No voice attacks

Abstract

In an approach to inhibiting audio attacks, in response to receiving an audio input from a source, a processor uses a predetermined filter including an allow list to determine that the audio input does not match any of the corresponding entries of the predetermined filter. The processor routes the audio input to an adversarial pipeline for processing. The processor uses a predetermined connectionist temporal classification method to identify adversarial examples for the audio input. The processor uses the identified adversarial examples to generate configurable distorted adversarial examples. In response to a reply from the user, the processor injects the configurable distorted adversarial examples as noise into the audio stream of the reply from the user in real time to modify the audio stream. The processor routes the modified audio stream to the source.

Description

[Technical Field]

[0001] FIELD OF THE DISCLOSURE The present disclosure relates generally to the field of audio security, and more particularly to inhibiting audio attacks. [Background technology]

[0002] Voice phishing, or vishing, is the use of telephone communications, often Voice over Internet Protocol (VoIP) communications, to conduct phishing attacks. Landline telephone service has traditionally been reliable. However, vishing scammers now often use modern VoIP features to thwart detection. Voice phishing is typically used to steal credit card numbers and other information used in identity theft schemes. Voice phishing attacks are sometimes carried out using automated voice synthesis systems to lure victims into calling numbers controlled by the attacker. Scammers may attempt to obtain victims' personal information, as well as personal and financial information related to credit cards and bank accounts. VoIP, also known as IP telephony, is a method and group of technologies for delivering voice communications and multimedia sessions over IP networks such as the Internet. Over the years, attackers have improved their methods and leveraged advanced technologies such as artificial intelligence to enhance their techniques and increase their effectiveness in deceiving subscribers. Summary of the Invention

[0003] An aspect of an embodiment of the present disclosure discloses an approach for inhibiting audio attacks. In response to receiving audio input from a source, a processor uses a predetermined filter including an allow list to determine that the audio input does not match any corresponding entries in the predetermined filter. The processor routes the audio input to an adversarial pipeline for processing. The processor uses a predetermined connectionist temporal classification method to identify adversarial examples for the audio input. The processor uses the identified adversarial examples to generate configurable distorted adversarial examples. In response to a reply from a user, the processor injects the configurable distorted adversarial examples as noise into the audio stream of the reply from the user in real time to modify the audio stream. The processor routes the modified audio stream to the source. [Brief explanation of the drawings]

[0004] [Figure 1] FIG. 1 illustrates a functional block diagram illustrating an audio anti-phishing environment, according to one embodiment of the present disclosure. [Figure 2] 2 shows a flowchart illustrating the operational steps of an audio anti-phishing module in the computing device of FIG. 1 according to one embodiment of the present disclosure. [Figure 3] 2 illustrates an exemplary functional diagram of an audio anti-phishing module within the computing device of FIG. 1 according to one embodiment of the present disclosure. [Figure 4] 2 illustrates an exemplary functional flowchart of an audio anti-phishing module in the computing device of FIG. 1 according to one embodiment of the present disclosure. [Figure 5] 2 illustrates an exemplary user scenario without using the voice anti-phishing module implemented in the computing device of FIG. 1 according to one embodiment of the present disclosure. [Figure 6] 2 illustrates an exemplary user scenario with a voice anti-phishing module implemented on the computing device of FIG. 1 according to one embodiment of the present disclosure. [Figure 7]2 illustrates a block diagram of components of the computing device of FIG. 1 according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0005] The present disclosure relates to systems and methods for inhibiting audio attacks.

[0006] Embodiments of the present disclosure recognize the need to utilize available media to inject adversarial data into voice patterns to combat artificial intelligence (AI)-based voice phishing schemes. Embodiments of the present disclosure disclose that AI-based voice phishing schemes fail to properly classify user responses, preventing attackers from continuing their fraud. Embodiments of the present disclosure disclose that 4G / 5G networks may include IP core networks for transmitting voice over IP networks through routing. Voice may be data passing through the network.

[0007] Embodiments of the present disclosure disclose improvements applied to the Adversarial Robustness Toolbox that enable real-time adversarial sample injection. Embodiments of the present disclosure disclose applying a real-time speech pipeline. Embodiments of the present disclosure disclose incorporating the Adversarial Robustness Toolbox into an IP core network that can inject real-time adversarial attacks into audio streams. Embodiments of the present disclosure disclose injecting noise into the audio stream, which may confuse the chatbot, causing it to hang up or ask a question that makes the user aware of the likelihood of fraud. Embodiments of the present disclosure disclose hashing the audio stream and adding the audio stream to a list. This list may be used as threat intelligence in the future, providing telecommunications operators with intelligence information about audio that is likely to be reused in attacks.

[0008] Embodiments of the present disclosure disclose an allow list, which may be a programmable hash table responsible for holding all authorized known chatbot sources that should not be sent to the adversarial pipeline. The adversarial pipeline may be a real-time adversarial sample generator that attacks voice samples and may contaminate the voices processed by untrusted (e.g., malicious) chatbots. Embodiments of the present disclosure disclose adding adversarial noise to render the chatbot unusable. Embodiments of the present disclosure disclose manually triggering adversarial noise to trick the chatbot.

[0009] The present disclosure will now be described in detail with reference to the figures, in which: Figure 1 is a functional block diagram illustrating an audio anti-phishing environment, generally designated 100, according to one embodiment of the present disclosure.

[0010] In the illustrated embodiment, the voice phishing prevention environment 100 includes a computing device 102, a voice input 104, a trusted source 120, an untrusted source 122, and a network 108. The voice input 104 may be voice converted to VoIP via routing over the network 108. The voice may also be data passing over the network 108. When an attacker conducts voice phishing, it may be processed on a chatbot system that is trained to operate on a question / answer basis and can convert voice to text to store information similar to the audio stream.

[0011] In various embodiments of the present disclosure, the computing device 102 may be a laptop computer, a tablet computer, a netbook computer, a personal computer (PC), a desktop computer, a mobile phone, a smartphone, a smartwatch, a wearable computing device, a personal digital assistant (PDA), or a server. In another embodiment, the computing device 102 represents a computing system that utilizes computers and components clustered to function as a single pool of seamless resources. In other embodiments, the computing device 102 may represent a server computing system that utilizes multiple computers as a server system, such as a cloud computing environment. In general, the computing device 102 may be any computing device or combination of devices that has access to the voice anti-phishing module 110 and the network 108 and is capable of processing program instructions and executing the voice anti-phishing module 110 in accordance with embodiments of the present disclosure. The computing device 102 may include internal and external hardware components, as depicted and described in further detail with respect to FIG. 7.

[0012] Additionally, in the illustrated embodiment, the computing device 102 includes a voice anti-phishing module 110. In the illustrated embodiment, the voice anti-phishing module 110 is located on the computing device 102. However, in other embodiments, the voice anti-phishing module 110 is located externally and can be accessed through a communications network, such as the network 108. The communications network can be, for example, a local area network (LAN), a wide area network (WAN) such as the Internet, or a combination thereof, and can include wired, wireless, fiber optic, or other connections known in the art. In general, the communications network can be any combination of connections and protocols that support communications between the computing device 102 and the voice anti-phishing module 110 in accordance with preferred embodiments of the present disclosure.

[0013] In the illustrated embodiment, the voice anti-phishing module 110 includes an allow list 112 and an adversarial pipeline 114. In one or more embodiments, the allow list 112 is a programmable hash table responsible for holding all authorized, known, trusted sources 120 (e.g., regular chatbots) that should not be sent to the adversarial pipeline 114. The allow list 112 may help determine which sources are trusted sources 120. The allow list 112 may allow known, trusted, and authenticated chatbots and may detect zero-day software attacks that are likely to exploit software vulnerabilities before software developers can find fixes.

[0014] In one or more embodiments, the adversarial pipeline 114 may be a real-time adversarial sample generator. Audio may be treated as a data stream passing through an IP network (e.g., network 108). As an example, an adversarial attack is a combination of techniques used to trick a machine learning model into misclassifying data. The adversarial attack may deceive an attacker and reduce the attacker's success rate on the audio network. The adversarial pipeline 114 may generate audio samples that may contaminate audio processed by an untrusted source 122 (e.g., a malicious chatbot). The adversarial pipeline 114 may include an input that receives audio traffic. The adversarial pipeline 114 may include an Adversarial Robustness Toolkit or any other adversarial attack library generator that can apply such attacks to audio streams. The adversarial pipeline 114 may include an output that can encapsulate audio traffic and provide continuity over the audio network (e.g., network 108). As an example, the adversarial pipeline 114 may be a real-time pipeline that can inject adversarial attacks into the audio input 104, causing sources calling the user to fail to transcribe the audio into text and therefore fail to steal information over the phone.

[0015] In one or more embodiments, the voice anti-phishing module 110 is configured to receive a voice input 104. The voice input 104 may be voice converted to VoIP that passes over the network 108 through routing. The voice may also be data that passes over the network 108. If an attacker performs a voice phishing attack, the voice phishing attack may be processed on a chatbot system that is trained to operate on a question / answer basis and can convert voice to text to preserve information similar to the audio stream. The voice anti-phishing module 110 may use predefined filters, including an allow list 112, to determine whether the voice input 104 matches a corresponding entry in the predefined filter. The allow list 112 may be a programmable hash table that serves to maintain a source of all known allowed chatbots that should not be sent to the adversarial pipeline 114.

[0016] In one or more embodiments, the voice anti-phishing module 110 is configured to identify the voice input 104 as coming from a trusted source 120. The voice anti-phishing module 110 may determine that the voice input 104 matches a corresponding entry in a predefined filter in the allow list 112. The voice anti-phishing module 110 may route the voice input 104 from the trusted source 120 for normal processing.

[0017] TIFF0007789099000001.tif191167

[0018] In one or more embodiments, the voice anti-phishing module 110 is configured to create a hash value representative of the poisoned audio stream. The voice anti-phishing module 110 can add the hash value to a list of audio samples likely to be used in voice phishing attacks. For example, after injecting noise into the audio stream, the voice anti-phishing module 110 may hash the audio stream and add the hash value to the list, which can provide intelligence information about audio likely to be reused in attacks. The voice anti-phishing module 110 can track the poisoned data and not only remove it if found, but also determine whether attackers have begun using the poisoned data.

[0019] FIG. 2 is a flowchart 200 illustrating the operational steps of the voice anti-phishing module 110 according to one embodiment of the present disclosure.

[0020] The voice anti-phishing module 110 operates to receive voice input 104. The voice input 104 may be converted to VoIP via routing over the network 108. The voice anti-phishing module 110 also operates to use predetermined filters, including an allow list 112, to determine whether the voice input 104 matches a corresponding entry in the predetermined filter. If the voice anti-phishing module 110 determines that the voice input 104 matches a corresponding entry in the predetermined filter, the voice anti-phishing module 110 operates to identify the voice input 104 as being from a trusted source 120. The voice anti-phishing module 110 operates to route the voice input 104 from the trusted source 120 for normal processing. If the voice anti-phishing module 110 determines that the voice input 104 does not match a corresponding entry in the predetermined filter in the allow list 112, the voice anti-phishing module 110 operates to identify the voice input 104 as being from an untrusted source 122. The voice anti-phishing module 110 operates to route audio input 104 from an untrusted source 122 to the adversarial pipeline 114 for processing. The voice anti-phishing module 110 operates to route a poison audio stream to the untrusted source 122. The voice anti-phishing module 110 operates to create a hash value representative of the poison audio stream. The voice anti-phishing module 110 operates to add the hash value to a list of audio samples likely to be used in voice phishing attacks.

[0021] In step 202, the voice phishing prevention module 110 receives a voice input 104. The voice input 104 may be voice converted to VoIP via routing over the network 108. The voice may also be data passing over the network 108. When an attacker conducts voice phishing, the voice phishing may be processed on a chatbot system that is trained to operate on a question / answer basis and can convert voice to text to store information similar to the audio stream.

[0022] At decision 204, the voice anti-phishing module 110 uses predetermined filters, including an allow list 112, to determine whether the voice input 104 matches a corresponding entry in the predetermined filter. The allow list 112 may be a programmable hash table responsible for holding all trusted sources (e.g., known, authorized chatbot sources) that should not be sent to the adversarial pipeline 114. If the voice anti-phishing module 110 determines that the voice input 104 matches a corresponding entry in the predetermined filter in the allow list 112 (YES at decision 204), then at step 206, the voice anti-phishing module 110 identifies the voice input 104 as being from a trusted source 120. At step 208, the voice anti-phishing module 110 routes the voice input 104 from the trusted source 120 for normal processing.

[0023] TIFF0007789099000002.tif186168

[0024] At step 214, the voice anti-phishing module 110 routes the poisoned audio stream to the untrusted source 122. At step 216, the voice anti-phishing module 110 creates a hash value representative of the poisoned audio stream. At step 218, the voice anti-phishing module 110 adds the hash value to a list of audio samples likely to be used in voice phishing attacks. For example, after injecting noise into the audio stream, the voice anti-phishing module 110 may hash the audio stream and add the hash value to the list, which can provide intelligence information about audio likely to be reused in attacks. The voice anti-phishing module 110 can track the poisoned data and not only remove it if found, but also determine whether an attacker has begun using the poisoned data.

[0025] FIG. 3 illustrates an exemplary functional diagram of the voice anti-phishing module 110, according to an embodiment of the present disclosure.

[0026] 3 , the voice anti-phishing module 110 receives the voice input 104 through a voice router 302 in the network 108. The voice input 104 may be queued in an input queue 304. The voice anti-phishing module 110 determines whether the voice input 104 matches a corresponding entry in the allowed list 112. If the voice anti-phishing module 110 determines that the voice input 104 matches a corresponding entry in the allowed list 112, the voice anti-phishing module 110 routes the voice input 104 for normal processing (e.g., a normal caller and a normal chatbot 312). If the voice anti-phishing module 110 determines that the voice input 104 does not match a corresponding entry in the allowed list 112, the voice anti-phishing module 110 may decap the voice input 104 in the decapper 306. The voice anti-phishing module 110 may route the voice input 104 to the adversarial pipeline 114 for processing. The voice anti-phishing module 110 can identify adversarial examples in the voice input 104 using the CTC method 308. The voice anti-phishing module 110 may use the identified adversarial examples to generate configurable distorted adversarial examples. The voice anti-phishing module 110 can inject the configurable distorted adversarial examples as noise into the audio stream in real time to generate a poison audio stream. By injecting noise into the audio stream, the voice anti-phishing module 110 can confuse malicious chatbots 314. The voice anti-phishing module 110 can route the poison audio stream to the malicious chatbots 314. The voice anti-phishing module 110 can create a hash value representative of the poison audio stream in the audit 310. The voice anti-phishing module 110 may add the hash value to a list of audio samples likely to be used in voice phishing attacks.

[0027] FIG. 4 illustrates an exemplary functional flowchart of the voice anti-phishing module 110 according to an embodiment of the present disclosure.

[0028] In the example of FIG. 4 , a malicious chatbot 314 may call a user 402. The voice anti-phishing module 110 may intercept the call to the user 402. The voice anti-phishing module 110 may analyze the data of the call. The voice anti-phishing module 110 may determine potential spam calls via an allow list 112. The voice anti-phishing module 110 may send the voice traffic through an adversarial pipeline 114. The adversarial pipeline 114 may establish a connection to the malicious chatbot 314. The adversarial pipeline 114 may establish a connection to the user 402. The malicious chatbot 314 may initiate a conversation with the user 402. The malicious chatbot 314 may ask a question to the user 402. The user 402 may answer the question to the adversarial pipeline 114. The voice anti-phishing module 110 may decap the audio stream through the adversarial pipeline 114. The voice anti-phishing module 110 may perform real-time adversarial testing through an adversarial pipeline 114. The voice anti-phishing module 110 may encapsulate the audio. The user 402 may use the processed audio stream to answer the malicious chatbot 314's questions. The malicious chatbot 314 may open a voice response. The malicious chatbot 314 may not receive the desired response. The malicious chatbot 314 may issue a misleading response or simply hang up the phone.

[0029] Figure 5 illustrates an exemplary user scenario without implementing the voice anti-phishing module 110, and Figure 6 illustrates an exemplary user scenario with implementing the voice anti-phishing module 110, according to an embodiment of the present disclosure.

[0030] In the example of Figure 5, if the voice phishing prevention module 110 is not implemented, the malicious chatbot 314 may perform voice phishing on the user 402. Once the malicious chatbot gets all of the scripted answers, the malicious chatbot 314 may hang up and turn to identity theft, voice recording, or something else entirely. In the example of Figure 6, because the voice phishing prevention module 110 is implemented, the malicious chatbot 314 may not receive the desired response. The malicious chatbot 314 may issue a misleading response or simply hang up without gaining any useful information from the user 402.

[0031] 7 illustrates a block diagram 700 of components of a computing device 102 in accordance with an exemplary embodiment of the present disclosure. It should be understood that FIG. 7 is intended to be an illustration of one implementation only and is not intended to imply any limitations with regard to the environments in which different embodiments may be implemented. Many modifications to the depicted environments may be made.

[0032] Computing device 102 may include a communications fabric 702 that provides communications between cache 716, memory 706, persistent storage 708, communications unit 710, and input / output (I / O) interface 712. Communications fabric 702 may be implemented with any architecture designed to pass data and / or control information between processors (such as microprocessors, communications and network processors), system memory, peripheral devices, and other hardware components in a system. For example, communications fabric 702 may be implemented using one or more buses or crossbar switches.

[0033] Memory 706 and persistent storage 708 are computer-readable storage media. In this embodiment, memory 706 includes random access memory (RAM). In general, memory 706 may include any suitable volatile or non-volatile computer-readable storage medium. Cache 716 is high-speed memory that enhances the performance of computer processor 704 by retaining recently accessed data and data near to the accessed data from memory 706.

[0034] The voice anti-phishing module 110 may be stored in persistent storage 708 and memory 706 for execution by one or more respective computer processors 704 via cache 716. In one embodiment, persistent storage 708 includes a magnetic hard disk drive. Alternatively, or in addition to a magnetic hard disk drive, persistent storage 708 may include a solid-state hard drive, a semiconductor memory device, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, or any other computer-readable storage medium capable of storing program instructions or digital information.

[0035] The media used by persistent storage 708 may also be removable. For example, a removable hard drive may be used for persistent storage 708. Other examples include optical and magnetic disks, thumb drives, and smart cards that are inserted into a drive for transfer onto another computer-readable storage medium that is also part of persistent storage 708.

[0036] Communications unit 710, in these examples, provides for communication with other data processing systems or devices. In these examples, communications unit 710 includes one or more network interface cards. Communications unit 710 may provide communications through the use of either or both physical and wireless communications links. Voice anti-phishing module 110 may be downloaded to persistent storage 708 via communications unit 710.

[0037] The I / O interface 712 allows for the input and output of data to and from other devices that may be connected to the computing device 102. For example, the I / O interface 712 may provide a connection to external devices 718, such as a keyboard, keypad, touchscreen, or other suitable input device or combination thereof. The external devices 718 may also include portable computer-readable storage media, such as thumb drives, portable optical or magnetic disks, and memory cards. Software and data used to practice embodiments of the present invention, such as the voice anti-phishing module 110, may be stored on such portable computer-readable storage media and loaded into the persistent storage 708 via the I / O interface 712. The I / O interface 712 also connects to a display 720.

[0038] Display 720 provides a mechanism for displaying data to a user and may be, for example, a computer monitor.

[0039] The programs described herein are identified based on the applications for which they are implemented in particular embodiments of the invention. However, it should be understood that the nomenclature of any particular program herein is used merely for convenience, and thus the invention should not be limited to use in any particular application specified and / or implied by such nomenclature.

[0040] The present invention may be a system, method, or computer program product, or combination thereof, integrated at any possible level of technical detail. The computer program product may include a computer-readable storage medium having stored thereon computer-readable program instructions for causing a processor to carry out aspects of the present invention.

[0041] A computer-readable storage medium may be a tangible device capable of retaining and storing instructions for use by an instruction execution device. By way of example, and not limitation, a computer-readable storage medium may be an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or a suitable combination thereof. A non-exhaustive list of more specific examples of computer-readable storage media includes portable computer diskettes, hard disks, RAM, ROM, EPROM (or flash memory), SRAM, CD-ROMs, DVDs, memory sticks, floppy disks, mechanically encoded devices having instructions recorded on punch cards or ridge-in-groove structures, or the like, and suitable combinations thereof. Computer-readable storage, as used herein, should not be construed as a transitory signal per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., light pulses passing through a fiber optic cable), or electrical signals transmitted over wires.

[0042] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device or to an external computer or external storage device via a network (e.g., the Internet, a local area network, a wide area network, a wireless network, or a combination thereof). The network may be comprised of copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, edge servers, or a combination thereof. A network adapter card or network interface of each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage on a computer-readable storage medium within the respective computing / processing device.

[0043] Computer-readable program instructions for carrying out operations of the present invention may be either source code or object code written in any combination of one or more programming languages, including assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, configuration data for integrated circuits, or object-oriented programming languages ​​such as Smalltalk, C++, etc., and procedural programming languages ​​such as the "C" programming language and similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, as a standalone software package, or partially on the user's computer. Alternatively, the computer may be executed partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA) can execute computer-readable program instructions by utilizing state information of the computer-readable program instructions to personalize the computer-readable program instructions in order to carry out aspects of the present invention.

[0044] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0045] These computer-readable program instructions can be provided to a processor of a computer or other programmable data processing apparatus to create a machine, such that the instructions, when executed by the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams. These computer-readable program instructions can also be stored in a computer-readable storage medium connectable to a computer, programmable data processing apparatus, or other device, or combination thereof, that functions in a particular way, such that the computer-readable program instructions stored therein configure one of the products including instructions that implement aspects of the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams.

[0046] Computer-readable program instructions, such as instructions to perform the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams on a computer, other programmable apparatus, or other device, can also be loaded into a computer, other programmable data processing apparatus, or other device to perform a series of operational steps on the computer, other programmable apparatus, or other device to produce a computer-implemented process.

[0047] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of executable implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowcharts or block diagrams may represent a module, segment, or portion of instructions, which constitute one or more executable instructions for implementing the specified logical function(s). In some alternative embodiments, the functions shown in the blocks may occur out of the order shown in the figures. For example, two blocks shown in succession may actually be accomplished as a single step, executed concurrently, substantially concurrently, partially, or fully in a time-overlapping manner, or the blocks may be executed in the reverse order depending on the functionality involved. It should also be noted that each block of the block diagrams and / or flowchart diagrams, and combinations of blocks in the block diagrams and / or flowchart diagrams, can be implemented by a special-purpose hardware-based system that performs the specified functions or operations or executes a combination of special-purpose hardware and computer instructions.

[0048] The description of various embodiments of the present invention is presented for illustrative purposes, but is not intended to be exhaustive or limited to the disclosed embodiments. It will be apparent to those skilled in the art that many modifications and variations are possible without departing from the scope and spirit of the invention. The terms used herein have been selected to best explain the principles of the embodiments, practical applications or technical improvements to technology found in the market, or to enable those skilled in the art to understand the embodiments described herein.

[0049] Although specific embodiments of the present invention have been described, those skilled in the art will recognize that there are other embodiments that are equivalent to the described embodiments. It is understood, therefore, that the present invention is not to be limited by the specifically illustrated embodiments, but rather is to be limited only by the scope of the appended claims.

Claims

1. In response to receiving an audio input comprising an audio stream from a source, determining, by one or more processors, using a predetermined filter comprising a permission list, that the audio input does not match any corresponding entry of the predetermined filter; routing, by one or more processors, the audio input to an adversarial pipeline; Identifying adversarial examples of the speech input using a predetermined connectionist temporal classification (CTC) method; generating configurable distorted adversarial examples using the identified adversarial examples; and processing the audio stream of user responses by injecting the configurable distorted adversarial examples as noise into the audio stream of user responses in real time to modify the audio stream in response to the user responses; and routing, by one or more processors, an audio input including the modified audio stream to the source.

2. creating a hash value representative of the modified audio stream; The computer-implemented method of claim 1 , further comprising adding the hash value to a list of audio samples likely to be used in an audio attack.

3. The computer-implemented method of claim 1 , further comprising injecting an adversarial attack into the audio stream in real time.

4. Adding adversarial noise to prevent the use of untrusted sources, The computer-implemented method of claim 1 , further comprising manually triggering the adversarial noise.

5. The computer-implemented method of claim 1 , further comprising tracking the modified audio stream.

6. The computer-implemented method of claim 1 , wherein the permission list is a programmable hash table.

7. The computer-implemented method of claim 1 , wherein the noise is white noise.

8. one or more computer-readable storage media and program instructions collectively stored on the one or more computer-readable storage media, the program instructions comprising: program instructions for, in response to receiving an audio input comprising an audio stream from a source, using a predetermined filter comprising a permission list to determine that the audio input does not match any corresponding entry of the predetermined filter; program instructions for routing the audio input to an adversarial pipeline, program instructions for identifying adversarial examples of the speech input using a predetermined CTC method; program instructions for generating configurable distorted adversarial examples using the identified adversarial examples; program instructions for injecting the configurable distorted adversarial examples as noise into the audio stream of user responses in real time to modify the audio stream in response to the user responses; and program instructions for routing an audio input including the modified audio stream to the source.

9. program instructions for creating a hash value representative of the modified audio stream; 9. The computer program product of claim 8, further comprising program instructions for adding the hash value to a list of audio samples likely to be used in an audio attack.

10. 10. The computer program product of claim 8, further comprising program instructions for injecting adversarial attacks into the audio stream in real time.

11. program instructions that add adversarial noise to prevent the use of untrusted sources; 10. The computer program product of claim 8, further comprising: program instructions for manually triggering the hostile noise.

12. 9. The computer program product of claim 8, further comprising program instructions for tracking the modified audio stream.

13. 9. The computer program product of claim 8, wherein the permission list is a programmable hash table.

14. The computer program product of claim 8 , wherein the noise is white noise.

15. one or more computer processors, one or more computer-readable storage media, and program instructions stored on the one or more computer-readable storage media that are executed by at least one of the one or more computer processors, the program instructions comprising: program instructions for, in response to receiving an audio input comprising an audio stream from a source, using a predetermined filter comprising a permission list to determine that the audio input does not match any corresponding entry of the predetermined filter; program instructions for routing the audio input to an adversarial pipeline, program instructions for identifying adversarial examples of the speech input using a predetermined CTC method; program instructions for generating configurable distorted adversarial examples using the identified adversarial examples; program instructions for injecting the configurable distorted adversarial examples as noise into the audio stream of user responses in real time to modify the audio stream in response to the user responses; and program instructions for routing audio input including the modified audio stream to the source.

16. program instructions for creating a hash value representative of the modified audio stream; 16. The computer system of claim 15, further comprising program instructions for adding the hash value to a list of audio samples likely to be used in an audio attack.

17. 16. The computer system of claim 15, further comprising program instructions for injecting adversarial attacks into the audio stream in real time.

18. program instructions that add adversarial noise to prevent the use of untrusted sources; 16. The computer system of claim 15, further comprising program instructions for manually triggering the hostile noise.

19. 16. The computer system of claim 15, further comprising program instructions for tracking the modified audio stream.

20. 16. The computer system of claim 15, wherein the permission list is a programmable hash table.

Images