Private number call processing method and device

Abstract

The invention belongs to the technical field of privacy communication, and provides a privacy number call processing method and device. The method comprises the following steps: when a call request initiated by a first user to a second user is received, distributing a private number for the first user, and dynamically binding the distributed private number with a real number of the first user; performing feature extraction on the collected voice signals to obtain voice signal features for establishing a training data set; and inputting the extracted voice signal features corresponding to the to-be-processed voice signal into a self-established voice recognition model to obtain a relative recognition result, further performing real-time translation, and in the real-time translation process, performing encryption processing on the to-be-processed voice signal and the text information corresponding to the to-be-processed voice signal and then sending the to-be-processed voice signal and the text information to a second user. According to the invention, customized privacy information protection of various users can be effectively realized.

Description

Technical Field

[0001] The present invention relates to the field of privacy communication technology, and in particular to a method and device for processing calls with a privacy number. Background Art

[0002] With the continuous development of communication technology, privacy protection has become an important issue that users are increasingly concerned about. The intermediate number service launched by China Unicom effectively protects users' real number information and avoids the risk of information leakage by providing a temporary, anonymous intermediate number for both parties of the call. However, in the application fields of mobile communications, social software, etc., while providing privacy protection, call quality and call security have also become the focus of users.

[0003] The existing technology has the following problems: the existing speech-to-speech conversion system is often limited in terms of real-time performance. Many systems need to convert speech into text first, then translate it, and then convert it back into speech. This process increases the delay and affects the fluency and real-time performance of the call; the language and dialect adaptation effect is poor. The existing system mainly focuses on the conversion of Mandarin or mainstream languages, and may not work well when dealing with local dialects. Especially in cases involving multiple dialects or mixed dialects, the language conversion quality and accuracy of the existing technology cannot meet actual needs; privacy protection is insufficient. Specifically, during voice calls, how to effectively protect user privacy while converting in real time is still a challenge. The existing system's protection measures for user identity information are usually not comprehensive enough, and there is a risk of information leakage.

[0004] Therefore, it is necessary to provide a method and device for processing private number calls. Summary of the invention

[0005] The present invention aims to provide a method and device for processing calls with a private number, so as to solve the technical problems in the prior art that the quality and accuracy of language conversion cannot meet actual needs, privacy protection is insufficient, there is a risk of information leakage, and it is impossible to effectively achieve real-time conversion while protecting user privacy. The technical problems to be solved by the present invention are achieved through the following technical solutions.

[0006] The first aspect of the present invention proposes a method for processing calls with a private number, comprising: when a call request is received from a first user to a second user, a private number is allocated to the first user, and the allocated private number is dynamically bound to the real number of the first user; voice signals of both parties of the call are collected, and features are extracted from the collected voice signals to obtain voice signal features for establishing a training data set, wherein the voice signal features include Mel-frequency cepstral coefficients and fundamental frequency; a voice recognition model is constructed according to a neural network algorithm, and the voice recognition model is trained using the established training data set to obtain a trained voice recognition model; feature extraction is performed on the voice signal to be processed, and the extracted voice signal features corresponding to the voice signal to be processed are input into the trained voice recognition model to obtain a recognition result corresponding to the voice signal to be processed, wherein the recognition result includes determining whether it is Mandarin or a dialect; real-time translation is performed according to the determined recognition result, and in the process of real-time translation, the voice signal to be processed and its corresponding text information are encrypted, specifically including determining the encryption strength and encrypting according to the determined encryption strength; the encrypted voice information is output and sent to the second user.

[0007] The second aspect of the present invention proposes a privacy number call processing device, which adopts the privacy number call processing method described in the first aspect of the present invention. The privacy number call processing device includes: a request processing module, when receiving a call request initiated by a first user to a second user, assigning a privacy number to the first user, and dynamically binding the assigned privacy number to the real number of the first user; a feature extraction module, for collecting voice signals of both parties of the call, and extracting features of the collected voice signals to obtain voice signal features for establishing a training data set, wherein the voice signal features include Mel-frequency cepstral coefficients and fundamental frequency; a model construction module, which constructs a speech recognition model according to a neural network algorithm, and uses the established training data to extract features of the voice signal. The speech recognition model is trained based on the data set to obtain a trained speech recognition model; a recognition determination module is used to extract features of the speech signal to be processed, and the extracted speech signal features corresponding to the speech signal to be processed are input into the trained speech recognition model to obtain a recognition result corresponding to the speech signal to be processed, and the recognition result includes determining whether it is Mandarin or a dialect; a real-time processing module performs real-time translation according to the determined recognition result, and in the process of real-time translation, encrypts the speech signal to be processed and its corresponding text information, specifically including determining the encryption strength and encrypting according to the determined encryption strength; a sending module is used to output the encrypted voice information and send it to a second user.

[0008] The third aspect of the present invention provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the privacy number call processing method described in the first aspect of the present invention.

[0009] A fourth aspect of the present invention provides a computer-readable medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for processing private number calls described in the first aspect of the present invention is implemented.

[0010] The embodiments of the present invention include the following advantages: Compared with the prior art, the present invention collects voice signals from both parties of the call and performs feature extraction on the collected voice signals to obtain voice signal features, so as to construct a voice recognition model, perform feature extraction on the voice signal to be processed, and input the extracted voice signal features corresponding to the voice signal to be processed into the trained voice recognition model to obtain a recognition result corresponding to the voice signal to be processed, which can more accurately identify the language type; perform real-time translation according to the determined recognition result, and in the process of real-time translation, output the voice information after encrypting the voice signal to be processed and its corresponding text information according to the encryption strength, and send it to the second user, which can effectively realize customized privacy information protection for various users.

[0011] In addition, by optimizing the end-to-end speech conversion process, specifically by integrating advanced speech recognition, speech synthesis and signal processing technologies, a more optimized end-to-end conversion process is created to minimize latency and improve conversion quality.

[0012] In addition, a dynamic privacy protection mechanism is implemented to allow authorized third-party services to access private number information when necessary while ensuring user privacy security. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flowchart of an example of a method for processing a private number call of the present invention; Figure 2 It is a schematic diagram of an application example of the privacy number call processing method of the present invention; Figure 3 It is a structural block diagram of the privacy number call processing device of the present invention; Figure 4 is a schematic structural diagram of an electronic device embodiment according to the present invention; Figure 5 It is a schematic diagram of the structure of a computer-readable medium embodiment according to the present invention. DETAILED DESCRIPTION

[0014] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0015] In view of the above problems, the present invention proposes a method and device for processing calls based on a privacy number. The method collects voice signals from both parties of the call, and extracts features from the collected voice signals to obtain voice signal features to build a voice recognition model, extracts features from the voice signals to be processed, and inputs the extracted voice signal features corresponding to the voice signals to be processed into the trained voice recognition model to obtain a recognition result corresponding to the voice signals to be processed, which can more accurately identify the language type; performs real-time translation according to the determined recognition results, and in the process of real-time translation, outputs the voice information after encryption processing of the voice signals to be processed and their corresponding text information according to the encryption strength, and sends it to the second user, which can effectively realize customized privacy information protection for various users.

[0016] In addition, by optimizing the end-to-end speech conversion process, specifically by integrating advanced speech recognition, speech synthesis and signal processing technologies, a more optimized end-to-end conversion process is created to minimize latency and improve conversion quality.

[0017] In addition, a dynamic privacy protection mechanism is implemented to allow authorized third-party services to access private number information when necessary while ensuring user privacy security.

[0018] Example 1 Refer to the following Figure 1 and Figure 2 , the contents of the present invention will be described in detail.

[0019] Figure 1 It is a step flow chart of an example of the private number call processing method of the present invention.

[0020] like Figure 1 As shown, in step S101, when a call request is received from a first user to a second user, a privacy number is allocated to the first user, and the allocated privacy number is dynamically bound to the real number of the first user.

[0021] exist Figure 2 In the application example, the privacy communication platform and the first user (corresponding to Figure 2 User 1 in the example) and the second user (corresponding to Figure 2 User 2 in ).

[0022] Specifically, the privacy communication platform includes a privacy number management module, a voice collection module, a voice recognition module, a translation module, and a voice synthesis module. The privacy management module includes a privacy number database, which is used to store various privacy numbers and allocate corresponding privacy numbers to various users based on call requests.

[0023] In an optional implementation, when the privacy number management module of the privacy communication platform receives a call request from the first user to the second user, the privacy number database of the privacy number management module allocates a corresponding privacy number to the first user.

[0024] Specifically, the allocated privacy number is dynamically bound to the real number of the first user.

[0025] For example, the allocated privacy number is dynamically bound to the real number of the first user according to the user ID and device ID. The specific binding relationship is dynamically changed, such as the AXB mode (A is the first user A, X is the privacy number, and B is the second user B). During the call, A and B do not know the real number of the other party. After the current call ends, the binding relationship is automatically released.

[0026] It should be noted that the above is only described as an optional example and should not be understood as a limitation to the present invention.

[0027] Next, in step S102, voice signals of both parties of the call are collected, and features are extracted from the collected voice signals to obtain voice signal features for establishing a training data set. The voice signal features include Mel-frequency cepstral coefficients and fundamental frequency.

[0028] Specifically, the voice acquisition module is used to collect the voice signals of both parties of the call and transmit them to the voice recognition module. For example, the voice signal sent by the first user to the second user through the privacy communication platform is collected, and the voice signal sent by the second user to the first user through the privacy communication platform is collected.

[0029] The preprocessed speech signal is subjected to feature extraction to obtain speech signal features such as Mel-frequency cepstral coefficients (MFCC), fundamental frequency (F0), etc. These speech information features will be used in the subsequent speech recognition process.

[0030] Optionally, the short-term waveform features of the speech signal are extracted through an autocorrelation function, specifically by calculating the similarity between the speech signal and itself at different delays.

[0031] By calculating the discrete degree of the signal value of the speech signal, the maximum value of the signal value of the speech signal is extracted to perform time domain feature extraction.

[0032] Optionally, the time domain signal is converted into a frequency domain signal by Fourier transform, i.e., FFT, so as to analyze the frequency component of the signal, or the array signal of the speech signal in the microphone array is processed by beam Fourier transform to extract frequency domain features.

[0033] Through linear prediction analysis, namely LPC, the characteristics of speech signals are extracted by minimizing the prediction error; and the perceptual linear prediction coefficient, namely PLP, is used to perform high-level feature extraction on speech signals.

[0034] Specifically, the features of speech signals are extracted by minimizing the prediction error based on the all-pole digital filter model through linear prediction analysis, i.e. LPC. Advanced feature extraction is performed through perceptual linear prediction coefficients, i.e. PLP, specifically extracting feature parameters based on the auditory model, which is particularly excellent in noise-resistant speech feature extraction.

[0035] It should be noted that the time domain feature extraction is performed through the autocorrelation function (used to describe the short-term waveform characteristics of the speech signal. The features are extracted by calculating the similarity between the speech signal and itself at different delays). In addition, the time domain feature extraction is also performed by calculating the variance (used to describe the waveform characteristics of the speech signal. The discreteness of the signal value of the speech signal is calculated to reflect the stability of the signal), the peak value (extracting the maximum value or local maximum value in the speech signal, which are usually related to the transient characteristics of the speech signal). In addition, in addition to using Fourier transform, that is, FFT (fast Fourier transform, which can efficiently calculate FFT), beam Fourier transform can also be used for frequency domain feature extraction. The beam Fourier transform is a special Fourier transform used to process array signals, such as speech signals in microphone arrays.

[0036] It should be noted that the above is only described as an optional example and should not be understood as a limitation to the present invention.

[0037] Next, in step S103, a speech recognition model is constructed according to a neural network algorithm, and the speech recognition model is trained using the established training data set to obtain a trained speech recognition model.

[0038] A speech recognition algorithm based on deep learning is adopted, specifically a recursive neural network (RNN), to build a speech recognition model.

[0039] The speech signal features extracted in step S102 are used to establish a training data set with speech signal features annotated with language types. In this example, the language types include dialect type and Mandarin type. However, other annotated types such as Chinese, English, and Japanese may also be included.

[0040] In the training dataset, X represents the input data, that is, the features of the speech signal after preprocessing and feature extraction. These features can be time domain features (such as short-time energy, short-time average zero-crossing rate, etc.), frequency domain features (such as spectrum, Mel-frequency cepstral coefficient MFCC, etc.) or other types of features (such as formant parameters, fundamental frequency, etc.). These features are extracted from the original speech signal and can reflect some important properties or patterns of the speech signal. Y represents the output data, that is, the label or target value corresponding to the input data X. In the context of speech signal processing, Y usually represents the language type or other related classification information of the speech signal. For example, if the training dataset is designed to recognize speech signals in different languages, Y may represent the label of the language category (such as Mandarin, dialects, English, French, and other annotation types).

[0041] Next, the speech recognition model is trained using the established training data set to obtain a trained speech recognition model.

[0042] It should be noted that in other implementations, convolutional neural networks (CNN) or Transformer algorithms are also used to establish a speech recognition model. The above are only optional implementations and should not be understood as limitations of the present invention.

[0043] Specifically, the mathematical function of the speech recognition algorithm of the speech recognition model is expressed as: Text=ASR(Feature) Among them, ASR represents the mathematical function of the speech recognition algorithm of the speech recognition model; Feature represents the extracted speech signal features; Text represents the recognized text information and language type. In this example, dialect judgment is regarded as a binary classification problem, and its mathematical function can be expressed as:

[0044] textDialect=Classifier(Feature) Among them, Classifier represents the language classifier, Feature represents the extracted speech signal features, and Dialect represents the recognition result (in this example, Mandarin or dialect).

[0045] In the process of model training, model parameter optimization and loss value optimization are two core links. Model parameters can be optimized by selecting appropriate optimization algorithms, adjusting learning rates, introducing batch normalization and regularization techniques, etc.; loss values ​​can be optimized by selecting appropriate loss functions, using cross-validation, continuously monitoring loss values, and adjusting hyperparameters, etc. These strategies will help improve the performance and stability of the model and make it better adapt to actual task requirements.

[0046] It should be noted that the above is only described as an optional example and should not be understood as a limitation to the present invention.

[0047] Next, in step S104, feature extraction is performed on the speech signal to be processed, and the extracted speech signal features corresponding to the speech signal to be processed are input into a trained speech recognition model to obtain a recognition result corresponding to the speech signal to be processed, wherein the recognition result includes determining whether it is Mandarin or a dialect.

[0048] Specifically, feature extraction is performed on the voice signal to be processed. In this example, it specifically refers to the voice signal sent by the first user to the second user through the privacy communication platform, and the voice signal sent by the second user to the first user through the privacy communication platform is collected.

[0049] It should be noted that, in this step, the feature extraction process of the speech signal adopts the same method as the feature extraction method of the speech signal in step S102, so the description of the same parts is omitted.

[0050] Furthermore, the extracted speech features are input into the trained speech recognition model to obtain the corresponding text information and language type (dialect or Mandarin).

[0051] During the recognition process, the privacy communication platform will determine whether the input speech is Mandarin or dialect based on the pre-trained dialect recognition model.

[0052] In an optional implementation, the following characteristic parameters that can reflect speech characteristics are specifically extracted from the preprocessed speech signal: Mel frequency cepstral coefficients, fundamental frequency. The extracted characteristic parameters are matched with the parameters in the model using a trained speech recognition model (e.g., an acoustic model) to obtain corresponding phonemes or word sequences, and decoding is performed based on the matching results of the acoustic model in combination with the language model to obtain the final dialect text information.

[0053] It should be noted that the above is only described as an optional example and should not be understood as a limitation to the present invention.

[0054] Next, in step S105, real-time translation is performed according to the determined recognition result. During the real-time translation process, the voice signal to be processed and its corresponding text information are encrypted, specifically including determining the encryption strength and performing encryption according to the determined encryption strength.

[0055] In a specific implementation, the translation module performs real-time translation according to the recognition result of step S104. During the real-time translation process, the transmitted voice signal and text information are encrypted.

[0056] The transmitted voice signal and text information can only be decrypted and accessed when it is ensured that the current user is an authorized user.

[0057] In the process of voice signal transmission, the user privacy protection is enhanced through the dynamic privacy protection mechanism. Specifically, a dynamic parameter α is used to indicate the strength of privacy protection. The value range of α is a value between [0,1]. The larger the value of α, the higher the privacy protection strength.

[0058] To determine the encryption strength, the following expression is constructed to calculate the encryption strength α, i.e., the dynamic parameter, based on the behavioral characteristics of the first user and the second user, the call duration, and the importance or sensitivity of the call content: α=w1*B+w2*log(D+1)+w3*C+w4*O Wherein, α represents the encryption strength, i.e., a dynamic parameter; B represents the influencing factor of the behavioral characteristics of the first user and the second user during the call, specifically a value between 0 and 1, wherein the closer B is to 1, the closer it is to the optimal behavior, and the closer B is to 0, the further it is from the optimal behavior; w1 represents the weight factor corresponding to the influencing factor of the behavioral characteristics; D represents the influencing factor of the call duration between the first user and the second user, the longer the call duration, the higher the encryption strength, log(D+1) is the logarithmic transformation of the duration D, which is used to process call duration data at different time scales; w2 represents the weight factor corresponding to the influencing factor of the call duration; C represents the influencing factor of the importance or sensitivity of the call content between the first user and the second user; w3 represents the weight factor corresponding to the importance or sensitivity; O represents other influencing factors affecting the encryption strength, specifically one or more influencing factors; w4 is the weight factor corresponding to other influencing factors, wherein, w1+w2+w3+w4=1, and w1>w2>w3>w4.

[0059] It should be noted that w1, w2, w3, and w4 are weight factors used to adjust the influence of each variable on the encryption strength. These weight factors should satisfy w1+w2+w3+w4=1 to ensure the rationality of the expression. The use of log(D+1) is a logarithmic transformation of the duration D, which is used to process call duration data at different time scales to effectively avoid the situation where D is meaningless in the above expression when D=0.

[0060] Furthermore, according to the calculated encryption strength α, the following parameters of the encryption algorithm are dynamically adjusted: key length, number of encryption rounds, to achieve different levels of privacy protection; The following expression is used to express the relationship between the key length K and the encryption strength α: K = K min +(α - αmin )*(K max -K min ) / ((α max - α min ) where K represents the key length, K min and K max are the minimum and maximum values of the key length respectively; α represents the encryption strength, α max and α min are the minimum and maximum values of the encryption strength α respectively; The relationship between the number of encryption rounds R and α is represented by the following expression: R = R min +(α - α min )*(R max -R min ) / ((α max - α min ) where R represents the number of encryption rounds, R min and R max are the minimum and maximum values of the number of encryption rounds R respectively; α represents the encryption strength, α max and α min are the minimum and maximum values of the encryption strength α respectively.

[0061] Optionally, the relationship between the encryption algorithm parameters (such as the key length K, the number of encryption rounds R, etc.) and the encryption strength α value is determined through experiments, simulations, or theoretical analyses.

[0062] Specifically, the following relationship has been determined through experiments: When the encryption strength α is relatively low, a shorter key length and a smaller number of encryption rounds are used.

[0063] Specifically, the range of α defined as "relatively low" is [0, α_low], where α_low is a positive number less than 1, representing a relatively low privacy protection requirement.

[0064] For the range of the key length K, a shortest key length K_min_low (corresponding to the minimum key length) is set, which is used when α = 0. As the encryption strength α increases, the key length K gradually increases, but before the encryption strength α reaches α_low, the key length K does not exceed a preset length K_max_low (corresponding to the maximum key length and satisfying K_min_low < K < K_max_low). Therefore, the range of the key length is [K_min_low, K_max_low].[[]END]]

[0065] For the range of the number of encryption rounds \(R\), a minimum number of encryption rounds \(R_{min\_low}\) (corresponding to the minimum number of encryption rounds) is set, and when \(\alpha = 0\), the number of encryption rounds \(R_{min\_low}\) is used. As the encryption strength \(\alpha\) increases, the number of encryption rounds \(R\) gradually increases, but before the encryption strength \(\alpha\) reaches \(\alpha_{low}\), the number of encryption rounds \(R\) does not exceed a preset number of rounds \(R_{max\_low}\) (corresponding to the maximum number of encryption rounds, and satisfying \(R_{min\_low}<R<R_{max\_low}\)). Therefore, the range of the number of encryption rounds is \([R_{min\_low}, R_{max\_low}]\).

[0066] When the encryption strength \(\alpha\) is high, a longer key length and a larger number of encryption rounds are used.

[0067] The range of the encryption strength \(\alpha\) defined as "high" is \([\alpha_{high}, 1]\), where \(\alpha_{high}\) is a number greater than 0 and less than 1, representing a relatively high privacy protection requirement.

[0068] For the range of the key length \(K\), a relatively long key length \(K_{min\_high}\) (corresponding to the minimum key length) is set as the starting point, and when \(\alpha=\alpha_{high}\), the key length \(K_{min\_high}\) is used. As the encryption strength \(\alpha\) increases, the key length \(K\) further increases until it reaches the preset key length \(K_{max}\) (corresponding to the maximum key length, and satisfying \(K_{min\_high}<K<K_{max}\)). Therefore, the range of the key length is \([K_{min\_high}, K_{max}]\).

[0069] For the range of the number of encryption rounds \(R\), a relatively large number of encryption rounds \(R_{min\_high}\) is set as the starting point, and when the encryption strength \(\alpha = \alpha_{high}\), this number of rounds is used. As the encryption strength \(\alpha\) increases, the number of encryption rounds \(R\) further increases until it reaches the preset number of encryption rounds \(R_{max}\) (corresponding to the maximum number of encryption rounds, and satisfying \(R_{min\_high}<R<R_{max}\)). Therefore, the range of the number of encryption rounds is \([R_{min\_high}, R_{max}]\).

[0070] In the actual application scenario, when determining the specific values of parameters such as \(K_{min\_low}\), \(K_{max\_low}\), \(K_{min\_high}\), \(K_{max}\), \(R_{min\_low}\), \(R_{max\_low}\), \(R_{min\_high}\) and \(R_{max}\), it is necessary to comprehensively consider security requirements, performance requirements and the characteristics of the encryption algorithm. The following is a specific numerical example based on a hypothetical scenario.

[0071] In a specific embodiment, for the range of key length K, when the encryption strength α is low, the minimum key length K_min_low is 128 bits. This is because a 128-bit key length can provide higher security in a symmetric encryption algorithm while maintaining good performance.

[0072] Specifically, before the encryption strength α reaches a certain lower threshold (such as 0.4), the maximum key length K_max_low is set to 256 bits. This length can provide higher security while maintaining performance.

[0073] When the encryption strength α is high, the minimum key length K_min_high is 256 bits. This means that when higher security is required, start with a 256-bit key length.

[0074] When the encryption strength α reaches the highest value (e.g., 1), the maximum key length K_max is 512 bits. A key length of 512 bits can provide extremely high security in the encryption algorithm.

[0075] For the range of the number of encryption rounds R, when the encryption strength α is low, the minimum number of encryption rounds R_min_low is 10. This number of rounds can maintain good encryption efficiency while ensuring basic security.

[0076] Before the encryption strength α reaches a lower threshold (such as 0.4), the maximum number of encryption rounds R_max_low is 20. This number of rounds can provide higher security while maintaining performance.

[0077] When the encryption strength α is high, the minimum number of encryption rounds R_min_high is 20. This means that when higher security is required, start with 20 rounds of encryption.

[0078] When the encryption strength α reaches the highest value (for example, 1), the maximum number of encryption rounds R_max is 30. Therefore, more encryption rounds can provide higher security.

[0079] It should be noted that the above values ​​are examples based on hypothetical scenarios and do not represent the only or best choice in actual applications. In actual applications, the specific values ​​of these parameters should be determined based on the security requirements, performance requirements, and characteristics of the encryption algorithm in the actual application scenario. In addition, these parameters may need to be regularly evaluated and updated based on the development of technology and changes in security threats.

[0080] In addition, when selecting encryption algorithms and parameters, best practices and security standards should be followed to ensure the overall security of the system. At the same time, security assessments and vulnerability scans should be performed regularly on the system to promptly discover and fix potential security risks.

[0081] By expressing the above relationships as mathematical functions or tables, the encryption strength α value can be calculated. In an optional implementation, during a call, the α value is calculated based on real-time calling behavior, call duration, call content, etc. This process may involve measurement and calculation of multiple variables, and one or more weighted summation steps.

[0082] Then, based on the calculated encryption strength, the transmitted voice signal and text information are encrypted to ensure that the call content receives an appropriate level of protection during transmission.

[0083] It should be noted that the above is only described as an optional example and should not be understood as a limitation to the present invention.

[0084] Next, in step S106, the encrypted voice information is output and sent to the second user.

[0085] Specifically, the encrypted voice information is output, for example, the voice synthesis module outputs the voice information after real-time translation and encryption, performs voice synthesis again and plays it to the second user (ie, the other party of the call).

[0086] For example, the text information corresponding to the dialect speech (ie, the dialect text Ta), the translation module converts the dialect text Ta into Mandarin text Tb, and then converts the Mandarin text into Mandarin speech Sb through the speech synthesis module, and sends the Mandarin speech to the second user, and the second user responds.

[0087] Specifically, when the second user receives the encrypted voice signal, when he responds to the first user with a voice signal, he performs feature extraction on the responded voice information, inputs the extracted voice signal features into the trained speech recognition model, determines whether it is Mandarin or dialect, and translates it in real time. After encryption, it is sent to the first user, thus completing a call interaction.

[0088] Further, the call interaction is repeatedly performed until the first user terminates the call or the second user terminates the call.

[0089] Preferably, the transmitted voice signal and text information can be decrypted and accessed only when it is ensured that the current user (eg, the second user) is an authorized user.

[0090] The voice data output through the privacy communication platform are all call data that have been processed for privacy protection. They can achieve accurate recognition and fluent output of dialects and Mandarin, while effectively protecting user privacy information, thereby solving the risk of user privacy information leakage.

[0091] In addition, the drawings are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the drawings do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0092] Compared with the prior art, the present invention collects voice signals from both parties of the call and performs feature extraction on the collected voice signals to obtain voice signal features, so as to construct a voice recognition model, perform feature extraction on the voice signal to be processed, and input the extracted voice signal features corresponding to the voice signal to be processed into the trained voice recognition model to obtain a recognition result corresponding to the voice signal to be processed, which can more accurately identify the language type; perform real-time translation according to the determined recognition result, and in the process of real-time translation, output the voice information after encrypting the voice signal to be processed and its corresponding text information according to the encryption strength, and send it to the second user, which can effectively realize customized privacy information protection for various users.

[0093] In addition, by optimizing the end-to-end speech conversion process, specifically by integrating advanced speech recognition, speech synthesis and signal processing technologies, a more optimized end-to-end conversion process is created to minimize latency and improve conversion quality.

[0094] In addition, a dynamic privacy protection mechanism is implemented to allow authorized third-party services to access private number information when necessary while ensuring user privacy security.

[0095] Example 2 The following are embodiments of the apparatus of the present invention, which can be used to implement the embodiments of the method of the present invention. For details not disclosed in the embodiments of the apparatus of the present invention, please refer to the embodiments of the method of the present invention.

[0096] Figure 3 It is a structural diagram of an example of a private number call processing device of the present invention.

[0097] Reference Figure 3 The second aspect of the present disclosure provides a privacy number call processing device, which adopts the privacy number call processing method described in the first aspect of the present invention.

[0098] Specifically, the privacy number call processing device 300 includes a request processing module 310, a feature extraction module 320, a model building module 330, an identification and determination module 340, a real-time processing module 350, and a sending module 360.

[0099] In a specific implementation, when the request processing module 310 receives a call request from the first user to the second user, it allocates a privacy number to the first user and dynamically binds the allocated privacy number to the real number of the first user. The feature extraction module 320 is used to collect the voice signals of both parties of the call, and extract the features of the collected voice signals to obtain voice signal features for establishing a training data set, wherein the voice signal features include Mel frequency cepstral coefficients and fundamental frequency. The model construction module 330 constructs a speech recognition model according to a neural network algorithm, and trains the speech recognition model using the established training data set to obtain a trained speech recognition model. The recognition determination module 340 is used to extract features of the voice signal to be processed, and input the extracted voice signal features corresponding to the voice signal to be processed into the trained voice recognition model to obtain a recognition result corresponding to the voice signal to be processed, wherein the recognition result includes determining whether it is Mandarin or a dialect. The real-time processing module 350 performs real-time translation according to the determined recognition result. In the process of real-time translation, the voice signal to be processed and its corresponding text information are encrypted, specifically including determining the encryption strength and performing encryption processing according to the determined encryption strength. The sending module 360 ​​is used to output the encrypted voice information and send it to the second user.

[0100] According to an optional implementation manner, determining the encryption strength includes: constructing the following expression to calculate the encryption strength α, i.e., the dynamic parameter, according to the behavioral characteristics of the first user and the second user, the call duration, and the importance or sensitivity of the call content: α=w1*B+w2*log(D+1)+w3*C+w4*O Wherein, α represents the encryption strength, i.e., a dynamic parameter; B represents the influencing factor of the behavioral characteristics of the first user and the second user during the call, specifically a value between 0 and 1, wherein the closer B is to 1, the closer it is to the optimal behavior, and the closer B is to 0, the further it is from the optimal behavior; w1 represents the weight factor corresponding to the influencing factor of the behavioral characteristics; D represents the influencing factor of the call duration between the first user and the second user, the longer the call duration, the higher the encryption strength, log(D+1) is the logarithmic transformation of the duration D, which is used to process call duration data at different time scales; w2 represents the weight factor corresponding to the influencing factor of the call duration; C represents the influencing factor of the importance or sensitivity of the call content between the first user and the second user; w3 represents the weight factor corresponding to the importance or sensitivity; O represents other influencing factors affecting the encryption strength, specifically one or more influencing factors; w4 is the weight factor corresponding to other influencing factors, wherein, w1+w2+w3+w4=1, and w1>w2>w3>w4.

[0101] According to an optional implementation, the following parameters of the encryption algorithm are dynamically adjusted according to the calculated encryption strength α: key length, number of encryption rounds, to achieve different levels of privacy protection.

[0102] The following expression is used to express the relationship between the key length K and the encryption strength α: K = K min +(α - α min )*(K max -K min ) / ((α max - α min ) Among them, K represents the key length, K min and K max are the minimum and maximum key lengths respectively; α represents the encryption strength, α max and α min are the minimum and maximum values ​​of encryption strength α, respectively; The following expression is used to express the relationship between the number of encryption rounds R and α: R = R min +(α - α min )*(R max -R min ) / ((α max - α min ) Among them, R represents the number of encryption rounds, R min and R max are the minimum and maximum values ​​of the number of encryption rounds R; α represents the encryption strength, α max and α min are the minimum and maximum values ​​of the encryption strength α respectively.

[0103] According to an optional implementation, the feature extraction of the collected speech signal to obtain the speech signal feature includes: extracting the maximum value of the signal value of the speech signal by calculating the discreteness of the signal value of the speech signal to extract the time domain feature; converting the time domain signal into a frequency domain signal by Fourier transform, i.e., FFT, so as to analyze the frequency component of the signal, or processing the array signal of the speech signal in the microphone array by beam Fourier transform to extract the frequency domain feature; extracting the feature of the speech signal by minimizing the prediction error through linear prediction analysis, i.e., LPC; and performing high-level feature extraction on the speech signal using perceptual linear prediction coefficient, i.e., PLP.

[0104] According to an optional implementation, after the second user receives the encrypted voice signal, when responding to the first user with a voice signal, feature extraction is performed on the responded voice information, the extracted voice signal features are input into a trained voice recognition model to determine whether it is Mandarin or a dialect, and real-time translation is performed. After encryption, it is sent to the first user, thereby completing a call interaction.

[0105] The call interaction is repeatedly performed until the first user terminates the call or the second user terminates the call.

[0106] According to an optional implementation, the real-time translation based on the determined recognition result includes: extracting the following characteristic parameters that can reflect the speech characteristics from the preprocessed speech signal: Mel frequency cepstral coefficients and fundamental frequency.

[0107] Using the trained acoustic model, the extracted feature parameters are matched with the parameters in the model to obtain the corresponding phonemes or word sequences. According to the matching results of the acoustic model, decoding is performed in combination with the language model to obtain the final dialect text information.

[0108] Compared with the prior art, the present invention collects voice signals from both parties of the call and performs feature extraction on the collected voice signals to obtain voice signal features, so as to construct a voice recognition model, perform feature extraction on the voice signal to be processed, and input the extracted voice signal features corresponding to the voice signal to be processed into the trained voice recognition model to obtain a recognition result corresponding to the voice signal to be processed, which can more accurately identify the language type; perform real-time translation according to the determined recognition result, and in the process of real-time translation, output the voice information after encrypting the voice signal to be processed and its corresponding text information according to the encryption strength, and send it to the second user, which can effectively realize customized privacy information protection for various users.

[0109] In addition, by optimizing the end-to-end speech conversion process, specifically by integrating advanced speech recognition, speech synthesis and signal processing technologies, a more optimized end-to-end conversion process is created to minimize latency and improve conversion quality.

[0110] In addition, a dynamic privacy protection mechanism is implemented to allow authorized third-party services to access private number information when necessary while ensuring user privacy security.

[0111] It should be noted that the privacy number call processing method executed by the privacy number call processing device in Example 2 is substantially the same as the privacy number call processing method in Example 1, and therefore, the description of the same parts is omitted.

[0112] Figure 4is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

[0113] like Figure 4 As shown, the electronic device is presented in the form of a general computing device. The processor may be one or more and work in coordination. The present invention does not exclude distributed processing, that is, the processor may be dispersed in different physical devices. The electronic device of the present invention is not limited to a single entity, but may also be the sum of multiple physical devices.

[0114] The memory stores a computer executable program, which is usually a machine-readable code. The computer-readable program can be executed by the processor to enable the electronic device to perform the method of the present invention, or at least part of the steps in the method.

[0115] The memory includes a volatile memory, such as a random access memory unit (RAM) and / or a cache memory unit, and may also be a non-volatile memory, such as a read-only memory unit (ROM).

[0116] Optionally, in this embodiment, the electronic device further includes an I / O interface, which is used for the electronic device to exchange data with an external device. The I / O interface can represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of a variety of bus structures.

[0117] It should be understood that Figure 4 The electronic device shown is only an example of the present invention, and the electronic device of the present invention may also include elements or components not shown in the above examples. For example, some electronic devices also include display units such as display screens, and some electronic devices also include human-computer interaction elements such as buttons, keyboards, etc. As long as the electronic device can execute the computer-readable program in the memory to implement the method of the present invention or at least part of the steps of the method, it can be considered as an electronic device covered by the present invention.

[0118] Through the above description of the implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by combining software with necessary hardware. Figure 5 As shown, the technical solution according to the embodiment of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes a number of commands to enable a computing device (which can be a personal computer, a server, or a network device, etc.) to execute the above method according to the embodiment of the present invention.

[0119] The software product may be 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, a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0120] The computer readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, device, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0121] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0122] The computer-readable medium carries one or more programs (eg, computer-executable programs). When the one or more programs are executed by a device, the computer-readable medium implements the method of the present disclosure.

[0123] Those skilled in the art will appreciate that the above modules can be distributed in the device according to the description of the embodiment, or can be changed accordingly and only used in one or more devices different from the embodiment. The modules of the above embodiments can be combined into one module, or further divided into multiple sub-modules.

[0124] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the exemplary embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several commands to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiment of the present invention.

[0125] The exemplary embodiments of the present invention are specifically shown and described above. It should be understood that the present invention is not limited to the detailed structure, configuration or implementation method described herein; on the contrary, the present invention is intended to cover various modifications and equivalent configurations included in the spirit and scope of the appended claims.

Claims

1. A method for processing private number calls, characterized in that: include: When receiving a call request from a first user to a second user, assigning a privacy number to the first user, and dynamically binding the assigned privacy number to the real number of the first user; Collecting voice signals from both parties of the call, and extracting features from the collected voice signals to obtain voice signal features for establishing a training data set, wherein the voice signal features include Mel frequency cepstral coefficients and fundamental frequency; According to the neural network algorithm, a speech recognition model is constructed, and the speech recognition model is trained using the established training data set to obtain a trained speech recognition model; Extracting features of the speech signal to be processed, and inputting the extracted speech signal features corresponding to the speech signal to be processed into a trained speech recognition model to obtain a recognition result corresponding to the speech signal to be processed, wherein the recognition result includes determining whether it is Mandarin or a dialect; Performing real-time translation according to the determined recognition result, and in the process of real-time translation, encrypting the speech signal to be processed and the corresponding text information, specifically including determining the encryption strength, and performing encryption according to the determined encryption strength; The encrypted voice information is output and sent to the second user.

2. The method for processing private number calls according to claim 1, characterized in that: Determining the encryption strength includes: constructing the following expression to calculate the encryption strength α, i.e., the dynamic parameter, according to the behavioral characteristics of both parties of the conversation between the first user and the second user, the duration of the conversation, and the importance or sensitivity of the conversation content: α=w1*B+w2*log(D+1)+w3*C+w4*O Wherein, α represents the encryption strength, i.e., a dynamic parameter; B represents the influencing factor of the behavioral characteristics of the first user and the second user during the call, specifically a value between 0 and 1, wherein the closer B is to 1, the closer it is to the optimal behavior, and the closer B is to 0, the further it is from the optimal behavior; w1 represents the weight factor corresponding to the influencing factor of the behavioral characteristics; D represents the influencing factor of the call duration between the first user and the second user, the longer the call duration, the higher the encryption strength, log(D+1) is the logarithmic transformation of the duration D, which is used to process call duration data at different time scales; w2 represents the weight factor corresponding to the influencing factor of the call duration; C represents the influencing factor of the importance or sensitivity of the call content between the first user and the second user; w3 represents the weight factor corresponding to the importance or sensitivity; O represents other influencing factors affecting the encryption strength, specifically one or more influencing factors; w4 is the weight factor corresponding to other influencing factors, wherein, w1+w2+w3+w4=1, and w1>w2>w3>w4.

3. The method for processing private number calls according to claim 2, characterized in that: include: According to the calculated encryption strength α, the following parameters of the encryption algorithm are dynamically adjusted: key length, number of encryption rounds, to achieve different levels of privacy protection; The following expression is used to express the relationship between the key length K and the encryption strength α: K = K min +(a - a min )*(K max -K min ) / ((a max - a min ) Among them, K represents the key length, K min and K max are the minimum and maximum key lengths respectively; α represents the encryption strength, α max and α min are the minimum and maximum values ​​of encryption strength α, respectively; The following expression is used to express the relationship between the number of encryption rounds R and α: R = R min +(a - a min )*(R max -R min ) / ((a max - a min ) Among them, R represents the number of encryption rounds, R min and R max are the minimum and maximum values ​​of the number of encryption rounds R; α represents the encryption strength, α max and α min are the minimum and maximum values ​​of the encryption strength α respectively.

4. The method for processing private number calls according to claim 1, characterized in that: The feature extraction of the collected speech signal to obtain the speech signal feature includes: By calculating the discrete degree of the signal value of the speech signal, the maximum value of the signal value of the speech signal is extracted to perform time domain feature extraction; The time domain signal is converted into a frequency domain signal by Fourier transform, i.e. FFT, so as to analyze the frequency components of the signal, or the array signal of the speech signal in the microphone array is processed by beam Fourier transform to extract frequency domain features; Through linear prediction analysis, namely LPC, the characteristics of speech signals are extracted by minimizing the prediction error; and the perceptual linear prediction coefficient, namely PLP, is used to perform high-level feature extraction on speech signals.

5. The method for processing private number calls according to claim 1, characterized in that: include: When the second user receives the encrypted voice signal, when responding to the first user, the voice information of the response is extracted, and the extracted voice signal features are input into the trained voice recognition model to determine whether it is Mandarin or dialect, and the voice signal is translated in real time, encrypted, and sent to the first user, thus completing a call interaction; The call interaction is repeatedly performed until the first user terminates the call or the second user terminates the call.

6. The method for processing private number calls according to claim 1, characterized in that: The real-time translation according to the determined recognition result includes: Extract the following characteristic parameters that can reflect the speech characteristics from the preprocessed speech signal: Mel frequency cepstral coefficient and fundamental frequency; Using the trained acoustic model, the extracted feature parameters are matched with the parameters in the model to obtain the corresponding phonemes or word sequences. According to the matching results of the acoustic model, decoding is performed in combination with the language model to obtain the final dialect text information.

7. A private number call processing device, characterized in that: It executes the private number call processing method according to any one of claims 1 to 6, and the private number call processing device comprises: a request processing module, which, when receiving a call request from a first user to a second user, allocates a privacy number to the first user and dynamically binds the allocated privacy number to the real number of the first user; A feature extraction module is used to collect voice signals from both parties of the call and perform feature extraction on the collected voice signals to obtain voice signal features for establishing a training data set. The voice signal features include Mel frequency cepstral coefficients and fundamental frequency. A model building module, which builds a speech recognition model according to a neural network algorithm, and trains the speech recognition model using the established training data set to obtain a trained speech recognition model; The recognition and determination module is used to extract features of the speech signal to be processed, and input the extracted speech signal features corresponding to the speech signal to be processed into the trained speech recognition model to obtain a recognition result corresponding to the speech signal to be processed, wherein the recognition result includes determining whether it is Mandarin or dialect; A real-time processing module performs real-time translation according to the determined recognition result. During the real-time translation process, the voice signal to be processed and the corresponding text information are encrypted, specifically including determining the encryption strength and performing encryption according to the determined encryption strength; The sending module is used to output the encrypted voice information and send it to the second user.

8. The private number call processing device according to claim 7, characterized in that: Determining the encryption strength includes: constructing the following expression to calculate the encryption strength α, i.e., the dynamic parameter, according to the behavioral characteristics of both parties of the conversation between the first user and the second user, the duration of the conversation, and the importance or sensitivity of the conversation content: α=w1*B+w2*log(D+1)+w3*C+w4*O Wherein, α represents the encryption strength, i.e., a dynamic parameter; B represents the influencing factor of the behavioral characteristics of the first user and the second user during the call, specifically a value between 0 and 1, wherein the closer B is to 1, the closer it is to the optimal behavior, and the closer B is to 0, the further it is from the optimal behavior; w1 represents the weight factor corresponding to the influencing factor of the behavioral characteristics; D represents the influencing factor of the call duration between the first user and the second user, the longer the call duration, the higher the encryption strength, log(D+1) is the logarithmic transformation of the duration D, which is used to process call duration data at different time scales; w2 represents the weight factor corresponding to the influencing factor of the call duration; C represents the influencing factor of the importance or sensitivity of the call content between the first user and the second user; w3 represents the weight factor corresponding to the importance or sensitivity; O represents other influencing factors affecting the encryption strength, specifically one or more influencing factors; w4 is the weight factor corresponding to other influencing factors, wherein, w1+w2+w3+w4=1, and w1>w2>w3>w4.

9. The basic private number call processing device according to claim 8, characterized in that: include: According to the calculated encryption strength α, the following parameters of the encryption algorithm are dynamically adjusted: key length, number of encryption rounds, to achieve different levels of privacy protection; The following expression is used to express the relationship between the key length K and the encryption strength α: K = K min +(a - a min )*(K max -K min ) / ((a max - a min ) Among them, K represents the key length, K min and K max are the minimum and maximum key lengths respectively; α represents the encryption strength, α max and α min are the minimum and maximum values ​​of encryption strength α, respectively; The following expression is used to express the relationship between the number of encryption rounds R and α: R = R min +(a - a min )*(R max -R min ) / ((a max - a min ) Among them, R represents the number of encryption rounds, R min and R max are the minimum and maximum values ​​of the number of encryption rounds R; α represents the encryption strength, α max and α min are the minimum and maximum values ​​of the encryption strength α respectively.

10. The basic private number call processing device according to claim 8, characterized in that: include: By calculating the discrete degree of the signal value of the speech signal, the maximum value of the signal value of the speech signal is extracted to perform time domain feature extraction; The time domain signal is converted into a frequency domain signal by Fourier transform, i.e. FFT, so as to analyze the frequency components of the signal, or the array signal of the speech signal in the microphone array is processed by beam Fourier transform to extract frequency domain features; Through linear prediction analysis, namely LPC, the characteristics of speech signals are extracted by minimizing the prediction error; using perceptual linear prediction coefficients, namely PLP, high-level feature extraction of speech signals is performed.

Images