WO2020119448A1 - 语音信息验证 - Google Patents

语音信息验证 Download PDF

Info

Publication number
WO2020119448A1
WO2020119448A1 PCT/CN2019/120972 CN2019120972W WO2020119448A1 WO 2020119448 A1 WO2020119448 A1 WO 2020119448A1 CN 2019120972 W CN2019120972 W CN 2019120972W WO 2020119448 A1 WO2020119448 A1 WO 2020119448A1
Authority
WO
WIPO (PCT)
Prior art keywords
voice
verified
verification code
verification
text
Prior art date
Application number
PCT/CN2019/120972
Other languages
English (en)
French (fr)
Inventor
陈欢
Original Assignee
北京三快在线科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京三快在线科技有限公司 filed Critical 北京三快在线科技有限公司
Publication of WO2020119448A1 publication Critical patent/WO2020119448A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/22—Interactive procedures; Man-machine interfaces
    • G10L17/24—Interactive procedures; Man-machine interfaces the user being prompted to utter a password or a predefined phrase
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/18—Artificial neural networks; Connectionist approaches
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L63/00—Network architectures or network communication protocols for network security
    • H04L63/08—Network architectures or network communication protocols for network security for authentication of entities
    • H04L63/083—Network architectures or network communication protocols for network security for authentication of entities using passwords
    • H04L63/0838—Network architectures or network communication protocols for network security for authentication of entities using passwords using one-time-passwords
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L63/00—Network architectures or network communication protocols for network security
    • H04L63/08—Network architectures or network communication protocols for network security for authentication of entities
    • H04L63/0861—Network architectures or network communication protocols for network security for authentication of entities using biometrical features, e.g. fingerprint, retina-scan

Definitions

  • the present disclosure relates to the field of artificial intelligence technology, and in particular, to a voice information verification method, a voice information verification device, an electronic device, and a computer-readable storage medium.
  • the present disclosure provides a voice information verification method, a voice information verification device, an electronic device, and a computer-readable storage medium, so as to overcome, at least to a certain extent, the problem of low authentication security due to defects in the prior art.
  • a voice information verification method including: acquiring a verification code and a voice to be verified about the verification code sent by a user terminal; in response to the voice to be verified matching the verification code, It is determined whether the voice to be verified is a non-machine; if the voice to be verified is a non-machine, it is determined that the voice to be verified passes verification.
  • a voice information verification device including: an information acquisition module for acquiring a verification code and a voice to be verified about the verification code sent by a user terminal; a man-machine judgment module for In response to that the voice to be verified matches the verification code, it is determined whether the voice to be verified is a non-machine; the voice verification module is used to determine that the voice to be verified passes the verification if the voice to be verified is a non-machine .
  • an electronic device including: a processor; and a memory for storing executable instructions of the processor; wherein, when the machine-executable instructions are executed, the processor It is configured to perform the method described in the first aspect above.
  • a computer-readable storage medium on which a computer program is stored, which when executed by a processor implements the method described in the first aspect above.
  • FIG. 1 schematically shows a schematic diagram of a system architecture of an operating environment of an exemplary embodiment
  • FIG. 2 schematically shows a step diagram of a voice information verification method in an exemplary embodiment
  • FIG. 3 schematically shows a flowchart of a voice information verification method in an exemplary embodiment
  • FIG. 4 schematically shows a flowchart of a method for obtaining a verification code in an exemplary embodiment
  • FIG. 5 schematically shows a flowchart of another method of voice information verification in an exemplary embodiment
  • FIG. 6 schematically shows a flowchart of a voice information verification method in an exemplary embodiment
  • FIG. 7 schematically shows a structural block diagram of a voice information verification apparatus in an exemplary embodiment
  • FIG. 8 schematically shows an electronic device for implementing the above method in an exemplary embodiment
  • FIG. 9 schematically shows a computer-readable storage medium for implementing the above method in an exemplary embodiment.
  • Example embodiments will now be described more fully with reference to the drawings.
  • the example embodiments can be implemented in various forms, and should not be construed as being limited to the examples set forth herein; on the contrary, providing these embodiments makes the present disclosure more comprehensive and complete, and fully conveys the idea of the example embodiments For those skilled in the art.
  • the described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner.
  • voice verification can be performed by matching the user's own voiceprint features.
  • the user is required to input voice registration information during registration and extract voiceprint feature information therefrom, so as to match the voiceprint feature of the voice verification information when the user logs in later.
  • the voice registration information or voice verification information is synthesized by the machine, for example, after crawling the user's daily voice information and using the voice verification information synthesized by its original sound, the server is difficult to recognize and needs to store the user's Voiceprint feature information increases the system's resource occupation.
  • the exemplary embodiment of the present disclosure first provides a voice information verification method, which can be applied to the user to identify the user through the verification code when logging in the App, web page, or performing sensitive operations such as payment confirmation and password modification. Verification scenario.
  • FIG. 1 shows a schematic diagram of a system architecture of an operating environment of this exemplary embodiment.
  • the system 110 may include a user terminal 111, a network 112, and a server 113.
  • the user terminal 111 may be various personal computers, tablet computers, smart phones, or wearable devices used by the user.
  • the server 113 may Acquire the voice to be verified from the user terminal 111 and verify the voice information.
  • the number of devices shown in FIG. 1 is only exemplary, and any number of user terminals 111 or networks 112 may be provided according to actual needs, and the server 113 may also be a server cluster composed of multiple servers.
  • the method in this exemplary embodiment can be applied to the server 113 shown in FIG. 1.
  • the voice information verification method may include the following steps S210-S230:
  • Step S210 Obtain the verification code and the voice to be verified about the verification code sent by the user terminal.
  • the verification code may be a text verification code, a voice verification code, an image verification code, etc., which may be generated by a specific program of the server, or may be obtained from another server.
  • the voice to be verified is the voice sent by the user according to the content of the verification code, for example, the voice recorded and uploaded by the user reading the content of the verification code.
  • the user terminal can collect the voice to be verified about the verification code through the user's trigger operation, and then send it to the server. For example, when entering the voice to be verified, the user can click a specific control to enter the voice to be verified, or choose to use the existing recording file as Voice to be verified, etc.
  • step S220 in response to the matching of the voice to be verified and the verification code, it is determined whether the voice to be verified is a non-machine.
  • the voice to be verified and the verification code can have multiple matching methods, for example: if the verification code is a text verification code , You can first convert the voice to be verified into text to be verified, and then compare the text to be verified with the text verification code to match; if the verification code is a voice verification code, you can compare the voice verification code with the voiceprint of the voice to be verified Features, etc. to match, etc.
  • the verification code matching of the voice to be verified is equivalent to the first verification, which is mainly used to verify identity consistency, and can identify situations where a user account is stolen and logged in or operated on an unbound terminal.
  • the human-machine judgment of the voice to be verified is continued, which is the second verification.
  • the voiceprint features of the voice to be verified can be analyzed and matched to compare whether the voiceprint features are the same or similar to the voiceprint features of the machine-synthesized sound, and whether the voiceprint features are also compared Match the voiceprint feature in the user's voice database.
  • Step S230 if the voice to be verified is a non-machine, it is determined that the voice to be verified passes the verification.
  • the situation that the machine synthesizes the voice can be excluded, thereby determining that the user who sent the voice to be verified received the correct verification code, and that for the real user, it can be determined that the voice to be verified finally passes verification.
  • step S220 may include the following steps:
  • the target spectrogram is the spectrogram corresponding to the voice to be verified.
  • the server obtains the voice to be verified, it can convert it into a voice file in a suitable format, perform time-domain analysis and frequency-domain analysis on it, and draw a target spectrum chart.
  • the target spectrogram contains the spectral features of the voice to be verified, and the spectral features can reflect the voiceprint features of the voice to be verified. Therefore, through the processing of the convolutional neural network model, the voiceprint features can be identified and compared with The voiceprint features of machine synthesized speech or real human speech learned during model training are compared, and the target spectrum is judged to be machine or non-machine, so that the human-machine classification results of the voice to be verified are obtained, so that the second verification can be completed process.
  • FIG. 3 shows a flowchart of a voice information verification method in this exemplary embodiment.
  • the verification code can be matched first. If the match fails, the verification failure result can be directly output. If the match is successful, the voice to be verified is converted into the target spectrogram, and then according to the volume
  • the product neural network outputs the machine decision probability of the target spectrogram, and judges whether the machine decision probability is greater than the probability threshold. If it is greater than the probability threshold, the verification fails. If it is not greater than the probability threshold, the final verification result can be output.
  • this exemplary embodiment matches the verification code and the voice to be verified about the verification code sent by the user terminal, and determines whether the voice to be verified is a non-machine if the match is successful, if the judgment result is a non-machine, then The verification result of the voice to be verified.
  • verification of the verification code matching and human-machine judgment of the voice to be verified can not only verify the identity of the user but also reduce the attacks on the system by the machine synthesized voice, and improve the security of the voice information verification method.
  • this exemplary embodiment there is no need for the user to input the registered voice information in advance, and there is no need to store the user's voiceprint feature information, thereby reducing the cost of using the voice information verification method, simplifying the user's operation flow, and reducing The system resources are occupied and the efficiency is improved.
  • the above-mentioned convolutional neural network model can be trained by the following steps:
  • the sample speech may be historical verification speech, or may include a part of machine synthesized speech; the classification label may refer to a label that each sample speech manually labeled belongs to "machine” or "non-machine".
  • the convolutional neural network model takes the sample spectrum map as input and outputs the classification results of each sample spectrum map. By adjusting the model parameters, the output classification results can be made closer to the classification label.
  • the target spectrogram belongs to a machine or a non-machine.
  • the convolutional neural network model is used to classify the target spectrum of the voice to be verified, which is equivalent to the second verification of the voice to be verified, which is mainly used to identify the situation where the machine synthesizes the voice. Therefore, when the human-machine classification result of the target spectrogram is non-machine, it can be considered that the voice to be verified is a real human voice, thereby finally determining that the voice to be verified passes verification.
  • the human-machine classification of the target spectrogram can be achieved by the machine decision probability output by the convolutional neural network model.
  • the convolutional neural network model analyzes the target spectrogram, it can output the machine decision probability according to the similarity between the target spectrogram and the feature spectrum of the machine synthesized speech. The higher the probability, the greater the probability that the target spectrogram belongs to the machine. If the judgment probability of the machine is greater than the probability threshold, the human-machine classification result of the target spectrogram can be regarded as a machine.
  • the probability threshold can be set and adjusted according to the training and actual application, so that it can accurately distinguish the target spectrogram between the machine and the non-machine.
  • acquiring the verification code in step S210 may include: acquiring the verification code subjected to the disturbance processing.
  • the disturbance processing may be to add information that is disturbing to the identification of the verification code, thereby increasing the difficulty of machine identification.
  • the disturbance processing can be to add interference characters to the text verification code; if the verification code is a voice verification code, the disturbance processing can be to add interference music, sound waves of a specific frequency to the voice verification code, etc. ; If the verification code is an image verification code, the disturbance processing can be to change the image color of the image verification code, reduce the image resolution, or disrupt the image composition sequence of the image verification code, etc.; if the verification code is an SMS verification code, the disturbance processing can be Add out-of-order letters or numeric strings to the text message, etc.
  • obtaining the verification code added with the disturbance information may include the following steps:
  • the image verification code subjected to the perturbation processing is generated by performing one or more processes of deforming, discoloring, blurring, and adding noise to the target picture.
  • FIG. 4 shows a flowchart of obtaining the verification code.
  • the preset text can be obtained from the text library, or it can be a random number generated, etc.
  • the obtained preset text is converted into a target picture
  • the target picture may contain a verification code Pictures of text information, such as pictures containing text verification code characters or pictures of image verification code fragments.
  • the deformation processing can be processing such as stretching, rotating, and liquefying the picture;
  • the color changing processing refers to changing the color gradation parameters of the target picture, such as converting the target picture into a grayscale image, adjusting the light and dark distribution, etc.;
  • the blurring processing means Reduce the resolution of the target picture, for example, perform local pixel compression on the target picture; increase the noise processing can add particles that affect the recognition of the target picture, etc.
  • Other interference processing methods that are easily conceived by those skilled in the art should also be included in the protection scope of the present disclosure.
  • the voice information verification method may further include the following steps:
  • the voice verification to be verified fails, and a failure prompt message is returned to the user terminal;
  • the voice to be verified is converted into text to be verified, and the text to be verified is matched with the verification code.
  • the length of the voice to be verified may be its duration, or its file size, etc. Considering that the too short voice to be verified may be caused by incorrect input, resulting in meaningless verification and affecting the working efficiency of the server, you can set a preset length as the judgment standard of the length of the voice to be verified, to filter out the above situation, for example Set the preset length to two seconds, and the voice less than two seconds will be automatically filtered by the server and return a failure message; for example, if the preset length is set to 5KB, the voice to be verified less than 5KB will also be automatically filtered by the server, etc. .
  • the voice to be verified After detecting that the length of the voice to be verified is greater than or equal to the preset length, the voice to be verified can be converted into text to be verified, where the text to be verified is text to be verified corresponding to the content of the voice to be verified, and the voice text can be used Conversion tool to achieve conversion.
  • the above preset length can be regarded as the lower limit of the length of the voice to be verified. It needs to be added that there may be a user mistakenly inputting or sending a wrong recording file, which makes the voice to be verified received by the server too complicated or lengthy, which may cause the server to process Difficulties increase the meaningless workload. Therefore, the upper limit of the length of the voice to be verified can also be set to filter out the voice that is too complex or lengthy to be verified. When the voice length exceeds the upper limit of the voice length, the result of verification failure can be output, and a failure prompt message is returned. In addition, you can set the upper limit of the recording time of the voice to be verified in the client program. When the upper limit is exceeded, the voice input is automatically terminated, and the recorded voice is sent to the server for voice information verification.
  • converting the voice to be verified into text to be verified can be achieved by the following steps:
  • the delayed neural network model is used to convert the pre-processed voice to be verified into text to be verified.
  • the voice to be verified obtained by the server may contain other noise and other influencing factors, the processing and recognition of the voice to be verified cause certain obstacles.
  • the voice to be verified can be preprocessed.
  • the channel conversion processing means that if the voice features of the voice to be extracted do not distinguish between channels, the multi-channel voice to be verified can be converted to mono; the pre-emphasis processing can be to reserve a certain frequency of the voice to be verified
  • the voice signal within the range is helpful for the server to analyze the voice information to be verified;
  • the voice enhancement process can be to filter out the noise from the voice to be verified to extract the pure voice signal;
  • the blank removal process refers to removing the actual signal from the voice to be verified ,
  • the voice to be verified may have noise or other blank invalid voice fragments, and removing the blank processing can also reduce the length of the voice to be verified and the file Size, reduce the processing capacity of
  • the voice activity detection can also be used to determine whether there is verifiable voice information in the voice to be verified , If it does not exist, you can output the result of verification failure.
  • other preprocessing methods that are easily conceived by those skilled in the art should also be included in the protection scope of the present disclosure.
  • time-delay neural network model After preprocessing the speech to be verified, it can be input into a time-delay neural network model, and the time-delay neural network model can perform frame-by-frame recognition of the speech to be verified, and finally convert it into corresponding text to be verified.
  • time-delay neural network model For the training and use of the time-delay neural network model, reference may be made to methods well known to those skilled in the art, which will not be described in detail here.
  • the verification code may include a text verification code
  • matching the text to be verified with the verification code may include the following steps: matching the text to be verified with the text verification code to obtain a typo ratio of the text to be verified.
  • step S220 may include the following step: if the typo ratio is lower than the matching threshold, it is determined whether the voice to be verified is a non-machine.
  • the typo ratio may be the percentage of the number of unmatched characters in the total number of characters.
  • the matching result may be different due to different matching methods selected. For example: if the text verification code is "today is really good” and the text to be verified is "today is really good", using forward matching, the proportion of typos in the matching result will be higher than that obtained by bidirectional matching.
  • the two-way matching method can improve the accuracy of server matching calculation, but it also increases the server's calculation task and has higher requirements for the server's configuration. Forward matching and reverse matching require lower server computing power and less processing capacity. You can choose the appropriate matching method according to the actual situation.
  • the matching threshold may be a set upper limit of an acceptable matching typo ratio. Considering that the server may convert the voice to be verified into text to be verified, and errors that may occur when matching the text to be verified and the text verification code, a certain inconsistency between the text to be verified and the text verification code may be allowed.
  • the matching threshold can be set according to the actual situation. For example, when the verification code is short or there is little interference information, the matching threshold can be set higher. When the verification code is long or the content is complex, the matching threshold can be set lower. This embodiment There is no specific limitation on this. When the proportion of typos is lower than the matching threshold, it can be considered that the text to be verified matches the verification code, that is, the voice to be verified matches the verification code successfully. After the first verification, the second verification can be continued, that is, the verification voice Man-machine judgment.
  • FIG. 5 shows a flowchart of a voice information verification method in this exemplary embodiment.
  • converting the voice to be verified into the target spectrogram may include: converting the voice to be verified into the target spectrogram through a short-time Fourier transform.
  • Short-time Fourier transform can transform complex sound signals into the frequency domain, and then analyze the characteristics of the time domain signal of the voice to be verified through the spectral characteristics. For example, multiple instantaneous segments can be extracted from the voice to be verified, and arranged in time series, and each segment is converted into a frequency-energy image, thereby obtaining a sequence of multiple target spectrograms, which can be subsequently used by convolutional nerves Network model for processing.
  • FIG. 6 shows a flowchart of a voice information verification method in this exemplary embodiment.
  • the server can obtain the preset text from the text library and obtain the verification code through interference processing, and the user terminal obtains the voice to be verified input by the user, and after preprocessing the voice to be verified, the delay neural network model is used Convert the voice to be verified into text to be verified, and match the text to be verified and the verification code by calculating the proportion of typos. If the matching is successful, the speech to be verified can be converted into the target spectrogram through the short-time Fourier transform. Further, the target spectrogram is analyzed through the convolutional neural network model completed by training and testing to obtain the human-machine classification results. If the human-machine classification result is non-machine, then the final verification result can be output.
  • the device 700 may include: an information acquisition module 710 for acquiring a verification code and a voice to be verified about the verification code sent by the user terminal; a man-machine judgment module 720 for responding to the voice to be verified and verification The codes match to determine whether the voice to be verified is a non-machine; the voice verification module 730 is used to determine that the voice to be verified passes the verification if the voice to be verified is a non-machine.
  • the man-machine judgment module 720 may include: a spectrogram conversion unit for converting the voice to be verified into a target spectrogram if the voice to be verified matches the verification code; a spectrogram analysis unit for The target spectrogram is analyzed through a convolutional neural network model to obtain the human-machine classification result of the target spectrogram; the voice judgment unit is used to determine whether the voice to be verified is non-machine based on the human-machine classification result.
  • the information acquisition module 710 can also be used to acquire the verification code subjected to the disturbance processing.
  • the information acquisition module 710 may include: a text acquisition unit for acquiring preset text and converting the preset text into a target picture; a picture processing unit for transforming and adding the target picture One or more of noise, discoloration, and blurring to generate a perturbed image verification code.
  • the voice information verification apparatus 700 may further include: a voice length detection unit for detecting the length of the voice to be verified; a preset length determination unit for when the length of the voice to be verified is less than a preset length At the time, it is determined that the voice verification fails to be verified, and a failure prompt message is returned to the user terminal, and when the length of the voice to be verified is greater than or equal to the preset length, the voice to be verified is converted into text to be verified, and the text to be verified and the verification code Make a match.
  • the preset length determination unit may further include: a preprocessing subunit, configured to perform one or more preprocessings of channel conversion, pre-emphasis, voice enhancement, and blank removal of the voice to be verified;
  • the model processing subunit is used to convert the pre-processed voice to be verified into text to be verified by using a time-delay neural network model.
  • the verification code may include a text verification code
  • the preset length determination unit may further include: a text matching unit for matching the text to be verified with the text verification code to obtain a typo ratio of the text to be verified.
  • the human-machine judgment module 720 can be used to judge whether the voice to be verified is a non-machine if the typo ratio is lower than the matching threshold.
  • Exemplary embodiments of the present disclosure also provide an electronic device capable of implementing the above method.
  • the electronic device 800 according to this exemplary embodiment of the present disclosure is described below with reference to FIG. 8.
  • the electronic device 800 shown in FIG. 8 is only an example, and should not bring any limitation to the functions and use scope of the embodiments of the present disclosure.
  • the electronic device 800 is represented in the form of a general-purpose computing device.
  • the components of the electronic device 800 may include, but are not limited to: the at least one processor 810, the at least one memory 820, a bus 830 connecting different system components (including the memory 820 and the processor 810), and a display 840.
  • the memory 820 stores program codes, and the program codes may be executed by the processor 810 so that the processor 810 executes the steps according to various exemplary embodiments of the present disclosure described in the "exemplary method" section of the above description of the present specification.
  • the processor 810 may execute steps S210 to S230 shown in FIG. 2 and so on.
  • the memory 820 may include a readable medium in the form of volatile memory, such as random access memory (Random Access Memory, RAM) 821 and/or cache memory 822, and may further include read-only memory (Read-Only, ROM) 823 .
  • RAM Random Access Memory
  • ROM read-only memory
  • the memory 820 may further include a program/utility tool 824 having a set of (at least one) program modules 825.
  • program modules 825 include but are not limited to: an operating system, one or more application programs, other program modules, and program data. These Each of the examples or some combination may include an implementation of the network environment.
  • the bus 830 may be one or more of several types of bus structures, including a memory bus or a memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of a variety of bus structures.
  • the electronic device 800 may also communicate with one or more external devices 1000 (eg, keyboard, pointing device, Bluetooth device, etc.), and may also communicate with one or more devices that enable a user to interact with the electronic device 800, and/or This enables the electronic device 800 to communicate with any device (eg, router, modem, etc.) that communicates with one or more other computing devices. Such communication may be performed through an input/output (Input/Output, I/O) interface 850.
  • the electronic device 800 may also communicate with one or more networks (such as a local area network (Local Area Network, LAN), a wide area network (Wide Area Network, WAN), and/or a public network, such as the Internet) through a network adapter 860. As shown in FIG.
  • the network adapter 860 communicates with other modules of the electronic device 800 through the bus 830. It should be understood that although not shown in the figure, other hardware and/or software modules may be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives And data backup storage system.
  • the example embodiments described here can be implemented by software, or can be implemented by software in combination with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, U disk, mobile hard disk, etc.) or on a network , Including several instructions to cause a computing device (which may be a personal computer, server, terminal device, or network device, etc.) to perform the method according to an exemplary embodiment of the present disclosure.
  • a computing device which may be a personal computer, server, terminal device, or network device, etc.
  • Exemplary embodiments of the present disclosure also provide a computer-readable storage medium having stored thereon a program product capable of implementing the above method of this specification.
  • various aspects of the present disclosure may also be implemented in the form of a program product, which includes program code, and when the program product runs on the terminal device, the program code is used to cause the terminal device to execute the The steps according to various exemplary embodiments of the present disclosure described in the "Exemplary Method" section.
  • a program product 900 for implementing the above method according to an exemplary embodiment of the present disclosure is described, which can adopt a portable compact disk read-only memory (CD-ROM) and include a program code, and can be used in a terminal Devices, such as personal computers.
  • CD-ROM portable compact disk read-only memory
  • the program product of the present disclosure is not limited thereto, and in this document, the readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus, or device.
  • the program product may use any combination of one or more readable media.
  • the readable medium may be a readable signal medium or a readable storage medium.
  • the readable storage medium may be, for example but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination of the above.
  • Readable storage media include, but are not limited to: electrical connection with one or more wires, portable disk, hard disk, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (Erasable Programmable Read-only Memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the foregoing.
  • the computer-readable signal medium may include a data signal that is transmitted in baseband or as part of a carrier wave, in which readable program code is carried. This propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • the readable signal medium may also be any readable medium other than a readable storage medium, and the readable medium may send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device.
  • the program code contained on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
  • Program code for performing the operations of the present disclosure can be written in any combination of one or more programming languages.
  • the programming language includes object-oriented programming languages such as Java, C++, etc., as well as conventional procedural programming Language-such as "C" language or similar programming language.
  • the program code may be executed entirely on the user's computing device, partly on the user's device, as an independent software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server On the implementation.
  • the remote computing device may be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (eg, using Internet service provision Business to connect via the Internet).
  • LAN local area network
  • WAN wide area network
  • Internet service provision Business to connect via the Internet
  • modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory.
  • the features and functions of the two or more modules or units described above may be embodied in one module or unit.
  • the features and functions of a module or unit described above can be further divided into multiple modules or units to be embodied.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Telephonic Communication Services (AREA)

Abstract

一种语音信息验证方法及装置、电子设备、计算机可读存储介质。方法包括：获取验证码以及用户终端发送的关于验证码的待验证语音(S210)；响应于待验证语音与验证码匹配，判断待验证语音是否为非机器(S220)；如果待验证语音为非机器，则确定待验证语音验证通过(S230)。

Description

语音信息验证 技术领域
本公开涉及人工智能技术领域，尤其涉及一种语音信息验证方法、语音信息验证装置、电子设备及计算机可读存储介质。
背景技术
随着计算机技术的发展，在很多App(Application，应用程序)及网站上出现了盗取账户、虚假账户、网络诈骗等恶意行为，对App及网站的正常运营造成了安全隐患。因此很有必要对各种账户进行身份验证，以阻止盗取账户或虚假账户进行各种网络行为，保障真实用户的利益。
需要说明的是，在上述背景技术部分公开的信息仅用于加强对本公开的背景的理解，因此可以包括不构成对本领域普通技术人员已知的现有技术的信息。
发明内容
本公开提供了一种语音信息验证方法、语音信息验证装置、电子设备及计算机可读存储介质，进而至少在一定程度上克服由于现有技术的缺陷而导致的身份验证安全性较低的问题。
本公开的其他特性和优点将通过下面的详细描述变得显然，或部分地通过本公开的实践而习得。
根据本公开的第一方面，提供一种语音信息验证方法，包括：获取验证码以及用户终端发送的关于所述验证码的待验证语音；响应于所述待验证语音与所述验证码匹配，则判断所述待验证语音是否为非机器；如果所述待验证语音为非机器，则判定所述待验证语音验证通过。
根据本公开的第二一方面，提供一种语音信息验证装置，包括：信息获取模块，用于获取验证码以及用户终端发送的关于所述验证码的待验证语音；人机判断模块，用于响应于所述待验证语音与所述验证码匹配，判断所述待验证语音是否为非机器；语音验证模块，用于如果所述待验证语音为非机器，则确定所述待验证语音验证通过。
根据本公开的第三方面，提供一种电子设备，包括：处理器；以及存储器，用于存 储所述处理器的可执行指令；其中，当执行所述机器可执行指令时，所述处理器配置为执行上述第一方面所述的方法。
根据本公开的第四方面，提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现上述第一方面所述的方法。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，并不能限制本公开。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，示出了符合本公开的实施例，并与说明书一起用于解释本公开的原理。显而易见地，下面描述中的附图仅仅是本公开的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1示意性示出示例性实施例运行环境的一种系统架构示意图；
图2示意性示出示例性实施例中一种语音信息验证方法的步骤图；
图3示意性示出示例性实施例中一种语音信息验证方法的流程图；
图4示意性示出示例性实施例中一种获取验证码方法的流程图；
图5示意性示出示例性实施例中另一种语音信息验证方法的流程图；
图6示意性示出示例性实施例中一种语音信息验证方法的流程图；
图7示意性示出示例性实施例中一种语音信息验证装置的结构框图；
图8示意性示出示例性实施例中一种用于实现上述方法的电子设备；
图9示意性示出示例性实施例中一种用于实现上述方法的计算机可读存储介质。
具体实施方式
现在将参考附图更全面地描述示例实施方式。然而，示例实施方式能够以多种形式实施，且不应被理解为限于在此阐述的范例；相反，提供这些实施方式使得本公开将更加全面和完整，并将示例实施方式的构思全面地传达给本领域的技术人员。所描述的特征、结构或特性可以以任何合适的方式结合在一个或更多实施方式中。
在相关技术的一种方案中，可通过用户自身的声纹特征匹配以进行语音验证。需要用户在注册时输入语音注册信息，从中提取声纹特征信息，以便于之后在用户登录时，对语音验证信息进行声纹特征匹配。但是，在该方案中，如果语音注册信息或语音验证信息是由机器合成的，例如爬取用户的日常语音信息后，利用其原声合成的语音验证信息，则服务器难以识别，且需要存储用户的声纹特征信息，增加了系统的资源占用。
鉴于上述问题，本公开的示例性实施例首先提供了一种语音信息验证方法，可以应用于用户在登录App、网页或进行支付确认、修改密码等敏感性操作时，通过验证码对用户进行身份验证的场景。
图1示出了本示例性实施例的运行环境的一种系统架构示意图。参考图1所示，该系统110可以包括用户终端111，网络112及服务器113。其中，用户终端111可以是用户使用的各种个人计算机、平板电脑、智能手机或可穿戴设备等终端设备，采集到关于验证码的待验证语音后，通过网络112发送至服务器113；服务器113可以从用户终端111获取待验证语音，并对其进行语音信息验证。
应当理解，图1中所示各装置的数目仅为示例性的，根据实际需要，可以设置任意数目的用户终端111或网络112，服务器113也可以是由多个服务器组成的服务器集群。
基于上述说明，本示例性实施例中的方法可以应用于图1所示的服务器113上。
下面结合附图2对本示例性实施例做进一步说明，如图2所示，语音信息验证方法可以包括以下步骤S210～S230：
步骤S210，获取验证码以及用户终端发送的关于该验证码的待验证语音。
其中，验证码可以是文本验证码、语音验证码、图像验证码等，可以由服务器的特定程序生成，也可以从其他服务器获取。待验证语音是用户根据验证码的内容发送的语音，例如用户朗读验证码的内容而录制并上传的语音。用户终端可以通过用户的触发操作来采集关于验证码的待验证语音，然后发送至服务器，例如在输入待验证语音时，用户可以点击特定控件录入待验证语音，或者选择将已有的录音文件作为待验证语音等。
步骤S220，响应于待验证语音与验证码匹配，判断待验证语音是否为非机器。
在本示例性实施例中，首先，可以检测待验证语音与验证码是否匹配，对于不同类型的验证码，待验证语音与验证码可以有多种匹配方式，例如：如果验证码为文本验证码，则可以先将待验证语音转换为待验证文本，再将待验证文本与文本验证码对比来进行匹配；如果验证码为语音验证码，则可以通过对比语音验证码与待验证语音的声纹特 征等来进行匹配等等。在步骤S220中，对待验证语音进行验证码匹配相当于第一次验证，主要用于验证身份一致性，可以识别出用户账户被窃取而在非绑定终端上登录或操作等情况。
在本示例性实施例中，如果上述第一次验证通过，则对待验证语音继续进行人机判断，为第二次验证。在判断待验证语音为机器或非机器时，可以对待验证语音的声纹特征进行分析匹配，比较其声纹特征是否与机器合成声音的声纹特征相同或相近，也可以比较其声纹特征是否与用户声音数据库中的声纹特征匹配等。
步骤S230，如果待验证语音为非机器，则确定待验证语音验证通过。
通过对待验证语音进行第二次验证，可以排除机器合成语音的情况，从而确定了发送待验证语音的用户收到了正确的验证码，并且为真实用户，可以确定待验证语音最终通过验证。
在一示例性实施例中，步骤S220可以包括以下步骤：
如果待验证语音与验证码匹配，则将待验证语音转换为目标频谱图；
通过卷积神经网络模型对目标频谱图进行分析，得到目标频谱图的人机分类结果；
基于人机分类结果，确定待验证语音是否为非机器。
目标频谱图是待验证语音所对应的频谱图。频谱图通常有两种：一种是以频率为横坐标，以信号能量为纵坐标的瞬时频谱图，待验证语音可以转换为多幅瞬时频谱图的序列；一种是以时间为横坐标，叠加频率为纵坐标的连续频谱图，待验证语音可以转换一幅连续频谱图。在服务器获取到待验证语音后，可以将其转换为合适格式的语音文件，对其进行时域分析和频域分析，绘制出目标频谱图。可见，目标频谱图包含了待验证语音的频谱特征，而频谱特征又能够体现出待验证语音的声纹特征，因此通过卷积神经网络模型处理，可以识别其中的声纹特征，并将其与模型训练时学习到的机器合成语音或人类真实语音的声纹特征进行比较，判断目标频谱图为机器或非机器，这样得到了待验证语音的人机分类结果，从而可以完成第二次验证的过程。
图3示出了本示例性实施例中一种语音信息验证方法的流程图。如图3所示，获取待检测语音后，可以首先进行验证码匹配，如果匹配失败，则可以直接输出验证失败的结果，如果匹配成功，则将待验证语音转换为目标频谱图，再根据卷积神经网络输出目标频谱图的机器判定概率，并判断机器判定概率是否大于概率阈值，如果大于概率阈值，则验证失败，如果未大于概率阈值，则可以输出最终验证通过的结果。
基于上述说明，本示例性实施例对验证码及用户终端发送的关于验证码的待验证语音进行匹配，并判断匹配成功的将待验证语音是否为非机器，如果判断结果为非机器，则得到待验证语音的验证结果。一方面，对待验证语音进行验证码匹配以及人机判断的验证，能够在验证用户身份一致性的同时，减少机器合成语音对系统进行的攻击，提高了语音信息验证方法的安全性。另一方面，本示例性实施例中，不需要用户预先输入注册语音信息，也不需要存储用户的声纹特征信息，从而降低了语音信息验证方法的使用成本，简化了用户的操作流程，减少了系统的资源占用，提高了效率。
在一示例性实施例中，上述卷积神经网络模型可以通过以下步骤实现训练：
获取多个样本语音以及样本语音的分类标签；
将各样本语音转换为样本频谱图；
利用样本频谱图以及分类标签训练并得到卷积神经网络模型。
其中，样本语音可以是历史验证语音，也可以包括一部分机器合成语音；分类标签可以指人工标注的每个样本语音属于“机器”或“非机器”的标签。
卷积神经网络模型以样本频谱图为输入，输出各样本频谱图的分类结果，通过调整模型参数，可以使输出的分类结果越来越接近分类标签。在训练中，也可以将样本频谱图及分类标签划分为训练集与验证集(例如8:2划分)。其中，训练集用于训练模型，设定初始学习率，以及每学习预设次数学习率降低的百分比，可以使模型更快的收敛；验证集用于验证模型的训练效果，如果在验证集上模型运行的准确率达到一定标准，则可以认为训练完成。
利用训练完成的卷积神经网络模型对目标频谱图进行分析，可以得到目标频谱图的人机分类结果，即目标频谱图属于机器或非机器。
利用卷积神经网络模型对待验证语音的目标频谱图进行人机分类，相当于对待验证语音进行了第二次验证，主要用于识别机器合成语音的情况。因此，当目标频谱图的人机分类结果为非机器时，可以认为待验证语音为真实的人类语音，从而最终判定待验证语音验证通过。
在一示例性实施例中，对目标频谱图的人机分类可以通过卷积神经网络模型输出的机器判定概率实现。卷积神经网络模型分析目标频谱图时，可以根据目标频谱图与机器合成语音的特征频谱图相似程度，输出机器判定概率，概率越高表示目标频谱图属于机器的可能性越大。如果机器判定概率大于概率阈值，则可以认为目标频谱图的人机分类 结果为机器。可以根据训练及实际应用的情况设定并调整概率阈值，使其可以准确区分机器与非机器的目标频谱图。
在一示例性实施例中，步骤S210中获取验证码可以包括：获取经过扰动处理的验证码。
其中，扰动处理可以是添加对验证码的识别具有干扰性的信息，从而增加机器识别的难度。例如：如果验证码是文本验证码，扰动处理可以是在文本验证码中加入干扰字符等；如果验证码是语音验证码，扰动处理可以是在语音验证码中加入干扰音乐、特定频率的声波等；如果验证码是图像验证码，扰动处理可以是改变图像验证码的图像颜色、降低图像分辨率、打乱图像验证码的图像组成顺序等；如果验证码是短信验证码，扰动处理可以是在短信中添加乱序字母或数字串等等。
在一示例性实施例中，获取添加扰动信息的验证码可以包括以下步骤：
获取预设文本，并将预设文本转换为目标图片；
通过对目标图片进行变形、变色、模糊化及增加噪点中的一种或多种处理，生成经过扰动处理的图像验证码。
图4示出了上述获取验证码的流程图。如图4所示，首先获取预设文本，预设文本可以从文本库中获取，也可以是生成的随机数等，将获得的预设文本转换为目标图片，目标图片可以是包含了验证码文本信息的图片，例如包含文本验证码字符的图片，或者图像验证码碎片的图片等。在转换目标图片时，可以生成验证码的艺术字，并嵌入到特定的背景中以得到目标图片；也可以如图4所示，将验证码的字符拆分，每个字符生成单字图片，然后打乱顺序并拼接为一幅目标图片。对于生成的目标图片，可以根据需要进行剪裁，然后进行扰动处理。其中，变形处理可以是将图片进行拉伸、旋转、液化等处理；变色处理是指改变目标图片的色阶参数，例如将目标图片转换为灰度图、调整明暗分布等；模糊化处理是指降低目标图片的分辨率，例如对目标图片进行局部像素压缩等；增加噪点处理可以是对目标图片添加影响辨认的颗粒等。另外，还可以在目标图片中随机添加干扰性元素，例如，在目标图片上添加横线或竖线等。本领域技术人员容易想到的其他干扰处理方式也应当包含在本公开的保护范围内。
在一示例性实施例中，在响应于待验证语音与验证码匹配之前，语音信息验证方法还可以包括以下步骤：
检测待验证语音的长度；
如果待验证语音的长度小于一预设长度，则判定待验证语音验证失败，并向用户终端返回失败提示信息；
如果待验证语音的长度大于等于预设长度，则将待验证语音转换为待验证文本，并将待验证文本与验证码进行匹配。
其中，待验证语音的长度可以是其时长，也可以是其文件大小等。考虑到过短的待验证语音可能是由误输入造成的，导致无意义的验证，影响服务器的工作效率，可以设置一个预设长度作为待验证语音长度的判断标准，以滤除上述情况，例如将预设长度设置为两秒，不足两秒的语音将被服务器自动过滤，并返回失败提示信息；又例如将预设长度设置为5KB，则小于5KB的待验证语音也将自动被服务器过滤等。
在检测到待验证语音的长度大于等于预设长度之后，可以将待验证语音转换为待验证文本，其中，待验证文本是与待验证语音内容对应的文本形式的待验证信息，可以利用语音文本转换工具实现转换。
上述预设长度可视为待验证语音的长度下限，需要补充的是，可能存在用户误输入或发送错误的录音文件，使得服务器接收的待验证语音过于复杂或冗长的情况，则可能造成服务器处理困难，增加无意义的工作量。因此，还可以设置待验证语音的长度上限，以滤除上述过于复杂或冗长的待验证语音。当语音长度超过语音长度上限时，可以输出验证失败的结果，返回失败提示信息。此外，可以在客户端程序内设置待验证语音的录制时间上限，当超过上限时自动终止语音的录入，并将已录入的语音发送到服务器以进行语音信息验证。
进一步的，将待验证语音转换为待验证文本可以通过以下步骤实现：
对待验证语音进行声道转换、预加重、语音增强、去除空白中的一种或多种预处理；
利用时延神经网络模型将预处理后的待验证语音转换为待验证文本。
考虑到服务器获取的待验证语音可能包含了其他噪声等影响因素，对待验证语音的处理和识别造成一定的障碍。在将待验证语音转换为待验证文本时，可以对待验证语音进行预处理。其中，声道转换处理是指如果所要提取的待验证语音的语音特征不区分声道，则可以将多声道的待验证语音转换为单声道；预加重处理可以是保留待验证语音一定频率范围内的语音信号，有利于服务器对待验证语音信息的分析；语音增强处理可以是从待验证语音中滤除噪音，以提取纯净的语音信号；去除空白处理是指去除待验证语音中无实际信号的片段，例如用户在输入待验证语音的过程中被打断或思考的情况下， 待验证语音可能存在噪声或其他空白的无效语音片段，去除空白处理也可以减小待验证语音的时长与文件大小，降低服务器的处理量。当外界噪声过大时，例如用户在人流量较多或受其他音响设备影响的情况下，对待验证语音进行处理之前，还可以通过语音活动检测来判断待验证语音中是否存在可以验证的语音信息，如果不存在则可以输出验证失败的结果。此外，本领域技术人员容易想到的其他预处理方式也应当包含在本公开的保护范围内。
在对待验证语音进行预处理后，可以将其输入到时延神经网络模型中，时延神经网络模型可以对待验证语音进行分帧识别，最终转换为对应的待验证文本。时延神经网络模型的训练和使用可以参考本领域技术人员熟知的方法，在此不再详述。
在一示例性实施例中，上述验证码可以包括文本验证码，将待验证文本与验证码进行匹配可以包括以下步骤：将待验证文本与文本验证码匹配，以得到待验证文本的错字比例。
相应的，步骤S220可以包括以下步骤：如果错字比例低于匹配阈值，则判断待验证语音是否为非机器。
其中，错字比例可以是未匹配成功的字符数占总字符数的百分比。匹配可以有多种方式，例如正向匹配、逆向匹配或双向匹配等。匹配结果可能由于选择的匹配方式不同而不同。例如：如果文本验证码为“今天天气真好”，待验证文本为“今天气真好”，采用正向匹配的方式，其匹配结果的错字比例将会高于采用双向匹配得到的错字比例。通常情况下，采用双向匹配的方式能够提高服务器匹配计算的准确率，但也增加了服务器的计算任务，对服务器的配置具有较高的要求。正向匹配及逆向匹配对服务器的计算能力要求较低，处理量较少。可以根据实际情况选择合适的匹配方式。
匹配阈值可以是设定的一个可接受的匹配错字比例的上限。考虑到服务器在将待验证语音转换待验证文本，以及匹配待验证文本与文本验证码时可能出现的误差，可以允许待验证文本与文本验证码之间存在一定的不一致。匹配阈值可以根据实际情况设定，例如验证码较短或干扰信息较少时，匹配阈值可以设置的较高，验证码较长或内容复杂时，匹配阈值可以设置的较低等，本实施例对此不做具体限定。当错字比例低于匹配阈值时，可以认为待验证文本与验证码匹配成功，也即待验证语音与验证码匹配成功，通过第一次验证，可以继续进行第二次验证，即对待验证语音进行人机判断。
图5示出了本示例性实施例中一种语音信息验证方法的流程图。如图5所示，获取 待验证语音后，可以先判断获取的待验证语音的长度是否达到预设长度，如果不足预设长度，可以直接输出验证失败的结果，如果达到预设长度，则可以对待验证语音进行预处理，并通过时延神经网络模型转换为待验证文本。然后对待验证文本和验证码进行匹配，判断错字比例是否低于匹配阈值，如果错字比例未低于匹配阈值，则可以输出验证失败的结果，如果错字比例低于匹配阈值，则可以将待验证语音转换为目标频谱图，以进行后续的验证。
在一示例性实施例中，将待验证语音转换为目标频谱图可以包括：通过短时傅里叶变换将待验证语音转换为目标频谱图。
短时傅里叶变换可以将复杂的声音信号转换到频域中，然后通过频谱特性分析待验证语音的时域信号特征。举例而言，可以从待验证语音中提取多个瞬时片段，并按照时序排列，每个片段转换为频率-能量图像，从而得到多个目标频谱图排成的序列，可以在后续利用卷积神经网络模型进行处理。
图6示出了本示例性实施例中一种语音信息验证方法的流程图。如图6所示，服务器可以由文本库获取预设文本并经过干扰处理得到验证码，由用户终端获取用户输入的待验证语音，并通过对待验证语音进行预处理后，利用时延神经网络模型将待验证语音转换为待验证文本，通过计算错字比例来匹配待验证文本和验证码。如果匹配成功，可以将待验证语音通过短时傅里叶变换转换为目标频谱图，进一步，再通过训练和测试完成的卷积神经网络模型对目标频谱图进行分析，得到人机分类结果，如果人机分类结果为非机器，则可以输出最终验证通过的结果。
本公开的实例性实施例还提供了一种语音信息验证装置。参考图7所示，该装置700可以包括：信息获取模块710，用于获取验证码以及用户终端发送的关于验证码的待验证语音；人机判断模块720，用于响应于待验证语音与验证码匹配，判断待验证语音是否为非机器；语音验证模块730，用于如果待验证语音为非机器，则确定待验证语音验证通过。
在一示例性实施例中，人机判断模块720可以包括：频谱图转换单元，用于如果待验证语音与验证码匹配，则将待验证语音转换为目标频谱图；频谱图分析单元，用于通过卷积神经网络模型对目标频谱图进行分析，得到目标频谱图的人机分类结果；语音判断单元，用于基于人机分类结果，确定待验证语音是否为非机器。
在一示例性实施例中，信息获取模块710还可以用于获取经过扰动处理的验证码。
在一示例性实施例中，信息获取模块710可以包括：文本获取单元，用于获取预设文本，并将预设文本转换为目标图片；图片处理单元，用于通过对目标图片进行变形、增加噪点、变色及模糊化中的一种或多种处理，生成经过扰动处理的图像验证码。
在一示例性实施例中，语音信息验证装置700还可以包括：语音长度检测单元，用于检测待验证语音的长度；预设长度判定单元，用于在待验证语音的长度小于一预设长度时，判定待验证语音验证失败，并向用户终端返回失败提示信息，以及在待验证语音的长度大于等于预设长度时，将待验证语音转换为待验证文本，并将待验证文本与验证码进行匹配。
在一示例性实施例中，预设长度判定单元还可以包括：预处理子单元，用于对待验证语音进行声道转换、预加重、语音增强、去除空白中的一种或多种预处理；模型处理子单元，用于利用时延神经网络模型将预处理后的待验证语音转换为待验证文本。
在一示例性实施例中，验证码可以包括文本验证码，预设长度判定单元还可以包括：文本匹配单元，用于将待验证文本与文本验证码匹配，以得到待验证文本的错字比例。在这种情况下，人机判断模块720可以用于如果错字比例低于匹配阈值，则判断待验证语音是否为非机器。
上述各模块/单元的具体细节已经在对应的方法部分实施例中进行了详细的描述，因此此处不再赘述。
本公开的示例性实施例还提供了一种能够实现上述方法的电子设备。
所属技术领域的技术人员能够理解，本公开的各个方面可以实现为系统、方法或程序产品。因此，本公开的各个方面可以具体实现为以下形式，即：完全的硬件实施方式、完全的软件实施方式(包括固件、微代码等)，或硬件和软件方面结合的实施方式，这里可以统称为“电路”、“模块”或“系统”。
下面参照图8来描述根据本公开的这种示例性实施例的电子设备800。图8显示的电子设备800仅仅是一个示例，不应对本公开实施例的功能和使用范围带来任何限制。
如图8所示，电子设备800以通用计算设备的形式表现。电子设备800的组件可以包括但不限于：上述至少一个处理器810、上述至少一个存储器820、连接不同系统组件(包括存储器820和处理器810)的总线830、显示器840。
其中，存储器820存储有程序代码，程序代码可以被处理器810执行，使得处理器810执行本说明书上述“示例性方法”部分中描述的根据本公开各种示例性实施方式的步 骤。例如，处理器810可以执行图2所示的步骤S210～S230等。
存储器820可以包括易失性存储器形式的可读介质，例如随机存取存储器(Random Access Memory，RAM)821和/或高速缓存存储器822，还可以进一步包括只读存储器(Read-Only，ROM)823。
存储器820还可以包括具有一组(至少一个)程序模块825的程序/实用工具824，这样的程序模块825包括但不限于：操作系统、一个或者多个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。
总线830可以为表示几类总线结构中的一种或多种，包括存储器总线或者存储器控制器、外围总线、图形加速端口、处理器或者使用多种总线结构中的任意总线结构的局域总线。
电子设备800也可以与一个或多个外部设备1000(例如键盘、指向设备、蓝牙设备等)通信，还可与一个或者多个使得用户能与该电子设备800交互的设备通信，和/或与使得该电子设备800能与一个或多个其它计算设备进行通信的任何设备(例如路由器、调制解调器等等)通信。这种通信可以通过输入/输出(Input/Output，I/O)接口850进行。并且，电子设备800还可以通过网络适配器860与一个或者多个网络(例如局域网(Local Area Network，LAN)，广域网(Wide Area Network，WAN)和/或公共网络，例如因特网)通信。如图8所示，网络适配器860通过总线830与电子设备800的其它模块通信。应当明白，尽管图中未示出，可以结合电子设备800使用其它硬件和/或软件模块，包括但不限于：微代码、设备驱动器、冗余处理器、外部磁盘驱动阵列、RAID系统、磁带驱动器以及数据备份存储系统等。
通过以上的实施方式的描述，本领域的技术人员易于理解，这里描述的示例实施方式可以通过软件实现，也可以通过软件结合必要的硬件的方式来实现。因此，根据本公开实施方式的技术方案可以以软件产品的形式体现出来，该软件产品可以存储在一个非易失性存储介质(可以是CD-ROM，U盘，移动硬盘等)中或网络上，包括若干指令以使得一台计算设备(可以是个人计算机、服务器、终端装置、或者网络设备等)执行根据本公开示例性实施例的方法。
本公开的示例性实施例还提供了一种计算机可读存储介质，其上存储有能够实现本说明书上述方法的程序产品。在一些可能的实施方式中，本公开的各个方面还可以实现为一种程序产品的形式，其包括程序代码，当程序产品在终端设备上运行时，程序代码 用于使终端设备执行本说明书上述“示例性方法”部分中描述的根据本公开各种示例性实施方式的步骤。
参考图9所示，描述了根据本公开的示例性实施例的用于实现上述方法的程序产品900，其可以采用便携式紧凑盘只读存储器(CD-ROM)并包括程序代码，并可以在终端设备，例如个人电脑上运行。然而，本公开的程序产品不限于此，在本文件中，可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。
程序产品可以采用一个或多个可读介质的任意组合。可读介质可以是可读信号介质或者可读存储介质。可读存储介质例如可以为但不限于电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。可读存储介质包括，但不限于：具有一个或多个导线的电连接、便携式盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(Erasable Programmable Read-only Memory，EPROM或闪存)、光纤、便携式紧凑盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。
计算机可读信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了可读程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。可读信号介质还可以是可读存储介质以外的任何可读介质，该可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。
可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于无线、有线、光缆、RF等等，或者上述的任意合适的组合。
可以以一种或多种程序设计语言的任意组合来编写用于执行本公开操作的程序代码，程序设计语言包括面向对象的程序设计语言—诸如Java、C++等，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算设备上执行、部分地在用户设备上执行、作为一个独立的软件包执行、部分在用户计算设备上部分在远程计算设备上执行、或者完全在远程计算设备或服务器上执行。在涉及远程计算设备的情形中，远程计算设备可以通过任意种类的网络，包括局域网(LAN)或广域网(WAN)，连接到用户计算设备，或者，可以连接到外部计算设备(例如利用因特网服务提供商来通过因特网连接)。
此外，上述附图仅是根据本公开示例性实施例的方法所包括的处理的示意性说明，而不是限制目的。易于理解，上述附图所示的处理并不表明或限制这些处理的时间顺序。另外，也易于理解，这些处理可以是例如在多个模块中同步或异步执行的。
应当注意，尽管在上文详细描述中提及了用于动作执行的设备的若干模块或者单元，但是这种划分并非强制性的。实际上，根据本公开的示例性实施例，上文描述的两个或更多模块或者单元的特征和功能可以在一个模块或者单元中具体化。反之，上文描述的一个模块或者单元的特征和功能可以进一步划分为由多个模块或者单元来具体化。
本领域技术人员在考虑说明书及实践这里公开的发明后，将容易想到本公开的其他实施例。本申请旨在涵盖本公开的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本公开的一般性原理并包括本公开未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的，本公开的真正范围和精神由权利要求指出。
应当理解的是，本公开并不局限于上面已经描述并在附图中示出的精确结构，并且可以在不脱离其范围进行各种修改和改变。本公开的范围仅由所附的权利要求来限。

Claims (11)

  1. 一种语音信息验证方法，包括：
    获取验证码以及用户终端发送的关于所述验证码的待验证语音；
    响应于所述待验证语音与所述验证码匹配，判断所述待验证语音是否为非机器；
    如果所述待验证语音为非机器，则确定所述待验证语音验证通过。
  2. 根据权利要求1所述的方法，其特征在于，判断所述待验证语音是否为非机器包括：
    将所述待验证语音转换为目标频谱图；
    通过卷积神经网络模型对所述目标频谱图进行分析，得到所述目标频谱图的人机分类结果；
    基于所述人机分类结果，确定所述待验证语音是否为非机器。
  3. 根据权利要求1所述的方法，其特征在于，获取所述验证码包括：
    获取经过扰动处理的验证码。
  4. 根据权利要求3所述的方法，其特征在于，获取所述经过扰动处理的验证码包括：
    获取预设文本；
    将所述预设文本转换为目标图片；
    通过对所述目标图片进行变形、变色、模糊化及增加噪点中的一种或多种处理，生成所述经过扰动处理的图像验证码。
  5. 根据权利要求1所述的方法，其特征在于，在响应于所述待验证语音与所述验证码匹配之前，所述方法还包括：
    检测所述待验证语音的长度；
    如果所述待验证语音的长度小于预设长度，则判定所述待验证语音验证失败，并向所述用户终端返回失败提示信息；
    如果所述待验证语音的长度大于等于所述预设长度，则将所述待验证语音转换为待验证文本，并将所述待验证文本与所述验证码进行匹配。
  6. 根据权利要求5所述的方法，其特征在于，将所述待验证语音转换为所述待验证文本包括：
    对所述待验证语音进行声道转换、预加重、语音增强、去除空白中的一种或多种预处理；
    利用时延神经网络模型将预处理后的所述待验证语音转换为所述待验证文本。
  7. 根据权利要求5所述的方法，其特征在于，所述验证码包括文本验证码，将所述待验证文本与所述验证码进行匹配包括：
    将所述待验证文本与所述文本验证码匹配，以得到所述待验证文本的错字比例；
    响应于所述待验证语音与所述验证码匹配，判断所述待验证语音是否为非机器包括：
    如果所述错字比例低于匹配阈值，则判断所述待验证语音是否为非机器。
  8. 根据权利要求2所述的方法，其特征在于，将所述待验证语音转换为所述目标频谱图包括：
    通过短时傅里叶变换将所述待验证语音转换为所述目标频谱图。
  9. 一种语音信息验证装置，包括：
    信息获取模块，用于获取验证码以及用户终端发送的关于所述验证码的待验证语音；
    人机判断模块，用于响应于所述待验证语音与所述验证码匹配，判断所述待验证语音是否为非机器；
    语音验证模块，用于如果所述待验证语音为非机器，则确定所述待验证语音验证通过。
  10. 一种电子设备，包括：
    处理器；以及
    存储器，用于存储所述处理器可执行的可执行指令；
    其中，当读取所述可执行指令时，所述处理器被促使执行权利要求1-8任一项所述的方法。
  11. 一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现权利要求1-8任一项所述的方法。
PCT/CN2019/120972 2018-12-13 2019-11-26 语音信息验证 WO2020119448A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811527293.3A CN109493872B (zh) 2018-12-13 2018-12-13 语音信息验证方法及装置、电子设备、存储介质
CN201811527293.3 2018-12-13

Publications (1)

Publication Number Publication Date
WO2020119448A1 true WO2020119448A1 (zh) 2020-06-18

Family

ID=65710079

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/120972 WO2020119448A1 (zh) 2018-12-13 2019-11-26 语音信息验证

Country Status (3)

Country Link
CN (1) CN109493872B (zh)
CA (1) CA3062359A1 (zh)
WO (1) WO2020119448A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112309404A (zh) * 2020-10-28 2021-02-02 平安科技（深圳）有限公司 机器语音的鉴别方法、装置、设备及存储介质
CN112927413A (zh) * 2021-01-20 2021-06-08 联仁健康医疗大数据科技股份有限公司 一种就医挂号方法、装置、设备及存储介质
CN115273859A (zh) * 2021-04-30 2022-11-01 清华大学 语音验证装置的安全性测试方法及装置

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109493872B (zh) * 2018-12-13 2021-12-14 北京三快在线科技有限公司 语音信息验证方法及装置、电子设备、存储介质
CN110689885B (zh) * 2019-09-18 2023-05-23 平安科技（深圳）有限公司 机器合成语音识别方法、装置、存储介质及电子设备
CN111128115B (zh) * 2019-12-11 2022-09-16 北京声智科技有限公司 信息验证方法、装置、电子设备及存储介质
CN111613230A (zh) * 2020-06-24 2020-09-01 泰康保险集团股份有限公司 声纹验证方法、装置、设备及存储介质
CN112308379A (zh) * 2020-09-30 2021-02-02 音数汇元(上海)智能科技有限公司 居家护理的服务订单评价方法、装置、设备和存储介质
CN112507316A (zh) * 2021-02-08 2021-03-16 北京远鉴信息技术有限公司 一种用户的验证方法、装置、可读存储介质及电子设备
CN113836509B (zh) * 2021-09-23 2024-03-01 百度在线网络技术（北京）有限公司 信息采集方法、装置、电子设备和存储介质
CN114422161B (zh) * 2021-11-08 2024-04-16 江苏鑫合易家信息技术有限责任公司 一种根据气象信息生成个性化情景验证码的方法和系统
CN114627881B (zh) * 2022-04-01 2022-10-04 上海财安金融服务集团股份有限公司 一种基于人工智能的语音呼叫处理方法及系统
CN115514550A (zh) * 2022-09-15 2022-12-23 中国电信股份有限公司 基于语音声调的交互验证方法、装置及电子设备
CN115565539B (zh) * 2022-11-21 2023-02-07 中网道科技集团股份有限公司 一种实现自助矫正终端防伪身份验证的数据处理方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104036780A (zh) * 2013-03-05 2014-09-10 阿里巴巴集团控股有限公司 一种人机识别方法及系统
CN104104664A (zh) * 2013-04-11 2014-10-15 腾讯科技（深圳）有限公司 对验证码进行验证的方法、服务器、客户端和系统
US20140307876A1 (en) * 2013-04-10 2014-10-16 Google Inc. Systems and Methods for Three-Dimensional Audio CAPTCHA
US20160072803A1 (en) * 2014-09-05 2016-03-10 Yahoo! Inc. System and method for authentication across devices
CN106330915A (zh) * 2016-08-25 2017-01-11 百度在线网络技术（北京）有限公司 语音验证处理方法及装置
CN108877813A (zh) * 2017-05-12 2018-11-23 阿里巴巴集团控股有限公司 人机识别的方法、装置和系统
CN109493872A (zh) * 2018-12-13 2019-03-19 北京三快在线科技有限公司 语音信息验证方法及装置、电子设备、存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8380503B2 (en) * 2008-06-23 2013-02-19 John Nicholas and Kristin Gross Trust System and method for generating challenge items for CAPTCHAs
KR101957277B1 (ko) * 2017-02-14 2019-03-12 윤종식 음성 인식을 이용한 코딩시스템 및 코딩방법
CN107147499A (zh) * 2017-05-17 2017-09-08 刘光明 使用语音输入进行验证的方法和系统

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104036780A (zh) * 2013-03-05 2014-09-10 阿里巴巴集团控股有限公司 一种人机识别方法及系统
US20140307876A1 (en) * 2013-04-10 2014-10-16 Google Inc. Systems and Methods for Three-Dimensional Audio CAPTCHA
CN104104664A (zh) * 2013-04-11 2014-10-15 腾讯科技（深圳）有限公司 对验证码进行验证的方法、服务器、客户端和系统
US20160072803A1 (en) * 2014-09-05 2016-03-10 Yahoo! Inc. System and method for authentication across devices
CN106330915A (zh) * 2016-08-25 2017-01-11 百度在线网络技术（北京）有限公司 语音验证处理方法及装置
CN108877813A (zh) * 2017-05-12 2018-11-23 阿里巴巴集团控股有限公司 人机识别的方法、装置和系统
CN109493872A (zh) * 2018-12-13 2019-03-19 北京三快在线科技有限公司 语音信息验证方法及装置、电子设备、存储介质

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112309404A (zh) * 2020-10-28 2021-02-02 平安科技（深圳）有限公司 机器语音的鉴别方法、装置、设备及存储介质
CN112309404B (zh) * 2020-10-28 2024-01-19 平安科技（深圳）有限公司 机器语音的鉴别方法、装置、设备及存储介质
CN112927413A (zh) * 2021-01-20 2021-06-08 联仁健康医疗大数据科技股份有限公司 一种就医挂号方法、装置、设备及存储介质
CN115273859A (zh) * 2021-04-30 2022-11-01 清华大学 语音验证装置的安全性测试方法及装置
CN115273859B (zh) * 2021-04-30 2024-05-28 清华大学 语音验证装置的安全性测试方法及装置

Also Published As

Publication number Publication date
CA3062359A1 (en) 2020-06-13
CN109493872B (zh) 2021-12-14
CN109493872A (zh) 2019-03-19

Similar Documents

Publication Publication Date Title
WO2020119448A1 (zh) 语音信息验证
US10957339B2 (en) Speaker recognition method and apparatus, computer device and computer-readable medium
WO2018166187A1 (zh) 服务器、身份验证方法、系统及计算机可读存储介质
US9300672B2 (en) Managing user access to query results
WO2021051544A1 (zh) 语音识别方法及其装置
CN110147726A (zh) 业务质检方法和装置、存储介质及电子装置
US20140379354A1 (en) Method, apparatus and system for payment validation
CN112949708B (zh) 情绪识别方法、装置、计算机设备和存储介质
CN109801638B (zh) 语音验证方法、装置、计算机设备及存储介质
US11335360B2 (en) Techniques to enhance transcript of speech with indications of speaker emotion
CN113177850A (zh) 一种保险的多方身份认证的方法及装置
JP2023549934A (ja) ユーザ認識のための方法及び機器
CN110516083B (zh) 相册管理方法、存储介质及电子设备
US11881224B2 (en) Multilingual speech recognition and translation method and related system for a conference which determines quantity of attendees according to their distances from their microphones
CN111400463A (zh) 对话响应方法、装置、设备和介质
CN112201254A (zh) 无感语音认证方法、装置、设备及存储介质
CN112087726A (zh) 彩铃识别的方法及系统、电子设备及存储介质
CN109887490A (zh) 用于识别语音的方法和装置
CN116030820A (zh) 音频验证方法及装置、音频取证方法及装置
CN112992155B (zh) 一种基于残差神经网络的远场语音说话人识别方法及装置
CN115206321A (zh) 语音关键词的识别方法、装置和电子设备
CN111899718A (zh) 用于识别合成语音的方法、装置、设备和介质
CN112116926A (zh) 音频数据的处理方法及装置、模型训练方法及装置
CN114003883A (zh) 一种便携式的数字化身份验证设备及身份验证方法
WO2021196458A1 (zh) 贷款智能进件方法、装置及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19895760

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19895760

Country of ref document: EP

Kind code of ref document: A1