Audio data desensitization method, device and system and storage medium

Abstract

The invention provides an audio data desensitization method, device and system and a storage medium, relates to the technical field of data processing, and can improve the accuracy of sensitive information identification. The method comprises the following steps: acquiring to-be-desensitized audio data, wherein the to-be-desensitized audio data comprises audio data of a plurality of objects; performing audio separation processing on the to-be-desensitized audio data to obtain audio data of each object; for the audio data of each object, carrying out sensitive information detection on the audio data of the object to obtain target sensitive data corresponding to the object so as to obtain all target sensitive data; and desensitizing the to-be-desensitized audio data based on all the target sensitive data to obtain desensitized audio data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to an audio data desensitization method, device, system and storage medium. BACKGROUND

[0002] With the rapid development of the data industry, sensitive data desensitization technology is applied more and more widely. At present, the traditional audio sensitive data desensitization technical solution mainly identifies sensitive information in audio data through voice recognition and keyword matching methods, and processes the identified sensitive information through sound elimination.

[0003] As described above in the audio sensitive data desensitization technical solution, due to the high complexity of the audio source scene (for example, multiple people speaking at the same time, environmental noise, speakers using dialects, and network slang, etc.), the traditional audio sensitive data desensitization technical solution has the problem of low accuracy of sensitive information identification. SUMMARY

[0004] The present application provides an audio data desensitization method, device, system and storage medium, which can improve the accuracy of sensitive information identification.

[0005] To achieve the above purpose, the present application adopts the following technical solution:

[0006] In a first aspect, the present application provides an audio data desensitization method, which comprises: obtaining to-be-desensitized audio data, the to-be-desensitized audio data comprising audio data of multiple objects; performing audio separation processing on the to-be-desensitized audio data to obtain audio data of each object; for the audio data of each object, performing sensitive information detection on the audio data of the object to obtain target sensitive data corresponding to the object, so as to obtain all target sensitive data; and performing desensitization processing on the to-be-desensitized audio data based on all target sensitive data to obtain desensitized audio data.

[0007] The above technical solution at least brings the following beneficial effects: the audio data desensitization method provided by the present application preprocesses the to-be-detected audio file to obtain to-be-desensitized audio data. Compared with the traditional method of directly using the original audio file for sensitive information detection, the present application first preprocesses the audio file to convert the audio data of multiple speaking objects into audio data of multiple single speaking objects, which eliminates the interference of overlapping voices and improves the accuracy of sensitive information detection. In addition, the present application also performs sensitive information detection on each single speaking object audio data obtained one by one to obtain more accurate sensitive information, further improving the accuracy of sensitive information detection.

[0008] In a possible implementation, the audio data of the object is subjected to speech recognition processing to obtain text data corresponding to the object; the text data corresponding to the object is subjected to sensitive information detection by using a target detection manner to obtain target sensitive data corresponding to the object, the target detection manner comprising: a first detection manner and / or a second detection manner, the first detection manner being used for detecting a first type of to-be-detected word in the text data based on a preset matching rule, and the second detection manner being used for detecting based on an association between the to-be-detected word and a second type of to-be-detected word in the text data; wherein the sensitivity of the first type of to-be-detected word is greater than the sensitivity of the second type of to-be-detected word.

[0009] In a possible implementation, the method further comprises: determining a current scene based on an input manner of the to-be-desensitized audio data; in a case where the current scene is a real-time stream processing scene, determining that the target detection manner comprises the first detection manner; in a case where the current scene is an offline batch processing scene, determining that the target detection manner comprises the first detection manner and the second detection manner.

[0010] In a possible implementation, the method further comprises: obtaining a current resource utilization rate of the processor; in a case where the current resource utilization rate is greater than a resource utilization rate threshold, switching the detection model in the second detection manner to a first mode; in a case where the current resource utilization rate is greater than the resource utilization rate threshold, switching the detection model in the second detection manner to a second mode; the resource occupied by the first mode is less than the resource occupied by the second mode.

[0011] In a possible implementation, the desensitization processing is performed on the to-be-desensitized audio data based on all the target sensitive data to obtain desensitized audio data, comprising: for each target sensitive data, performing timestamp positioning processing on the target sensitive data to obtain a sensitive data position corresponding to the target sensitive data, the sensitive data position being a position of the target sensitive data in the to-be-desensitized audio data; performing desensitization processing on the to-be-desensitized audio data based on the sensitive data position corresponding to each target sensitive data to obtain the desensitized audio data.

[0012] In a possible implementation, the desensitization processing is performed on the to-be-desensitized audio data based on the sensitive data position corresponding to each target sensitive data to obtain desensitized audio data, comprising: performing desensitization processing on the to-be-desensitized audio data based on the sensitive data position corresponding to each target sensitive data and a desensitization strategy to obtain the desensitized audio data; the desensitization strategy comprises at least one of the following: when the sensitivity of the sensitive level of the target sensitive data is greater than or equal to a sensitivity threshold, performing white noise covering processing and spectrum confusion processing on the target sensitive data; in a case where the sensitivity of the target sensitive data is less than or equal to the sensitivity threshold, performing mute replacement processing on the target sensitive information.

[0013] In a possible implementation, the method further includes: obtaining a signal-to-noise ratio of the to-be-desensitized audio data; performing noise reduction processing on the to-be-desensitized audio data based on Wiener filtering in a case where the signal-to-noise ratio is greater than a second signal-to-noise ratio threshold; and performing noise reduction processing on the to-be-desensitized audio data based on frequency domain spectral subtraction and wavelet thresholding in a case where the signal-to-noise ratio is less than or equal to the second signal-to-noise ratio threshold.

[0014] In a second aspect, the present application provides an audio data desensitization device, which includes: a communication unit and a processing unit; the communication unit is configured to obtain to-be-desensitized audio data, the to-be-desensitized audio data including audio data of a plurality of objects; the processing unit is configured to perform audio separation processing on the to-be-desensitized audio data to obtain audio data of each object; the processing unit is further configured to, for the audio data of each object, perform sensitive information detection on the audio data of the object to obtain target sensitive data corresponding to the object, so as to obtain all target sensitive data; and the processing unit is further configured to perform desensitization processing on the to-be-desensitized audio data based on all target sensitive data to obtain desensitized audio data.

[0015] In a possible implementation, the processing unit is specifically configured to: perform speech recognition processing on the audio data of the object to obtain text data corresponding to the object; and perform sensitive information detection on the text data corresponding to the object by using a target detection manner, to obtain target sensitive data corresponding to the object, the target detection manner including: a first detection manner and / or a second detection manner, the first detection manner being configured to detect a first type of to-be-detected word in the text data based on a preset matching rule, and the second detection manner being configured to detect based on associated content of the to-be-detected word and a second type of to-be-detected word in the text data; and wherein the sensitivity of the first type of to-be-detected word is greater than the sensitivity of the second type of to-be-detected word.

[0016] In a possible implementation, the processing unit is specifically configured to: determine a current scene based on an input manner of the to-be-desensitized audio data; determine that the target detection manner includes the first detection manner in a case where the current scene is a real-time stream processing scene; and determine that the target detection manner includes the first detection manner and the second detection manner in a case where the current scene is an offline batch processing scene.

[0017] In a possible implementation, the communication unit is further configured to obtain a current resource utilization rate of the processor; the processing unit is further configured to switch a detection model in the second detection manner to the first mode in a case where the current resource utilization rate is greater than a resource utilization rate threshold; and the processing unit is further configured to switch the detection model in the second detection manner to the second mode in a case where the current resource utilization rate is greater than the resource utilization rate threshold.

[0018] In a possible implementation, the determining unit is specifically configured to: for each target sensitive data, performing timestamp positioning processing on the target sensitive data to obtain a sensitive data position corresponding to the target sensitive data, the sensitive data position being a position of the target sensitive data in the to-be-desensitized audio data; and performing desensitization processing on the to-be-desensitized audio data based on the sensitive data position corresponding to each target sensitive data to obtain the desensitized audio data.

[0019] In a possible implementation, the determining unit is specifically configured to: perform desensitization processing on the to-be-desensitized audio data based on the sensitive data position corresponding to each target sensitive data and a desensitization strategy to obtain the desensitized audio data; and the desensitization strategy includes at least one of the following: performing white noise covering processing and spectrum confusion processing on the target sensitive data in a case where a sensitivity of the target sensitive data is greater than or equal to a sensitivity threshold; and performing mute replacing processing on the target sensitive information in a case where the sensitivity of the target sensitive data is less than or equal to the sensitivity threshold.

[0020] In a possible implementation, the communication unit is further configured to obtain a signal-to-noise ratio of the to-be-desensitized audio data; and the determining unit is further configured to: in a case where the signal-to-noise ratio is greater than a signal-to-noise ratio threshold, perform noise reduction processing on the to-be-desensitized audio data based on Wiener filtering; and in a case where the signal-to-noise ratio is less than or equal to the signal-to-noise ratio threshold, perform noise reduction processing on the to-be-desensitized audio data based on frequency domain spectral subtraction and wavelet threshold.

[0021] In a third aspect, the present application provides an audio data desensitization system, comprising: an audio preprocessing module, configured to process to-be-desensitized audio data, remove abnormal data in the to-be-desensitized audio data, and obtain first audio data, wherein the abnormal data comprises at least one of the following: damaged data, repeated data, invalid data, and noise data; a multi-person overlapping speech detection and separation cutting module, configured to perform audio separation processing on the first audio data to obtain audio data of each object, wherein the audio separation processing comprises multi-object audio data position identification and multi-object audio data separation; a speech recognition module, configured to perform speech recognition processing on the audio data of the object to obtain text data corresponding to the object; a sensitive information classification and grading identification module, configured to perform sensitive information detection on the text data corresponding to the object in a target detection manner to obtain target sensitive data corresponding to the object; a sensitive data timestamp positioning processing module, configured to, for each target sensitive data, perform timestamp positioning processing on the target sensitive data to obtain a sensitive data position corresponding to the target sensitive data, wherein the sensitive data position is a position of the target sensitive data in the to-be-desensitized audio data; a sound elimination and desensitization processing module, configured to perform desensitization processing on the to-be-desensitized audio data based on all the target sensitive data to obtain desensitized audio data, and further configured to save the desensitized audio data and corresponding desensitized text data; and a distributed processing base, configured to optimize the audio data desensitization system based on a desensitization process and a desensitization result of the target audio data.

[0022] In a fourth aspect, the present application provides an audio data desensitization device, comprising: a processor and a communication interface; the communication interface is coupled with the processor, and the processor is configured to run a computer program or instructions to implement the audio data desensitization method as described in the first aspect and any possible implementation manner of the first aspect.

[0023] In a fifth aspect, the present application provides a computer readable storage medium, which stores instructions, and when the instructions run on a terminal, the terminal executes the audio data desensitization method as described in the first aspect and any possible implementation manner of the first aspect.

[0024] In a sixth aspect, the present application provides a computer program product comprising instructions, and when the computer program product runs on an audio data desensitization device, the audio data desensitization device executes the audio data desensitization method as described in the first aspect and any possible implementation manner of the first aspect.

[0025] In a seventh aspect, the present application provides a chip, comprising a processor and a communication interface, the communication interface is coupled with the processor, and the processor is configured to run a computer program or instructions to implement the audio data desensitization method as described in the first aspect and any possible implementation manner of the first aspect.

[0026] Specifically, the chip provided in the present application further comprises a memory for storing a computer program or instructions. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 A structural schematic diagram of an audio data desensitization system provided by an embodiment of the present application;

[0028] Figure 2 A structural schematic diagram of an audio data desensitization device provided by an embodiment of the present application;

[0029] Figure 3 A structural schematic diagram of another audio data desensitization system provided by an embodiment of the present application;

[0030] Figure 4 A flowchart of an audio data desensitization method provided by an embodiment of the present application;

[0031] Figure 5 A flowchart of a method for separating multi-object audio data provided by an embodiment of the present application;

[0032] Figure 6 A flowchart of another audio data desensitization method provided by an embodiment of the present application;

[0033] Figure 7 A flowchart of another audio data desensitization method provided by an embodiment of the present application;

[0034] Figure 8 A flowchart of another audio data desensitization method provided by an embodiment of the present application;

[0035] Figure 9 A structural schematic diagram of another audio data desensitization device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0036] The audio data desensitization method, device, system and storage medium provided by the embodiments of the present application are described in detail below in combination with the drawings.

[0037] The term “and / or” in the present document is merely used to describe an association relationship of associated objects, and indicates that there can be three relationships, for example, A and / or B can represent three cases of existence of A alone, existence of A and B simultaneously, and existence of B alone.

[0038] The terms “first” and “second” and the like in the present specification and the drawings are used to distinguish different objects, or to distinguish different treatments of the same object, and are not used to describe a specific order of the objects.

[0039] Moreover, the terms "comprising" and "having" and any variations thereof in the description and in the claims are intended to cover both the inclusive and exclusive aspects thereof. For example, a process, method, system, product, or apparatus that comprises a list of steps or elements is not necessarily limited to only those steps or elements but can include other steps or elements not expressly listed or inherent to such process, method, system, product, or apparatus.

[0040] It should be noted that the terms "exemplary" and "for example" are used herein to mean "an example of" or "one example among others." Any embodiment or design scheme described as "exemplary" or "for example" in the present application is not necessarily to be construed as preferred or advantageous over other embodiments or design schemes. Rather, use of the terms "exemplary" or "for example" is intended to present concepts in a particular manner.

[0041] In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0042] With the rapid development of the data industry, sensitive data desensitization technology is applied more and more widely. At present, the traditional audio sensitive data desensitization technical scheme mainly identifies sensitive information in audio data through voice recognition and keyword matching methods, and processes the identified sensitive information through sound elimination.

[0043] At present, some audio desensitization methods propose to obtain audio data in the process of vehicle driving, and automatically identify audio data segments containing sensitive information through voice recognition or keyword detection algorithm, and then perform sound elimination processing on the audio data segments. The scene applicable to this audio desensitization method is relatively simple, and when facing a scene with high complexity, this audio desensitization method has the problem of low accuracy of sensitive information recognition.

[0044] As described in the foregoing audio sensitive data desensitization technical scheme, due to the high complexity of the audio source scene (for example, multiple people speaking at the same time, environmental noise, speakers using dialects, and network slang, etc.), the traditional audio sensitive data desensitization technical scheme has the problem of low accuracy of sensitive information recognition.

[0045] In view of this, the audio data desensitization method provided by the application is used to preprocess the audio file to be detected, and the audio data to be desensitized is obtained. Compared with the traditional method of directly using the original audio file for sensitive information detection, the audio file is preprocessed in the application to convert the audio data of multiple speaking objects into audio data of multiple single speaking objects, which eliminates the interference of overlapping voices and improves the accuracy of sensitive information detection. In addition, the application also detects the sensitive information of each single speaking object audio data one by one to obtain more accurate sensitive information, further improving the accuracy of sensitive information detection.

[0046] The technical scheme provided by the embodiments of the application can be applied to various communication systems, for example, a new radio (NR) communication system using the fifth generation mobile communication technology (5th generation mobile communication technology, 5G), a future evolution system, or a variety of communication fusion systems, etc.

[0047] Exemplarily, Figure 1 A structural schematic diagram of an audio data desensitization system 10 provided by the embodiments of the application is shown in FIG. 1. The audio data desensitization system 10 can include at least one end device 101 and at least one computing device 102, and the end device 101 can be in communication connection with the computing device 102. Figure 1 In FIG. 1, only one end device 101 and one computing device 102 are shown.

[0048] The end device 101 is configured to generate audio data to be desensitized, the audio data to be desensitized including audio data of multiple objects, and send the audio data to be desensitized to the computing device 102.

[0049] Optionally, the computing device 102 and the end device 101 can be independent different devices, and can also be the same device, which is not limited in the application.

[0050] The computing device 102 is configured to perform audio separation processing on the audio data to be desensitized to obtain audio data of each object; for the audio data of each object, perform sensitive information detection on the audio data of the object to obtain target sensitive data corresponding to the object, so as to obtain all target sensitive data; and perform desensitization processing on the audio data to be desensitized based on all target sensitive data to obtain desensitized audio data.

[0051] Optionally, the computing device 102 can be a large-capacity device (for example, a server integrating a large number of boards) with wireless communication function, which can be deployed on land, including indoor or outdoor, handheld or vehicle-mounted. It can also be deployed on the water surface (such as ships, etc.). It can also be deployed in the air (such as airplanes, balloons and satellites, etc.).

[0052] Alternatively, the end device 101 may be a device with wireless transceiver functionality or a chip or chip system that can be provided in the device. The end device 101 includes, but is not limited to, a small base station, a wireless access point, a transmission receive point (TRP), a transmission point (TP), a macro base station, a relay base station, and any other access node.

[0053] It should be noted that Figure 1 This is just an illustrative framework diagram. Figure 1 The number of nodes included in the name of each device is not limited, and Figure 1 In addition to the functional nodes shown, the audio data desensitization system 10 may also include other nodes, such as core network equipment, and this application does not impose any restrictions on this.

[0054] The application scenarios of the embodiments of this application are not limited. The system architecture and business scenarios described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. It is known to those skilled in the art that with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0055] When implementing it specifically, Figure 1 All devices in the Figure 2 The structure shown, or including Figure 2 Parts shown. Figure 2 This is a schematic diagram of the composition of an audio data desensitization device 20 provided in an embodiment of the present application. The audio data desensitization device 20 can be a computing device 102 or a chip or system on chip in the computing device 102. Alternatively, the audio data desensitization device 20 can be a terminal device 101 or a chip or system on chip in the terminal device 101. Figure 2 As shown, the audio data desensitization device 20 may include a processor 201 and a bus 202.

[0056] Furthermore, the audio data desensitization device 20 may further include a communication interface 203 and a memory 204. The processor 201, the memory 204 and the communication interface 203 may be connected via a bus 202.

[0057] The processor 201 can be a CPU, a general-purpose processor, a network processor (NP), a digital signal processor (DSP), a microprocessor, a microcontroller, a programmable logic device (PLD), or any combination thereof. The processor 201 can also be other devices with processing capability, such as a circuit, a device, or a software module, without limitation.

[0058] The bus 202 is used to transmit information between components included in the audio data desensitization apparatus 20.

[0059] The communication interface 203 is used to communicate with other devices or other communication networks. The other communication networks can be an Ethernet, a radio access network (RAN), a wireless local area network (WLAN), or the like. The communication interface 203 can be a module, a circuit, a communication interface, or any device capable of communication.

[0060] The memory 204 is used to store instructions. The instructions can be a computer program.

[0061] The memory 204 can be a read-only memory (ROM) or other type of static storage device that can store static information and / or instructions, or a random access memory (RAM) or other type of dynamic storage device that can store information and / or instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disk storage, a magneto-optical disk, a magnetic disk storage or other magnetic storage devices, and the like, without limitation.

[0062] It should be noted that the memory 204 can exist independently of the processor 201, or can be integrated with the processor 201. The memory 204 can be used to store instructions or program codes or some data, and the like. The memory 204 can be located in the audio data desensitization apparatus 20, or can be located outside the audio data desensitization apparatus 20, without limitation.

[0063] In an example, the processor 201 can include one or more CPUs, for example,Figure 2 CPU0 and CPU1 (not shown in the figure) in the computer.

[0064] As an optional implementation, the audio data desensitization apparatus 20 comprises a plurality of processors.

[0065] As an optional implementation, the audio data desensitization apparatus 20 further comprises an output device 205 and an input device 206. Exemplarily, the input device 206 is a keyboard, a mouse, a microphone, a joystick or the like, and the output device 205 is a display screen, a speaker or the like.

[0066] It should be noted that the audio data desensitization apparatus 20 can be a desktop computer, a portable computer, a network server, a mobile phone, a tablet computer, a wireless terminal, an embedded device, a chip system or a device having a similar structure as shown in the computer. Figure 2 Moreover, Figure 2 The constituent structures shown in the computer do not constitute a limitation on the computer, and in addition to the components shown in the computer, Figure 1 and Figure 2 The components shown in the computer can comprise more or fewer components than shown in the computer, or some components can be combined, or different components can be arranged. Figure 2 Figure 1 and Figure 2 The components shown in the computer can comprise more or fewer components than shown in the computer, or some components can be combined, or different components can be arranged.

[0067] Exemplarily, Figure 3 Another structure of an audio data desensitization system provided by the embodiment of the present application is shown in the figure, which comprises an audio preprocessing module 301, a multi-person overlapping speech detection and separation cutting module 302, a speech recognition module 303, a sensitive information recognition module 304, a timestamp positioning processing module 305, an audio data desensitization module 306, and a distributed processing base 307.

[0068] The audio preprocessing module 301 is configured to process the audio data to be desensitized, remove abnormal data in the audio data to be desensitized, and obtain first audio data; wherein the abnormal data comprises at least one of the following: damaged data, repeated data, and noise data.

[0069] The multi-person overlapping speech detection and separation cutting module 302 is configured to perform audio separation processing on the first audio data to obtain audio data of each object, wherein the audio separation processing process comprises a plurality of object audio data position recognitions and a plurality of object audio data separations.

[0070] The speech recognition module 303 is configured to perform speech recognition processing on the audio data of the object to obtain text data corresponding to the object, wherein the speech recognition process comprises feature extraction, similarity calculation, object clustering, and conversion of positioning points. ​

[0071] The sensitive information identification module 304 is configured to detect sensitive information in the text data corresponding to the object by using a target detection manner, to obtain target sensitive data corresponding to the object, and the obtained target sensitive data can be target sensitive data including sensitive information such as abuse, discrimination, and violence, or target sensitive data including private information such as a phone number, an email address, an ID number, a name, and an address.

[0072] The timestamp positioning processing module 305 is configured to, for each target sensitive data, perform timestamp positioning processing on the target sensitive data, to obtain a sensitive data position corresponding to the target sensitive data, where the sensitive data position is a position of the target sensitive data in the to-be-desensitized audio data.

[0073] The audio data desensitization module 306 is configured to perform desensitization processing on the to-be-desensitized audio data based on all the target sensitive data, to obtain desensitized audio data, and is further configured to save the obtained desensitized audio data and corresponding desensitized text data. The desensitized audio data and the corresponding desensitized text data jointly constitute high-quality multi-modal data.

[0074] The distributed processing base 307 is configured to optimize the audio data desensitization system based on a desensitization process and a desensitization result of the target audio data, and the process can be implemented by reinforcement learning. In addition, the distributed processing base 307 further includes a distributed framework configured to analyze and calculate data generated by the desensitization process and the desensitization result, and determine an optimization strategy based on an analysis and calculation result.

[0075] In the embodiments of the application, the chip system can be composed of a chip, or can include a chip and other discrete devices.

[0076] In addition, the actions, terms, and the like involved among the embodiments of the application can be mutually referenced and are not limited. The message name or parameter name in the message exchanged between the devices in the embodiments of the application is only an example, and other names can also be used in specific implementation, which is not limited.

[0077] The audio data desensitization method provided by the embodiments of the application will be described below in combination with the cross-border network system shown in Figure 1 The actions, terms, and the like involved among the embodiments of the application can be mutually referenced and are not limited. The message name or parameter name in the message exchanged between the devices in the embodiments of the application is only an example, and other names can also be used in specific implementation, which is not limited. The actions involved in the embodiments of the application are only an example, and other names can also be used in specific implementation, such as: “include” in the embodiments of the application can be replaced by “carry” or “carry” and the like.

[0078] As Figure 4As shown, an audio data desensitization method provided by an embodiment of the present application includes the following steps.

[0079] S401, obtaining audio data to be desensitized.

[0080] For example, the audio data to be desensitized can be the communication recording of the customer service personnel and the customer of an enterprise providing online customer service, can also be the communication recording of the intelligent voice robot and the customer, can also be the audio data recorded by the customer himself / herself, and can also be the audio data publicly available on the Internet, and the present application does not limit this.

[0081] Further, optionally, the manner of obtaining the audio data to be desensitized can be to directly read the pre-input audio data to be desensitized from the audio data desensitization system, can also be to obtain the audio data to be desensitized input by other devices or systems in real time, and can also be to collect the audio data to be desensitized uploaded on the Internet, and the present application does not limit this.

[0082] S402, performing audio separation processing on the audio data to be desensitized to obtain audio data of each object.

[0083] In a possible implementation manner, as shown in Figure 5 As shown in a flowchart of a method for separating multi-object audio data provided by an embodiment of the present application, the implementation process of performing audio separation processing on the audio data to be desensitized can include overlap region identification, multi-object simultaneous speech position separation, semantic reorganization, and timestamp restoration, and the specific process is as follows: overlap region identification: based on double-threshold energy detection and long short-term memory (LSTM) time series prediction model, the audio data to be desensitized is detected to determine the audio data position of the multi-object simultaneous speech. Multi-object simultaneous speech position separation: the audio data position of the multi-object simultaneous speech is separated to obtain the audio segment of the multi-object simultaneous speech. The audio segment is input into a voiceprint recognition model (context-aware masking, CAM++), which identifies the audio segment through the context-aware masking feature enhancement manner and separates different objects in the audio through the speaker embedding vector extraction module. Semantic reorganization: detecting the audio data of each object, obtaining the audio data with incomplete semantics, and based on a Chinese pre-training language (robustly optimized bert approach, RoBERTa) model, the audio data is completed to obtain the audio data with complete semantics. Time stamp restoration: supplementing the separated audio data to the corresponding position of the audio data to be desensitized.

[0084] In a possible implementation, before the audio separation processing of the to-be-desensitized audio data, the damaged to-be-desensitized audio data can also be removed, and the specific process can be: the format of the to-be-desensitized audio data and the encoding of the to-be-desensitized audio data can be detected, and the to-be-desensitized audio data whose format or encoding is abnormal or damaged can be removed.

[0085] In a possible implementation, the to-be-desensitized audio data without valid content can also be removed, and the specific process can be: the to-be-desensitized audio data whose audio data is less than or equal to a preset threshold can be removed.

[0086] For example, when the time length of the to-be-desensitized audio data is less than or equal to 5 s, and for example, when the size of the to-be-desensitized audio data is less than or equal to 1 KB, the to-be-desensitized audio data can be considered as invalid audio data.

[0087] In a possible implementation, the to-be-desensitized audio data with repeated content can also be removed, and the specific process can be: when there are multiple to-be-desensitized audio data, the waveforms or spectrums of the multiple to-be-desensitized audio data can be analyzed, and for the multiple to-be-desensitized audio data whose waveform or spectrum similarity is greater than a waveform or spectrum similarity threshold, only one corresponding to-be-desensitized audio data is retained.

[0088] In a possible implementation, the background noise in the to-be-desensitized audio data can also be removed, and the specific process can be: the audio spectrum in the to-be-desensitized audio data can be analyzed based on a frequency domain signal processing method, and compared with the audio spectrum of noise, and when the similarity between the audio spectrum in the to-be-desensitized audio data and the audio spectrum of noise is greater than or equal to a spectrum similarity threshold, it can be considered that the part of to-be-desensitized audio data is noise data. Then, the part of to-be-desensitized audio data is cut off to remove the background noise in the to-be-desensitized audio data.

[0089] In a possible implementation, the silent segment data in the to-be-desensitized audio data can also be removed, and the specific process can be: the to-be-desensitized audio data is detected based on an energy level threshold and a zero-crossing rate to determine the silent segment data, and then the part of to-be-desensitized audio data is cut off to remove the silent segment data in the to-be-desensitized audio data.

[0090] In a possible implementation, the to-be-desensitized audio data can also be normalized, and the specific process can be: based on root mean square (RMS) normalization processing, the audio signal amplitude of the to-be-desensitized audio data is adjusted to a preset audio signal amplitude range.

[0091] S403, for the audio data of each object, sensitive information detection is performed on the audio data of the object to obtain target sensitive data corresponding to the object, so as to obtain all target sensitive data.

[0092] In a possible implementation, the implementation process of S403 can be: performing speech recognition processing on the audio data of the object to obtain text data corresponding to the object; and performing sensitive information detection on the text data corresponding to the object by using a target detection manner to obtain target sensitive data corresponding to the object, the target detection manner including: a first detection manner and / or a second detection manner, the first detection manner being used for detecting a first type of to-be-detected word in the text data based on a preset matching rule, and the second detection manner being used for detecting based on associated content of the to-be-detected word and a second type of to-be-detected word in the text data; and the sensitivity of the first type of to-be-detected word is greater than the sensitivity of the second type of to-be-detected word.

[0093] In a possible implementation, the current scene is determined based on an input manner of the to-be-desensitized audio data; in a case where the current scene is a real-time stream processing scene, it is determined that the target detection manner includes the first detection manner; and in a case where the current scene is an offline batch processing scene, it is determined that the target detection manner includes the first detection manner and the second detection manner.

[0094] In a possible implementation, the desensitization processing is performed on the to-be-desensitized audio data based on all the target sensitive data to obtain desensitized audio data, including the following steps: for each target sensitive data, performing timestamp positioning processing on the target sensitive data to obtain a sensitive data position corresponding to the target sensitive data, the sensitive data position being a position of the target sensitive data in the to-be-desensitized audio data; and performing desensitization processing on the to-be-desensitized audio data based on the sensitive data position corresponding to each target sensitive data to obtain the desensitized audio data.

[0095] S404, desensitization processing is performed on the to-be-desensitized audio data based on all the target sensitive data to obtain desensitized audio data.

[0096] Optionally, the target sensitive information position can be covered by white noise or the target sensitive information can be replaced by silence to obtain audio data that does not contain sensitive data, and in addition, the corresponding text data can be desensitized.

[0097] Further, optionally, the method for desensitizing the text data includes: specific symbol replacement, ciphertext encryption.

[0098] For example, the middle several digits of an ID card number can be replaced by a symbol “*” by using the specific symbol replacement method, for example, “110105********1234”, and the sensitive data in the text data can be encrypted by using the ciphertext encryption method to achieve desensitization of the text data.

[0099] It can be understood that the above audio data desensitization method can be implemented by an audio data desensitization device. In order to realize the above functions, the audio data desensitization device comprises hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the modules and algorithm steps of each example described in the embodiments disclosed herein, the embodiments disclosed in the present application can be realized in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed in hardware or computer software driven hardware depends on the specific application of the technical solution and the design constraint conditions. Professional technicians can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments disclosed in the present application.

[0100] In some embodiments, as previously described in relation to the description of "S403", the audio data to be desensitized can be converted into corresponding text data, and the target sensitive data can be obtained by detecting sensitive information from the text data.

[0101] In view of this, as Figure 6 shown, the above S403 can be specifically determined by the following S601 to S602.

[0102] S601, performing speech recognition processing on the audio data of the object to obtain text data corresponding to the object.

[0103] In a possible implementation, the implementation process of performing speech recognition processing on the audio data of the object to obtain text data corresponding to the object can be: based on a high-efficiency speech endpoint detection (finite state machine network monophone, FSMN-Monophone) voice activity detection (VAD) model, obtaining the start position and end position of the voice appearing in the audio data corresponding to the object, and then based on a Paraformer speech recognition model, processing the start position and end position of the voice appearing to obtain the text data corresponding to the object.

[0104] In a possible implementation, after obtaining the text data corresponding to the object, the following process is further included: in order to improve the readability of the text content of the text data, based on a punctuation (Controllable Time-delayTransformer) model in a real-time post-processing framework, adding punctuation marks to the text data corresponding to the object. Then, by performing reverse text standardization on the obtained text data, the text is converted into a unified format, for example, Chinese is converted into numbers.

[0105] For example, after performing speech recognition processing on the audio data of the object, the text data corresponding to the object is obtained. In the audio data, "123" can be recognized as "one two three" or "123", and "2024-07-25" can be recognized as "2024-07-25". Through reverse text normalization processing, "one two three" and "2024-07-25" can be formatted as "123" and "2024-07-25", respectively.

[0106] S602, sensitive information detection is performed on the text data corresponding to the object by using a target detection method to obtain target sensitive data corresponding to the object.

[0107] The target detection method includes a first detection method and / or a second detection method. The first detection method is used to detect a first type of to-be-detected word in the text data based on a preset matching rule, and the second detection method is used to detect based on the association content of the to-be-detected word and a second type of to-be-detected word in the text data. The sensitivity of the first type of to-be-detected word is greater than the sensitivity of the second type of to-be-detected word.

[0108] Optionally, the first detection method can include string multi-mode (aho-corasick, AC) automatic machine matching, regular expression recognition, and conditional random field (CRF) sequence labeling.

[0109] Further, optionally, the first detection method recognition process can be based on an AC automatic machine matching algorithm to match the first type of to-be-detected word. Regular expression recognition and CRF sequence labeling can be based on a preset regular expression combined with CRF sequence labeling to detect mobile phone numbers, email addresses, ID numbers, and bank card numbers.

[0110] The second detection method can include a two-level sensitive model recognition and a pre-training model recognition.

[0111] The second detection method recognition process can be based on a named entity class recognition (BiLSTM-CRF) model to recognize the second type of to-be-detected word, wherein the second type of to-be-detected word includes context-associated sensitive words.

[0112] For example, homophones or abbreviations of the first category of words to be detected. The pre-trained models include the BERT-WWM pre-trained model and the enhanced representation through knowledge integration (ERNIE) pre-trained model. The pre-trained whole word masking (BERT-WWM) model recognition process can be: based on the BERT-WWM pre-trained model, the second category of words to be detected containing names are recognized. In addition, the BERT-WWM pre-trained model can also be trained, and the names containing polyphones can be recognized based on the trained BERT-WWM pre-trained model. The ERNIE pre-trained model recognition process can be: based on the ERNIE pre-trained model, sensitive data containing address information in text data is recognized.

[0113] For example, when the recognized text is "Zhang Chongyang and Li Xingchang attended the meeting", the BERT-wwm pre-trained model will encode the name containing polyphones, determine the correct pronunciation and meaning of the polyphones based on the context, and output the sensitive data recognition result.

[0114] In some embodiments, the current scenario may be determined based on the target sensitive data input method, and then different target detection methods may be determined based on different scenarios.

[0115] In view of this, if Figure 7 As shown, the specific determination of different target detection methods can be determined through the following S701 to S703.

[0116] S701: Determine the current scene based on the input method of the audio data to be desensitized.

[0117] Optionally, the process of determining the current scene may be: when the method for obtaining the audio data to be desensitized is to obtain the input audio data to be desensitized in real time, generating a real-time stream processing scene identifier, and then determining that the current scene is a real-time stream processing scene based on the real-time stream processing scene identifier. When the method for obtaining the audio data to be desensitized is to read internal data of the audio data desensitization system to obtain the audio data to be desensitized, generating an offline batch processing scene identifier, and determining that the current scene is an offline batch processing scene based on the offline batch processing scene identifier.

[0118] S702: When the current scenario is a real-time stream processing scenario, determine that the target detection mode includes a first detection mode.

[0119] The real-time stream processing scenario refers to a scenario of instant processing and analysis of continuously generated to-be-desensitized audio data, in which processing needs to be performed at the moment of data generation or within a very short time to quickly extract valuable information or make decisions.

[0120] In a case where the current scenario is the offline batch processing scenario, the target detection manner is determined to include the first detection manner and the second detection manner.

[0121] The offline batch processing scenario refers to a scenario of centralized processing of to-be-desensitized audio data through a batch processing manner after the to-be-desensitized audio data accumulates to a certain scale. The offline batch processing scenario pays more attention to the integrity of the to-be-desensitized audio data and the accuracy of the processing result.

[0122] In some embodiments, as previously described in relation to the foregoing "S404", the target sensitive data can be subjected to timestamp positioning processing to obtain a sensitive data position corresponding to the target sensitive data, and the to-be-desensitized audio data can be subjected to desensitization processing based on the sensitive data position.

[0123] In view of this, as shown in the foregoing S404, the foregoing S404 can be specifically determined through the following S801 to S802. Figure 8

[0124] S801, for each target sensitive data, the target sensitive data is subjected to timestamp positioning processing to obtain a sensitive data position corresponding to the target sensitive data.

[0125] In a possible implementation manner, the implementation process of the timestamp positioning processing of the target sensitive data can be that, based on the sensitive data positions obtained through the first detection manner and the second detection manner, a data stamp of the sensitive data in the text data is obtained, and then through the data stamp, a position of the sensitive data in the to-be-desensitized audio data is determined, so as to complete the timestamp positioning processing of the target sensitive data.

[0126] Optionally, the sensitive data can also be verified based on the position of the sensitive data in the to-be-desensitized audio data in combination with the context, so as to improve the accuracy of the sensitive data identification result.

[0127] S802, the to-be-desensitized audio data is subjected to desensitization processing based on the sensitive data position corresponding to each target sensitive data, to obtain desensitized audio data.

[0128] ​In some embodiments, data in the desensitization process can also be acquired; the data in the desensitization process includes at least one of the following: pre-processing process data, speech recognition process data, sensitive information recognition process data, and sound elimination processing process data. Based on reinforcement learning technology and dynamic parameter adjustment technology, analyze the desensitization node data, and obtain the analysis result; based on the analysis result and the optimization rule, optimize the desensitization process.

[0129] In a possible implementation, the process of optimizing the pre-processing process includes: acquiring the signal-to-noise ratio of the audio data to be desensitized; in the case where the signal-to-noise ratio is greater than the signal-to-noise ratio threshold, based on Wiener filtering, the audio data to be desensitized is processed for noise reduction; in the case where the signal-to-noise ratio is less than or equal to the signal-to-noise ratio threshold, based on frequency domain spectral subtraction and wavelet threshold, the audio data to be desensitized is processed for noise reduction.

[0130] In a possible implementation, the process of optimizing the speech recognition process includes: acquiring the current resource utilization rate of the processor; in the case where the current resource utilization rate is greater than the resource utilization rate threshold, the detection model in the second detection mode is switched to the first mode; in the case where the current resource utilization rate is greater than the resource utilization rate threshold, the detection model in the second detection mode is switched to the second mode.

[0131] In a possible implementation, the process of optimizing the sound elimination processing process includes: based on the sensitive data position corresponding to each target sensitive data and the desensitization strategy, the audio data to be desensitized is desensitized to obtain desensitized audio data; the desensitization strategy includes at least one of the following: in the case where the sensitivity of the target sensitive data is greater than or equal to the sensitivity threshold, the target sensitive data is processed for white noise covering and spectral confusion; in the case where the sensitivity of the target sensitive information is less than or equal to the sensitivity threshold, the target sensitive information is processed for mute replacement.

[0132] For example, when the obtained desensitized audio data contains more than or equal to 1 number of prohibited sensitive words, or more than or equal to 2 number of violent sensitive words, the target sensitive information position can be covered by white noise, and then the target sensitive information position is processed for spectral confusion to prevent the desensitized audio data from being restored. When the obtained desensitized audio data contains only address, mailbox and other sensitive data, the target sensitive information position can be replaced by mute to improve the processing speed.

[0133] Optionally, the process of switching the detection mode of the detection model may be: when the current resource utilization is greater than the resource utilization threshold, the detection model in the second detection mode is switched to half-precision floating-point (FP16) mode to reduce the model recognition accuracy and thus reduce the resource utilization. When the current resource utilization is less than the resource utilization threshold, the detection model in the second detection mode is switched to single-precision floating-point (FP32) mode to improve the model recognition accuracy.

[0134] The embodiment disclosed in the present application can divide the functional modules of the audio data desensitization device generated according to the above method example. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment disclosed in the present application is schematic and is only a logical functional division. There may be other division methods in actual implementation.

[0135] Figure 9 This is a schematic diagram of the structure of another audio data desensitization device provided by an embodiment of the present invention. Figure 9 As shown, the audio data desensitization device 90 can be used to perform Figure 4 The audio data desensitization method shown in FIG.

[0136] The communication unit 901 is used to obtain the audio data to be desensitized, which includes audio data of multiple objects; the processing unit 902 is used to perform audio separation processing on the audio data to be desensitized to obtain the audio data of each object; the processing unit 902 is also used to perform sensitive information detection on the audio data of each object, obtain the target sensitive data corresponding to the object, and obtain all target sensitive data; the processing unit 902 is also used to desensitize the audio data to be desensitized based on all target sensitive data to obtain desensitized audio data.

[0137] In a possible implementation, the processing unit 902 is specifically configured to: perform speech recognition processing on the audio data of the object to obtain text data corresponding to the object; and perform sensitive information detection on the text data corresponding to the object by using a target detection manner to obtain target sensitive data corresponding to the object, the target detection manner including: a first detection manner and / or a second detection manner, the first detection manner being used for detecting a first type of to-be-detected word in the text data based on a preset matching rule, and the second detection manner being used for detecting based on associated content of the to-be-detected word and a second type of to-be-detected word in the text data; and wherein the sensitivity of the first type of to-be-detected word is greater than the sensitivity of the second type of to-be-detected word.

[0138] In a possible implementation, the processing unit 902 is specifically configured to: determine a current scene based on an input manner of the to-be-desensitized audio data; and in a case where the current scene is a real-time stream processing scene, determine that the target detection manner includes the first detection manner; and in a case where the current scene is an offline batch processing scene, determine that the target detection manner includes the first detection manner and the second detection manner.

[0139] In a possible implementation, the communication unit 901 is further configured to acquire a current resource utilization rate of the processor; and the processing unit 902 is further configured to: in a case where the current resource utilization rate is greater than a resource utilization rate threshold, switch a detection model in the second detection manner to the first mode; and in a case where the current resource utilization rate is greater than the resource utilization rate threshold, switch the detection model in the second detection manner to the second mode.

[0140] In a possible implementation, the processing unit 902 is specifically configured to: for each target sensitive data, perform timestamp positioning processing on the target sensitive data to obtain a sensitive data position corresponding to the target sensitive data, the sensitive data position being a position of the target sensitive data in the to-be-desensitized audio data; and perform desensitization processing on the to-be-desensitized audio data based on the sensitive data position corresponding to each target sensitive data to obtain desensitized audio data.

[0141] In a possible implementation, the processing unit 902 is specifically configured to: perform desensitization processing on the to-be-desensitized audio data based on the sensitive data position corresponding to each target sensitive data and a desensitization strategy to obtain desensitized audio data; and the desensitization strategy includes at least one of the following: in a case where the sensitivity of the target sensitive data is greater than or equal to a sensitivity threshold, performing white noise covering processing and spectrum confusion processing on the target sensitive data; and in a case where the sensitivity of the target sensitive data is less than or equal to the sensitivity threshold, performing mute replacement processing on the target sensitive information.

[0142] In a possible implementation, the communication unit 901 is further configured to obtain a signal-to-noise ratio of the to-be-desensitized audio data; the processing unit 902 is further configured to, when the signal-to-noise ratio is greater than a signal-to-noise ratio threshold, perform noise reduction processing on the to-be-desensitized audio data based on Wiener filtering; and the processing unit 902 is further configured to, when the signal-to-noise ratio is less than or equal to the signal-to-noise ratio threshold, perform noise reduction processing on the to-be-desensitized audio data based on frequency domain spectral subtraction and wavelet thresholding.

[0143] Those skilled in the art can clearly understand the technical solutions and the technical effects of the present application from the above description of the embodiments. For the convenience and brevity of description, only the division of the functional modules is taken as an example in the above description. In actual applications, the above functions can be completed by different functional modules according to the needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0144] The present disclosure also provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor of an electronic device, enable the electronic device to perform the audio data desensitization method provided in the embodiments of the present disclosure.

[0145] The embodiments of the present disclosure further provide a computer program product containing instructions, which, when running on an electronic device, enable the electronic device to perform the audio data desensitization method provided in the embodiments of the present disclosure.

[0146] Among them, the computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a register, a hard disk, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above, or any other form of computer-readable storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and the storage medium can be located in an application-specific integrated circuit (ASIC). In the embodiments of the present application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0147] The above are only specific embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or replacements within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An audio data desensitization method, characterized by, The method comprises: obtaining to-be-desensitized audio data, the to-be-desensitized audio data comprising audio data of a plurality of objects; performing audio separation processing on the to-be-desensitized audio data to obtain audio data of each object; for the audio data of each object, performing sensitive information detection on the audio data of the object to obtain target sensitive data corresponding to the object, to obtain all the target sensitive data; based on all the target sensitive data, performing desensitization processing on the to-be-desensitized audio data to obtain desensitized audio data.

2. The method of claim 1, wherein, The method further comprises: performing speech recognition processing on the audio data of the object to obtain text data corresponding to the object; performing sensitive information detection on the text data corresponding to the object by using a target detection manner, to obtain target sensitive data corresponding to the object, the target detection manner comprising: a first detection manner and / or a second detection manner, the first detection manner being used for detecting a first type of to-be-detected word in the text data based on a preset matching rule, and the second detection manner being used for detecting based on associated content of the to-be-detected word and a second type of to-be-detected word in the text data; wherein the sensitivity of the first type of to-be-detected word is greater than the sensitivity of the second type of to-be-detected word.

3. The method of claim 2, wherein, The method further comprises: based on an input manner of the to-be-desensitized audio data, determining a current scene; in a case where the current scene is a real-time stream processing scene, determining that the target detection manner comprises the first detection manner; in a case where the current scene is an offline batch processing scene, determining that the target detection manner comprises the first detection manner and the second detection manner.

4. The method according to claim 2 or 3, characterized in that, The second detection manner is detection by a sensitive detection model, the sensitive detection model being used for detecting sensitive data; the method further comprises: obtaining a current resource utilization rate of a processor; in a case where the current resource utilization rate is greater than a resource utilization rate threshold, switching a detection model in the second detection manner to a first mode; in a case where the current resource utilization rate is greater than a resource utilization rate threshold, switching a detection model in the second detection manner to a second mode; the resource occupied by the first mode being less than the resource occupied by the second mode.

5. The method of claim 1, wherein, The method further comprises: for each target sensitive data, performing timestamp positioning processing on the target sensitive data to obtain a sensitive data position corresponding to the target sensitive data, the sensitive data position being a position of the target sensitive data in the to-be-desensitized audio data; based on the sensitive data position corresponding to each target sensitive data, performing desensitization processing on the to-be-desensitized audio data to obtain the desensitized audio data.

6. The method of claim 5, wherein, The method further comprises: The desensitization processing is performed on the to-be-desensitized audio data based on the sensitive data position and the desensitization strategy corresponding to each target sensitive data, to obtain the desensitized audio data. The desensitization strategy comprises at least one of the following: In a case where the sensitivity of the target sensitive data is greater than or equal to a sensitivity threshold, performing white noise covering processing and spectrum confusion processing on the target sensitive data; In a case where the sensitivity of the target sensitive data is less than or equal to the sensitivity threshold, performing mute replacement processing on the target sensitive information.

7. The method according to claim 1 or 2, characterized in that, The method further comprises: obtaining a signal-to-noise ratio of the to-be-desensitized audio data; in a case where the signal-to-noise ratio is greater than a signal-to-noise ratio threshold, performing noise reduction processing on the to-be-desensitized audio data based on Wiener filtering; in a case where the signal-to-noise ratio is less than or equal to the signal-to-noise ratio threshold, performing noise reduction processing on the to-be-desensitized audio data based on frequency domain spectrum subtraction and wavelet threshold.

8. An audio data desensitizing device, characterized by, The audio data desensitization device comprises a communication unit and a processing unit. The communication unit is configured to obtain to-be-desensitized audio data, wherein the to-be-desensitized audio data comprises audio data of multiple objects. The processing unit is configured to perform audio separation processing on the to-be-desensitized audio data to obtain audio data of each object. The processing unit is further configured to, for audio data of each object, perform sensitive information detection on the audio data of the object to obtain target sensitive data corresponding to the object, so as to obtain all target sensitive data. The processing unit is further configured to perform desensitization processing on the to-be-desensitized audio data based on all target sensitive data to obtain desensitized audio data.

9. An audio data desensitization system, characterized by, The audio data desensitization system comprises: an audio preprocessing module configured to process the to-be-desensitized audio data to remove abnormal data in the to-be-desensitized audio data, to obtain first audio data, wherein the abnormal data comprises at least one of the following: damaged data, repeated data, invalid data, and noise data; a multi-person overlapping speech detection and separation cutting module configured to perform audio separation processing on the first audio data to obtain audio data of each object, wherein the audio separation processing comprises multiple object audio data position identification and multiple object audio data separation; a speech recognition module configured to perform speech recognition processing on the audio data of the object to obtain text data corresponding to the object. a sensitive information classification and grading identification module configured to perform sensitive information detection on the text data corresponding to the object by using the target detection method to obtain target sensitive data corresponding to the object. a sensitive data timestamp positioning processing module configured to, for each target sensitive data, perform timestamp positioning processing on the target sensitive data to obtain a sensitive data position corresponding to the target sensitive data, wherein the sensitive data position is a position of the target sensitive data in the to-be-desensitized audio data. a sound elimination and desensitization processing module configured to perform desensitization processing on the to-be-desensitized audio data based on all target sensitive data to obtain desensitized audio data, and further configured to save the desensitized audio data and corresponding desensitized text data. A distributed processing base for optimizing the audio data desensitization system based on a desensitization process and a desensitization result of the target audio data.

10. An audio data desensitizing device, comprising: Comprising: a processor and a communication interface; the communication interface and the processor are coupled, the processor is used to run computer programs or instructions to realize the audio data desensitization method as claimed in any one of claims 1-7.

11. A computer-readable storage medium having stored therein instructions, the computer-readable storage medium comprising: When a computer executes the instructions, the computer executes the audio data desensitization method as claimed in any one of claims 1-7.

Images