and, at the same time, draw a noise limit range; Extract all time periods of the audio noise curve that are outside the limited range of the drawn noise to obtain the abnormal noise period.

[0013] As a further solution of the present invention: the process of generating the audio secondary evaluation signal is: Based on the relevant signal, the forged coefficient adjustment threshold is calculated, and then the forged coefficient is compared and analyzed with the forged coefficient adjustment threshold; If the forged coefficient is less than or equal to the forged coefficient adjustment threshold, a true audio signal is generated; If the forged coefficient is greater than the forged coefficient adjustment threshold, an audio forged signal is generated; Among them, the process of obtaining the forgery coefficient adjustment threshold is: The forged coefficient threshold is multiplied by the correlation difference ratio to obtain an adjustment value, and then the adjustment value is summed with the forged coefficient threshold to obtain the forged coefficient adjustment threshold.

[0014] In a second aspect, the present invention provides a forged audio detection system for noisy scenes, the system comprising: Data acquisition module: detect and analyze the audio and obtain detection data; Among them, the detection data includes: abnormal degree value and continuous abnormal value; Data processing module: calculate the counterfeiting coefficient based on the detection data; Audio evaluation module: based on the forgery coefficient, evaluates whether the audio is forged audio and generates an audio evaluation signal; The audio evaluation signal includes: an audio forgery warning signal and an audio authenticity signal; Correlation analysis module: Based on the audio forgery warning signal, it analyzes the impact of noise on the forged audio evaluation and generates correlation signals; Among them, the associated signals include: associated signals and unassociated signals; Secondary evaluation module: based on the relevant signal, re-evaluate whether the audio is fake audio and generate an audio secondary evaluation signal; The audio secondary evaluation signal includes: an audio forged signal and an audio real signal.

[0015] Beneficial effects of the present invention: (1) The present invention first analyzes the audio data to construct a two-dimensional waveform curve graph with time as the X-axis and amplitude as the Y-axis. On this basis, the audio waveform is evenly divided into multiple sub-curve segments to more carefully analyze the characteristics of each part; then, the abnormal characterization value of each waveform sub-curve is calculated, including obtaining the amplitude slope and amplitude mean of the sub-curve, and comparing them with the corresponding values ​​of the entire audio waveform curve. By calculating the sub-slope difference ratio and the sub-amplitude difference ratio and summing the two, the abnormal characterization value of each sub-curve is obtained. The value reflects the degree of difference between the sub-curve and the overall waveform. The higher the abnormal characterization value, the greater the possibility that the audio part is forged; after estimating the abnormal characterization values ​​of all sub-curves, the abnormal degree value and the continuous abnormal value of the audio are calculated. The abnormal degree value is obtained by comparing the abnormal representation values ​​of all abnormal sub-curves. The eigenvalue is obtained by taking the average value, which reflects the overall severity of the abnormality in the audio, while the continuous abnormality value is obtained by calculating the variance of all abnormal characteristic values, which reflects the stability of the speech speed and rhythm in the audio; finally, based on the above detection data, the forgery coefficient is calculated by a preset formula, which combines the information of the abnormality degree value and the continuous abnormality value, and is used to evaluate the possibility of audio forgery. By comparing with the preset forgery coefficient threshold, an audio evaluation signal can be generated, including an audio forgery warning signal and an audio true signal; the core of the present invention lies in multi-dimensional evaluation: not only the abnormality degree of each sub-curve in the audio waveform is analyzed, but also the persistence and stability of these abnormalities in the entire audio are considered, so as to construct a comprehensive forgery coefficient. The multi-dimensional evaluation method gives the method a high degree of automation, efficiency and accuracy; (2) The present invention firstly constructs a noise curve and a limited interval to accurately identify the abnormal noise period in the audio. Then, the abnormal noise period is compared with the pre-forgery period (i.e., the audio part corresponding to all abnormal sub-curves). By calculating the overlapping period, the correlation difference and the correlation difference ratio, the potential impact of noise on the forgery assessment is deeply analyzed. The preset correlation difference ratio threshold becomes the key indicator for judging the degree of noise influence. Based on this, an uncorrelated or correlated signal is generated, which provides an important reference for the subsequent secondary evaluation. In the secondary evaluation stage, the forgery coefficient threshold is dynamically adjusted according to the correlation signal. By calculating the adjustment value (the product of the forgery coefficient threshold and the correlation difference ratio) and summing it with the original threshold, a new forgery coefficient adjustment threshold is obtained. This adjustment mechanism makes the evaluation result closer to reality and effectively reduces the impact of noise interference on the final judgment. Finally, based on the adjusted forgery coefficient threshold, the technology performs a secondary evaluation on the audio and generates an audio real or forged signal. The present invention not only improves the accuracy of detection, but also ensures the stability and reliability of audio forgery detection under different noise environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the accompanying drawings.

[0017] Figure 1 It is a flowchart of the present invention; Figure 2 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0019] Embodiment 1: See also Figure 1 As shown, a method for detecting forged audio in a noisy scene according to an embodiment of the present invention includes the following steps: Step 1: Detect and analyze the audio to obtain detection data; Among them, the detection data includes: abnormal degree value and continuous abnormal value; In some implementation schemes, an audio analyzer is used to analyze the audio, and an audio waveform curve is obtained based on a two-dimensional coordinate system with time as the X-axis and amplitude as the Y-axis; It should be explained that the waveform of a real audio is usually continuous and smooth, while a forged or tampered audio may have mutations, discontinuities or abnormal features on the waveform. Evenly divide the audio waveform curve to obtain a number of waveform sub-curves; It should be explained that the number of waveform sub-curves is determined according to the total duration of the audio, and the number of waveform sub-curves should not be too small, so as to facilitate the subsequent analysis of the slope of each waveform sub-curve; Calculate the abnormality characterization value of each waveform sub-curve respectively, and calculate the abnormality degree value and the continuous abnormality value respectively based on the abnormality characterization value of each waveform sub-curve; First, specifically, the process of obtaining the abnormal characterization value is: Based on any waveform sub-curve, respectively obtain the amplitude sub-slope and the amplitude sub-mean of the waveform sub-curve; It should be explained that the amplitude sub-slope is the ratio of the amplitude end value to the amplitude start value of the waveform sub-curve, and the amplitude sub-mean is the amplitude average value of the waveform sub-curve; At the same time, the total amplitude slope and the total amplitude mean of the audio waveform curve are obtained respectively; The amplitude sub-slope and the amplitude total slope are calculated by difference, and the absolute value is taken to obtain the sub-slope difference, and then the sub-slope difference and the amplitude total slope are calculated by ratio to obtain the sub-slope difference ratio; The sub-mean of the amplitude is calculated with the total mean of the amplitude, and the absolute value is taken to obtain the sub-amplitude difference, and then the sub-amplitude difference is calculated with the total mean of the amplitude to obtain the sub-amplitude difference ratio; The sub-slope difference ratio and the sub-amplitude difference ratio are summed and calculated to obtain an abnormal characterization value; It should be explained that the abnormal characterization value reflects the degree of difference between the waveform sub-curve and the entire audio waveform curve; the larger the value of the abnormal characterization value, the greater the difference between the waveform sub-curve and the overall audio waveform curve, and the greater the possibility of forgery or tampering; Second, specifically, the process of obtaining the abnormality value is as follows: Preset an abnormal characterization threshold, and compare and analyze the abnormal characterization values ​​of all waveform sub-curves with the abnormal characterization threshold; If the abnormal characterization value is less than or equal to the abnormal characterization threshold, it means that the difference between the waveform sub-curve and the overall audio waveform curve is small, and it is determined that the possibility that the part of the audio corresponding to the waveform sub-curve is forged is small, that is, a normal sub-curve signal is generated; If the abnormal characterization value is greater than the abnormal characterization threshold, it means that the waveform sub-curve is significantly different from the overall audio waveform curve, and it is determined that the portion of audio corresponding to the waveform sub-curve is likely to be forged, that is, an abnormal sub-curve signal is generated, and the waveform sub-curve is marked as an abnormal sub-curve; Extract the abnormal characterization values ​​of all abnormal sub-curves, and calculate the mean of all abnormal characterization values ​​to obtain the abnormal degree value; It should be explained that the abnormality level value indicates the overall severity of the abnormality in the audio; the higher the abnormality level value, the more serious the abnormality of the audio, that is, the higher the possibility of forgery or tampering; Third, specifically, the process of obtaining the continuous outlier value is: Extract the abnormal characterization values ​​of all waveform sub-curves, and perform variance calculation on all abnormal characterization values ​​to obtain continuous abnormal values; It should be explained that the continuous outlier value reflects the stability of the speech speed and rhythm in the audio. In real audio, the speech speed and rhythm are usually natural and consistent, while in forged or tampered audio, the speech speed and rhythm may become unnatural or inconsistent due to splicing, editing or other tampering methods; if the continuous outlier value is high, it means that there are large differences between the abnormal representation values ​​of each waveform sub-curve, which means that the speech speed and rhythm in the audio are unstable, and the possibility of forgery or tampering is high; if the continuous outlier value is low, it means that the difference between the abnormal representation values ​​of each waveform sub-curve is small, the speech speed and rhythm of the audio are relatively stable, and the possibility of forgery or tampering is low; Step 2: Calculate the counterfeiting coefficient based on the detection data; In some embodiments, by the formula: , calculate the forgery coefficient WZ, where a1 and a2 are preset proportional factors, both greater than 0, CD is the abnormality degree value, and CX is the continuous abnormality value; Step 3: Based on the forgery coefficient, evaluate whether the audio is forged audio and generate an audio evaluation signal; The audio evaluation signal includes: an audio forgery warning signal and an audio authenticity signal; In some embodiments, a forgery coefficient threshold is preset, and the forgery coefficient is compared and analyzed with the forgery coefficient threshold; If the forgery coefficient is less than or equal to the forgery coefficient threshold, it means that the possibility that the audio is forged is low, that is, a true audio signal is generated; If the forgery coefficient is greater than the forgery coefficient threshold, it means that the audio is highly likely to be forged, and an audio forgery warning signal is generated; The technical solution of the embodiment of the present invention is mainly as follows: first, by parsing the audio data, a two-dimensional waveform curve graph with time as the X-axis and amplitude as the Y-axis is constructed. On this basis, the audio waveform is evenly divided into multiple sub-curve segments so as to analyze the characteristics of each part in more detail; then, the abnormal characterization value of each waveform sub-curve is calculated, including obtaining the amplitude slope and amplitude mean of the sub-curve, and comparing them with the corresponding values ​​of the entire audio waveform curve, and by calculating the sub-slope difference ratio and the sub-amplitude difference ratio, and summing the two, the abnormal characterization value of each sub-curve is obtained, which reflects the degree of difference between the sub-curve and the overall waveform. The higher the abnormal characterization value, the greater the possibility that the part of the audio is forged; after estimating the abnormal characterization values ​​of all sub-curves, the abnormal degree value and the continuous abnormal value of the audio are calculated. The abnormal degree value is obtained by comparing all abnormal sub-curves. The abnormal characterization value is obtained by taking the average value, which reflects the overall severity of the abnormality in the audio, while the continuous abnormality value is obtained by calculating the variance of all abnormal characterization values, which reflects the stability of the speech speed and rhythm in the audio; finally, based on the above detection data, the forgery coefficient is calculated by a preset formula, which combines the information of the abnormality degree value and the continuous abnormality value, and is used to evaluate the possibility of audio forgery. By comparing with the preset forgery coefficient threshold, an audio evaluation signal can be generated, including an audio forgery warning signal and an audio true signal; the core of the embodiment of the present invention lies in multi-dimensional evaluation: not only the abnormality degree of each sub-curve in the audio waveform is analyzed, but also the persistence and stability of these anomalies in the entire audio are considered, so as to construct a comprehensive forgery coefficient. The multi-dimensional evaluation method gives the method a high degree of automation, efficiency and accuracy.

[0020] Embodiment 2: Based on Example 1, please refer to Figure 1 As shown, the forged audio detection method for noisy scenes described in the embodiment of the present invention further includes the following steps: Step 4: Based on the audio forgery warning signal, analyze the impact of noise on the forged audio evaluation and generate a correlation signal; Among them, the associated signals include: associated signals and unassociated signals; In some implementation schemes, based on the audio forgery warning signal, the audio noise is detected and analyzed to obtain the abnormal noise period; Compare and analyze the abnormal noise period with the audio pre-forgery period, extract the overlapping period between the abnormal noise period and the audio pre-forgery period, and mark it as the overlapping period; The audio pre-forgery period is a portion of the audio period corresponding to all abnormal sub-curves; Calculate the difference between the audio pre-forgery period and the overlap period to obtain a correlation difference, and then calculate the ratio of the correlation difference to the audio pre-forgery period to obtain a correlation difference ratio; Preset a correlation difference ratio threshold, and compare and analyze the correlation difference ratio with the correlation difference ratio threshold; If the correlation difference ratio ≤ the correlation difference ratio threshold, it means that the noise has little influence on the forged audio evaluation, that is, an uncorrelated signal is generated; If the correlation difference ratio > the correlation difference ratio threshold, it means that the noise has a greater impact on the forged audio evaluation, that is, a correlation signal is generated; Specifically, the process of obtaining the abnormal noise period is as follows: Based on a two-dimensional coordinate system with time as the X-axis and noise decibels as the Y-axis, draw an audio noise curve and, at the same time, draw a noise limit range; It should be explained that the upper and lower limits of the noise limit interval are preset by those skilled in the art, and the interval represents the boundary between the normal noise level and the abnormal noise level; Extract all time periods of the audio noise curve that are outside the limited range of the drawn noise to obtain the abnormal noise period; Step 5: Based on the relevant signal, re-evaluate whether the audio is forged audio and generate an audio secondary evaluation signal; The audio secondary evaluation signal includes: an audio forged signal and an audio real signal; In some embodiments, a forged coefficient adjustment threshold is calculated based on the associated signal, and then the forged coefficient is compared and analyzed with the forged coefficient adjustment threshold; If the forgery coefficient is less than or equal to the forgery coefficient adjustment threshold, it means that after removing the influence of noise, the possibility of forgery is low, that is, a real audio signal is generated; If the forgery coefficient is greater than the forgery coefficient adjustment threshold, it means that the audio is highly likely to be forged after removing the influence of noise, that is, an audio forgery signal is generated; Specifically, the process of obtaining the forged coefficient adjustment threshold is: The forged coefficient threshold is multiplied by the correlation difference ratio to obtain an adjustment value, and then the adjustment value is summed with the forged coefficient threshold to obtain the forged coefficient adjustment threshold; The technical solution of the embodiment of the present invention is mainly as follows: first, by constructing a noise curve and a limited interval, the abnormal noise period in the audio is accurately identified; then, the abnormal noise period is compared with the pre-forgery period (i.e., the audio part corresponding to all abnormal sub-curves), and by calculating the overlapping period, the correlation difference and the correlation difference ratio, the potential impact of noise on the forgery assessment is deeply analyzed. The preset correlation difference ratio threshold becomes a key indicator for judging the degree of noise influence, and an unrelated or related signal is generated based on it, providing an important reference for the subsequent secondary evaluation; in the secondary evaluation stage, the forgery coefficient threshold is dynamically adjusted according to the related signal, and a new forgery coefficient adjustment threshold is obtained by calculating the adjustment value (the product of the forgery coefficient threshold and the correlation difference ratio) and summing it with the original threshold. This adjustment mechanism makes the evaluation result closer to reality and effectively reduces the impact of noise interference on the final judgment; finally, based on the adjusted forgery coefficient threshold, the technology performs a secondary evaluation on the audio to generate an audio real or forged signal; the embodiment of the present invention not only improves the accuracy of detection, but also ensures the stability and reliability of audio forgery detection under different noise environments.

[0021] Embodiment 3: Based on Example 1 and Example 2, please refer to Figure 2 As shown, a forged audio detection system for noisy scenes according to an embodiment of the present invention includes: Data acquisition module: detect and analyze the audio and obtain detection data; Among them, the detection data includes: abnormal degree value and continuous abnormal value; Data processing module: calculate the counterfeiting coefficient based on the detection data; Audio evaluation module: based on the forgery coefficient, evaluates whether the audio is forged audio and generates an audio evaluation signal; The audio evaluation signal includes: an audio forgery warning signal and an audio authenticity signal; Correlation analysis module: Based on the audio forgery warning signal, it analyzes the impact of noise on the forged audio evaluation and generates correlation signals; Among them, the associated signals include: associated signals and unassociated signals; Secondary evaluation module: based on the relevant signal, re-evaluate whether the audio is fake audio and generate an audio secondary evaluation signal; The audio secondary evaluation signal includes: an audio forged signal and an audio real signal.

[0022] The above thresholds are set for the convenience of comparison. The thresholds depend on the amount of sample data and the number of bases set by technicians in this field for each set of sample data. The above formulas are obtained by collecting a large amount of data and performing software simulation to select a formula close to the actual value. The factors in the formula are set by technicians in this field according to actual conditions; for example: Formula ; A technician in this field collects multiple groups of test data and sets a corresponding counterfeiting coefficient for each group of test data; Substitute the set counterfeiting coefficient and the collected test data into the formula, any two formulas constitute a binary linear equation system, screen the calculated factors and take the average, and obtain the values ​​of a1 and a2 as 2.65 and 2.23 respectively; The size of the factor is to quantify each parameter to obtain a specific value for subsequent comparison. The size of the factor depends on the amount of test data and the technical personnel in this field preliminarily setting the corresponding falsification coefficient for each set of test data; as long as it does not affect the proportional relationship between the parameter and the quantified value, such as the falsification coefficient is proportional to the value of the continuous abnormal value.

[0023] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for detecting forged audio in noisy scenes, characterized in that: The following steps are involved: Step 1: Detect and analyze the audio to obtain detection data; Among them, the detection data includes: abnormal degree value and continuous abnormal value; Step 2: Calculate the counterfeiting coefficient based on the detection data; Step 3: Based on the forgery coefficient, evaluate whether the audio is forged audio and generate an audio evaluation signal; The audio evaluation signal includes: an audio forgery warning signal and an audio authenticity signal; Step 4: Based on the audio forgery warning signal, analyze the impact of noise on the forged audio evaluation and generate a correlation signal; Among them, the associated signals include: associated signals and unassociated signals; Step 5: Based on the relevant signal, re-evaluate whether the audio is forged audio and generate an audio secondary evaluation signal; The audio secondary evaluation signal includes: an audio forged signal and an audio real signal.

2. A method for detecting forged audio in noisy scenes according to claim 1, characterized in that: The process of acquiring detection data is as follows: Analyze the audio and obtain the audio waveform curve based on a two-dimensional coordinate system with time as the X-axis and amplitude as the Y-axis; Evenly divide the audio waveform curve to obtain a number of waveform sub-curves; The abnormal characterization value of each waveform sub-curve is calculated respectively, and based on the abnormal characterization value of each waveform sub-curve, the abnormal degree value and the continuous abnormal value are calculated respectively.

3. A method for detecting forged audio in noisy scenes according to claim 2, characterized in that: The process of obtaining the abnormal characterization value is as follows: Based on any waveform sub-curve, respectively obtain the amplitude sub-slope and the amplitude sub-mean of the waveform sub-curve; At the same time, the total amplitude slope and the total amplitude mean of the audio waveform curve are obtained respectively; The amplitude sub-slope and the amplitude total slope are calculated by difference, and the absolute value is taken to obtain the sub-slope difference, and then the sub-slope difference and the amplitude total slope are calculated by ratio to obtain the sub-slope difference ratio; The sub-mean of the amplitude is calculated with the total mean of the amplitude, and the absolute value is taken to obtain the sub-amplitude difference, and then the sub-amplitude difference is calculated with the total mean of the amplitude to obtain the sub-amplitude difference ratio; The sub-slope difference ratio and the sub-amplitude difference ratio are summed and calculated to obtain the abnormal characterization value.

4. The method for detecting forged audio in noisy scenes according to claim 3, characterized in that: The process of obtaining the abnormality value is as follows: Preset an abnormal characterization threshold, and compare and analyze the abnormal characterization values ​​of all waveform sub-curves with the abnormal characterization threshold; If the abnormal characterization value is less than or equal to the abnormal characterization threshold, a normal sub-curve signal is generated; If the abnormal characterization value is greater than the abnormal characterization threshold, an abnormal sub-curve signal is generated, and the waveform sub-curve is marked as an abnormal sub-curve; The abnormal characterization values ​​of all abnormal sub-curves are extracted, and the mean of all abnormal characterization values ​​is calculated to obtain the abnormal degree value.

5. The method for detecting forged audio in noisy scenes according to claim 3, characterized in that: The process of obtaining continuous outliers is as follows: The abnormal characterization values ​​of all waveform sub-curves are extracted, and the variance of all abnormal characterization values ​​is calculated to obtain the continuous abnormal value.

6. The method for detecting forged audio in noisy scenes according to claim 1, characterized in that: The process of obtaining the forged coefficient is: By formula: , calculate the forgery coefficient WZ, where a1 and a2 are preset proportional factors, both greater than 0, CD is the abnormality degree value, and CX is the continuous abnormality value.

7. The method for detecting forged audio in noisy scenes according to claim 1, characterized in that: The process of generating the audio evaluation signal is: Preset a forgery coefficient threshold, and compare and analyze the forgery coefficient with the forgery coefficient threshold; If the forged coefficient is less than or equal to the forged coefficient threshold, a true audio signal is generated; If the forgery coefficient is greater than the forgery coefficient threshold, an audio forgery warning signal is generated.

8. The method for detecting forged audio in noisy scenes according to claim 1, characterized in that: The process of generating the correlation signal is: Based on the audio forgery warning signal, the audio noise is detected and analyzed to obtain the abnormal noise period; Compare and analyze the abnormal noise period with the audio pre-forgery period, extract the overlapping period between the abnormal noise period and the audio pre-forgery period, and mark it as the overlapping period; The audio pre-forgery period is a portion of the audio period corresponding to all abnormal sub-curves; Calculate the difference between the audio pre-forgery period and the overlap period to obtain a correlation difference, and then calculate the ratio of the correlation difference to the audio pre-forgery period to obtain a correlation difference ratio; Preset a correlation difference ratio threshold, and compare and analyze the correlation difference ratio with the correlation difference ratio threshold; If the correlation difference ratio ≤ the correlation difference ratio threshold, a no-correlation signal is generated; If the correlation difference ratio>the correlation difference ratio threshold, a correlation signal is generated; The process of obtaining the abnormal noise period is as follows: Based on a two-dimensional coordinate system with time as the X-axis and noise decibels as the Y-axis, draw an audio noise curve and, at the same time, draw a noise limit range; Extract all time periods of the audio noise curve that are outside the limited range of the drawn noise to obtain the abnormal noise period.

9. The method for detecting forged audio in noisy scenes according to claim 1, characterized in that: The process of generating the audio secondary evaluation signal is: Based on the relevant signal, the forged coefficient adjustment threshold is calculated, and then the forged coefficient is compared and analyzed with the forged coefficient adjustment threshold; If the forged coefficient is less than or equal to the forged coefficient adjustment threshold, a true audio signal is generated; If the forged coefficient is greater than the forged coefficient adjustment threshold, an audio forged signal is generated; Among them, the process of obtaining the forgery coefficient adjustment threshold is: The forged coefficient threshold is multiplied by the correlation difference ratio to obtain an adjustment value, and then the adjustment value is summed with the forged coefficient threshold to obtain the forged coefficient adjustment threshold.

10. A forged audio detection system for noisy scenes, characterized in that: The system is used to execute the method described in any one of claims 1 to 9, and the system comprises: Data acquisition module: detect and analyze the audio and obtain detection data; Among them, the detection data includes: abnormal degree value and continuous abnormal value; Data processing module: calculate the counterfeiting coefficient based on the detection data; Audio evaluation module: based on the forgery coefficient, evaluates whether the audio is forged audio and generates an audio evaluation signal; The audio evaluation signal includes: an audio forgery warning signal and an audio authenticity signal; Correlation analysis module: Based on the audio forgery warning signal, it analyzes the impact of noise on the forged audio evaluation and generates correlation signals; Among them, the associated signals include: associated signals and unassociated signals; Secondary evaluation module: based on the relevant signal, re-evaluate whether the audio is fake audio and generate an audio secondary evaluation signal; The audio secondary evaluation signal includes: an audio forged signal and an audio real signal.

Images