Content pushing method and device, electronic equipment and computer readable storage medium

Abstract

The embodiment of the invention discloses a content pushing method and device, electronic equipment and a computer readable storage medium, and belongs to the technical field of content pushing. According to the embodiment of the invention, feature extraction is performed on the first voice information of the target user to obtain the first feature information and the second feature information, and comprehensive analysis is performed on the first feature information and the second feature information, so that the accuracy of the target classification result can be remarkably improved, and then proper content is pushed according to the target classification result. And the content access security of family members is effectively ensured while the use experience of the user is improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of content pushing, and in particular to a content pushing method and device, electronic equipment and computer readable storage medium. BACKGROUND

[0002] An intelligent television is usually used as a shared device in a family to provide entertainment and information pushing services for all family members. The intelligent television is usually configured with a voice assistant, and a user can use the voice assistant to order content.

[0003] However, the current intelligent television usually only performs semantic recognition on a user voice and pushes related content according to the recognized user semantics. This method cannot effectively distinguish between family members, resulting in a lack of personalization and adaptability of the content pushed by the television and problems such as content security. SUMMARY

[0004] Embodiments of the present application provide a content pushing method and device, electronic equipment and computer readable storage medium, which can accurately identify a user and improve the accuracy and adaptability of content recommendation.

[0005] In a first aspect, embodiments of the present application provide a content pushing method, comprising: obtaining first voice information of a target user; performing feature extraction on the first voice information to obtain first feature information and second feature information; determining a target classification result of the target user based on the first feature information and the second feature information; performing content pushing on the target user based on the target classification result.

[0006] In one embodiment, the feature extraction on the first voice information to obtain the first feature information and the second feature information comprises: performing preprocessing on the first voice information to obtain second voice information; the preprocessing comprises one or more of denoising processing, normalization processing, frame processing and windowing processing; performing first feature extraction on the second voice information to obtain the first feature information; the first feature information comprises one or more of pitch, fundamental frequency, formant, mel-frequency cepstral coefficient and zero-crossing rate; performing second feature extraction on the second voice information to obtain the second feature information; the second feature information comprises one or more of fundamental frequency cycle perturbation parameter, amplitude perturbation parameter and recurrence quantification analysis parameter.

[0007] In one of the embodiments, the determining the target classification result of the target user based on the first feature information and the second feature information comprises: identifying the first feature information to obtain first identification information; identifying the second feature information to obtain second identification information; obtaining the target classification result of the target user based on the first identification information and the second identification information.

[0008] In one of the embodiments, the identifying the first feature information to obtain first identification information comprises: inputting the first feature information into a first identification model to obtain the first identification information output by the first identification model; the identifying the second feature information to obtain second identification information comprises: inputting the second feature information into a second identification model to obtain the second identification information output by the second identification model.

[0009] In one of the embodiments, the first identification information and the second identification information are respectively used to represent the confidence degree of the target user being a minor; the first identification model is obtained based on a first training data set, and the first training data set comprises a first minor voice data set and an adult voice data set; the second identification model is obtained based on a second training data set, and the second training data set is obtained in the following manner: extracting a second minor voice data set corresponding to minors in a target age range from the first minor voice data set; obtaining the second training data set based on the second minor voice data set and the adult voice data set.

[0010] In one of the embodiments, the obtaining the target classification result of the target user based on the first identification information and the second identification information comprises: performing weighted calculation on the first identification information and the second identification information to obtain a target confidence degree; in a case where the target confidence degree is greater than a confidence degree threshold, determining that the target classification result of the target user is a first classification result; the first classification result represents that the target user is a minor; in a case where the target confidence degree is less than or equal to the confidence degree threshold, determining that the target classification result of the target user is a second classification result; the second classification result represents that the target user is an adult.

[0011] In one of the embodiments, the content pushing to the target user based on the target classification result comprises: In a case where the target classification result is a first classification result, the first classification result is broadcast to a target application, and / or a system mode is set to a minor mode, so that the target application filters to-be-pushed content and pushes the filtered to-be-pushed content to the target user in a case where the first classification result is listened to and / or the system mode is detected as the minor mode; wherein the first classification result represents that the target user is a minor.

[0012] In a second aspect, the embodiments of the present application provide a content pushing device, the device comprising: an information acquisition module configured to acquire first voice information of a target user; a feature extraction module configured to perform feature extraction on the first voice information to obtain first feature information and second feature information; a result determination module configured to determine a target classification result of the target user based on the first feature information and the second feature information; a content pushing module configured to perform content pushing to the target user based on the target classification result.

[0013] In one of the embodiments, the feature extraction module comprises: a preprocessing submodule configured to perform preprocessing on the first voice information to obtain second voice information; the preprocessing comprises one or more of denoising processing, normalization processing, frame processing, and windowing processing; a first feature extraction submodule configured to perform first feature extraction on the second voice information to obtain the first feature information; the first feature information comprises one or more of pitch, fundamental frequency, formant, mel-frequency cepstral coefficient, and zero-crossing rate; a second feature extraction submodule configured to perform second feature extraction on the second voice information to obtain the second feature information; the second feature information comprises one or more of fundamental frequency cycle perturbation parameter, amplitude perturbation parameter, and recurrence quantification analysis parameter.

[0014] In one of the embodiments, the result determination module comprises: a first identification submodule configured to identify the first feature information to obtain first identification information; a second identification submodule configured to identify the second feature information to obtain second identification information; a result determination submodule configured to obtain the target classification result of the target user based on the first identification information and the second identification information.

[0015] In one of the embodiments, the first identification sub-module is specifically configured to input the first feature information into a first identification model to obtain the first identification information output by the first identification model. The second identification sub-module is specifically configured to input the second feature information into a second identification model to obtain the second identification information output by the second identification model.

[0016] In one of the embodiments, the first identification information and the second identification information are respectively used to represent the confidence that the target user is a minor. The first identification model is obtained based on a first training data set, and the first training data set includes a first minor voice data set and an adult voice data set. The second identification model is obtained based on a second training data set, and the second training data set is obtained in the following manner: In the first minor voice data set, a second minor voice data set corresponding to minors in a target age range is extracted; The second training data set is obtained based on the second minor voice data set and the adult voice data set.

[0017] In one of the embodiments, the result determination sub-module includes: a weighting calculation unit configured to perform weighting calculation on the first identification information and the second identification information to obtain a target confidence; a first classification unit configured to determine a target classification result of the target user as a first classification result in a case where the target confidence is greater than a confidence threshold; the first classification result represents that the target user is a minor; a second classification unit configured to determine the target classification result of the target user as a second classification result in a case where the target confidence is less than or equal to the confidence threshold; the second classification result represents that the target user is an adult.

[0018] In one of the embodiments, the content pushing module includes: a first content pushing sub-module configured to broadcast the first classification result to a target application and / or set a system mode to a minor mode in a case where the target classification result is the first classification result, so that the target application filters to-be-pushed content and pushes the filtered to-be-pushed content to the target user in a case where the first classification result is listened to and / or the system mode is detected as the minor mode; the first classification result represents that the target user is a minor.

[0019] Thirdly, embodiments of this application also provide an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps in the above-described content push method.

[0020] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the above-described content push method.

[0021] Fifthly, embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described in the embodiments of this application.

[0022] In summary, in this embodiment, by acquiring the first voice information of the target user and extracting features from it to obtain first feature information and second feature information, the target user's target classification result can be determined based on the first and second feature information. Furthermore, content can be pushed to the target user based on the target classification result. Thus, by comprehensively analyzing the first and second feature information, the accuracy of user identification can be significantly improved, allowing for the push of appropriate content to different users. This enhances the user experience while effectively ensuring the content access security of family members. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram illustrating the steps of a content push method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a content push device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0026] It's important to note that smart TVs, typically shared home devices, usually need to push content to different users. However, current smart TVs generally only push content based on user semantics, failing to effectively differentiate between users. This deficiency results in a lack of personalization and adaptability in TV content delivery, not only degrading the user's viewing experience but also potentially triggering privacy leaks and content security issues.

[0027] For example, when a teenager uses a TV voice assistant to request a program or movie (such as searching for "dinosaur movie"), the existing system will push relevant content based on semantic analysis. However, because it cannot identify the user as a teenager, the system may recommend or play resources containing gore and violence, affecting the teenager's mental health and behavioral habits. Furthermore, in multi-member family environments, once one user completes their task and leaves, others cannot receive personalized recommendations, further reducing the system's convenience and security.

[0028] To address the current problem of ineffective user differentiation leading to a lack of personalization and adaptability in TV content delivery, this application aims to provide a content delivery method. By extracting multi-dimensional features from the target user's first voice information to obtain first and second feature information, and then comprehensively analyzing the first and second feature information, the accuracy of target classification results can be significantly improved. Based on the target classification results, appropriate content can be delivered, enhancing the user experience while effectively ensuring the content access security of family members.

[0029] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the priority of the embodiments.

[0030] Figure 1 A flowchart illustrating a content push method according to an embodiment of this application is shown. The entity executing this content push method may be a content push device, which may be integrated into an electronic device, such as a server or a terminal.

[0031] The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), as well as big data and artificial intelligence platforms.

[0032] The terminal can be a smart TV, smartphone, smart speaker, tablet computer, laptop computer, desktop computer, washing machine, etc., but is not limited to these. The terminal and the server can be connected directly or indirectly through wired or wireless communication, which is not limited herein.

[0033] Furthermore, in the embodiments of this application, "multiple" refers to two or more. The terms "first" and "second," etc., in the embodiments of this application are used for distinguishing descriptions and should not be construed as implying relative importance.

[0034] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.

[0035] In this embodiment, the description will be from the perspective of a content push device, which can be integrated into a server or terminal. To facilitate the explanation of the content push method of this application, the following will describe the content push device integrated into a television in detail, that is, the television will be used as the execution subject for detailed explanation.

[0036] Reference Figure 1 This application illustrates a content push method, which may specifically include the following steps: S101: Obtain the target user's first voice information.

[0037] In this embodiment, the television has a voice interaction function. The television can obtain the target user's first voice information through an integrated voice module, or through a voice remote control connected to the television. The first voice information can be voice information used to request multimedia resources, including television programs, music, dramas, movies, etc. For example, the first voice information could be "I want to hear a story about dinosaurs" or "Play a dinosaur movie."

[0038] S102: Extract features from the first speech information to obtain first feature information and second feature information.

[0039] In this embodiment, multi-dimensional feature extraction can be performed on the first speech information to extract multi-dimensional feature information. Specifically, the multi-dimensional feature information may include first feature information and second feature information.

[0040] In this embodiment, the first feature information represents the basic acoustic feature information of the target user, which can reflect the macroscopic acoustic attributes of the target user, such as the physical and spectral information of the first speech information; the second feature information represents the medical acoustic features of the target user, which can reflect the microscopic physiological information of the target user, such as the vocal cord state and vocalization stability of the target user.

[0041] In this embodiment, considering that the recognition accuracy of the first feature information is limited in certain scenarios, especially in the case of voice change or blurred voice feature boundaries, if only basic acoustic feature information is considered, it will be impossible to accurately identify the target user. For example, it is easy to identify a minor with a voice changer as an adult. Therefore, by recognizing the second feature information of the first voice information, deeper physiological information, such as vocal cord maturity and stability, can be provided, thereby providing a more granular basis for physiological differentiation and reducing the risk of misclassification.

[0042] S103: Based on the first feature information and the second feature information, determine the target classification result of the target user.

[0043] In this embodiment, the target classification result may include one or more of the target user's gender, age, and user type. The user type can be determined based on the user's age group, with different age groups corresponding to different user types.

[0044] For example, referring to Table 1, user types can include both minors and adults.

[0045] Table 1 Example Table of User Types

[0046] For example, referring to Table 2, user types can also include various types such as children, teenagers, adolescents, adults, and the elderly.

[0047] Table 2. Example Table of User Types (Part Two)

[0048] In this embodiment, the number of user types can be set according to the recognition accuracy requirements; this embodiment does not impose any limitations.

[0049] In this embodiment, by comprehensively analyzing the first feature information and the second feature information, the recognition accuracy of the target classification result can be significantly improved, especially when the recognition accuracy of the voices of teenagers and adults overlaps in basic acoustic features.

[0050] S104: Based on the target classification results, push content to the target users.

[0051] In this embodiment, the television is equipped with a mapping table that represents the correspondence between classification results and push modes. After the target classification result is identified, the target push mode corresponding to the target classification result can be determined by querying the mapping table, and then content can be pushed to the target user according to the target push mode.

[0052] For example, when the target classification result is the first classification result, which indicates that the target user is a minor, content can be pushed to the target user in the minor mode, thereby automatically blocking sensitive resources and protecting the content of minors; when the target classification result is the second classification result, which indicates that the target user is an adult, content can be pushed to the target user in the adult mode, and the resource blocking can be removed to meet the personalized usage needs of adults.

[0053] In one example, a child in the family says "Play a dinosaur movie" to the TV. Based on this voice message, the TV recognizes that the target user is a minor and automatically switches to minor mode, blocking dinosaur movies with bloody and violent elements, and recommending children's cartoon movies related to dinosaurs to the target user. Subsequently, an adult in the family says the same voice message to the TV. Based on this voice message, the TV recognizes that the target user is an adult and automatically switches to adult mode, recommending live-action movies related to dinosaurs to the target user.

[0054] In this embodiment, by extracting multi-dimensional features from the target user's first voice information to obtain first feature information and second feature information, and by comprehensively analyzing the first feature information and second feature information, the accuracy of the target classification result can be significantly improved. Then, appropriate content can be pushed according to the target classification result, which can improve the user experience while effectively ensuring the content access security of family members.

[0055] In one feasible implementation, the step of extracting features from the first speech information to obtain first feature information and second feature information may specifically include the following sub-steps: S102-1: Preprocess the first speech information to obtain the second speech information.

[0056] In this embodiment, preprocessing includes one or more of the following: denoising, normalization, frame splitting, and windowing.

[0057] In this embodiment, the denoising process may specifically include: filtering the first speech information using a high-pass filter and a low-pass filter to obtain denoised first speech information. The high-pass filter allows frequencies in the first speech information above a first frequency threshold to pass through, thereby filtering out low-frequency noise; the low-pass filter allows frequencies in the first speech information below a second frequency threshold to pass through, thereby filtering out high-frequency noise; wherein the second frequency threshold is greater than the first frequency threshold.

[0058] In practical implementation, the first speech information can be denoised using the following formula: y1[n]=x[n]×h[n](1; Where x[n] represents the nth sampling point of the audio signal corresponding to the first speech information, h[n] represents the filter kernel (high-pass or low-pass), and y1[n] represents the nth sampling point of the denoised audio signal.

[0059] In this embodiment, noise reduction processing can eliminate or suppress background noise (such as wind noise, electromagnetic interference, and environmental noise) in the original audio signal, thereby improving the signal-to-noise ratio and ensuring the accuracy of subsequent feature extraction.

[0060] In this embodiment, the normalization process may specifically include: traversing the denoised audio signal to obtain the maximum amplitude value; and scaling the denoised audio signal proportionally based on the maximum amplitude value to obtain the normalized audio signal.

[0061] In practical implementation, the denoised audio signal can be normalized using the following formula: y2[n]=y1[n] / max(|x[n]|)(2); Where max(|x[n]| represents the maximum amplitude, y1[n] represents the nth sampling point of the denoised audio signal, and y2[n] is the normalized audio signal.

[0062] In this embodiment, by performing normalization processing, the signal amplitude difference caused by the difference in speaker volume or the difference in recording device gain can be eliminated, and the audio signal amplitude can be scaled to a uniform standard range to avoid the amplitude value affecting the recognition accuracy.

[0063] In this embodiment, the framing process may specifically include: dividing and overlapping the normalized audio signal according to a preset frame length and frame shift to obtain a framed audio signal. Here, the frame length represents the set time length of each frame, and the frame shift represents the set overlap between two adjacent frames.

[0064] In this embodiment, by dividing a long-term continuous non-stationary speech signal into multiple short-term quasi-stationary signal frames, it is convenient to perform stable local feature analysis within each frame.

[0065] In this embodiment, the windowing process may specifically include: multiplying each frame of the segmented audio signal point by point using a Hamming window function to obtain a windowed audio signal, which can be used as the final second speech information.

[0066] In practical implementation, the windowing process for the framed audio signal can be performed according to the following formula: X_windowed[n]=x i [n]×w[n](3; w[n]=0.54-0.46×cos((2π×n) / (N-1))(4); Where, x i [n] represents the nth sampling point of the i-th frame of the segmented audio signal, w[n] represents the Hamming window function, X_windowed[n] represents the windowed audio signal, n represents the sampling point index within the Hamming window function (0≤n≤N-1), and N is the total number of sampling points in a frame (N=sampling rate*frame length).

[0067] In this embodiment, by multiplying each frame signal by a Hamming window function, the discontinuity (spectral leakage) effect at both ends of the signal caused by frame truncation can be effectively reduced, so that the two ends of the frame smoothly transition to zero, thereby improving the quality of subsequent frequency domain transformation.

[0068] In this embodiment, by performing noise reduction, normalization, frame segmentation, and windowing on the first speech information, the data quality can be effectively improved, thereby enhancing the subsequent feature recognition effect.

[0069] S102-2: Extract the first feature from the second speech information to obtain the first feature information.

[0070] In this embodiment, the first feature information includes one or more of pitch, fundamental frequency, formant, Mel frequency cepstral coefficient, and zero crossover rate.

[0071] It should be noted that pitch is the primary frequency of human sound, usually corresponding to the periodic frequency of vocal cord vibration. The pitch range of teenagers and adults usually differs, with teenagers generally having higher pitches.

[0072] In this embodiment, the pitch can be extracted in the following way: calculate the autocorrelation function R[τ] on the second speech information to find the periodicity of the signal; then, in the autocorrelation function R[τ], exclude the peak at zero delay, find the delay value corresponding to the first significant peak, and determine the delay value as the pitch period T0; determine the reciprocal of the pitch period T0 as the pitch F0_pitch.

[0073] It should be noted that the fundamental frequency is the strongest frequency component in an audio signal and can be used as a physical quantity to supplement pitch, enhancing the robustness of the characteristics.

[0074] In this embodiment, the fundamental frequency can be extracted in the following way: a fast Fourier transform can be performed on the second speech information to convert it from the time domain to the frequency domain, resulting in a spectrum X[k]. In the spectrum X[k], the frequency value F0_spectral corresponding to the spectral peak with the largest amplitude is found, and this frequency value F0_spectral is determined as the fundamental frequency.

[0075] It should be noted that formants are the resonant frequencies of the vocal tract, which can reflect characteristics such as the speaker's age and gender. The formant distribution of adolescents is usually different from that of adults.

[0076] In this embodiment, the formants can be extracted in the following way: perform linear predictive coding (LPC) analysis on the second speech information to obtain the prediction signal X_hat[n], calculate the LPC coefficients by the error between the second speech information and the prediction signal X_hat[n], construct a polynomial from the LPC coefficients and solve for its roots, or calculate the LPC spectrum H(f) to obtain a preset number of peaks, and perform calculation processing on the preset number of peaks to obtain the formants.

[0077] It should be noted that Mel-Frequency Cepstral Coefficients (MFCC) is a speech feature extraction method designed based on the characteristics of human hearing. By simulating the nonlinear perception of frequency by the human ear, it converts speech signals into low-dimensional feature representations.

[0078] In this embodiment, the Mel frequency cepstral coefficients can be extracted as follows: The second speech information is subjected to an FFT (fast Fourier transform) to obtain the spectrum X[k], and the power spectrum P[k] = |X[k]|^2 is calculated; the power spectrum P[k] is weighted and summed through a set of Mel-scale triangular bandpass filter banks H_m[k] (where m is the filter index) to obtain the filter bank energy E[m] of each filter output; the natural logarithm of the filter bank energy E[m] is taken to obtain the logarithmic filter bank energy S[m]; the logarithmic filter bank energy S[m] is subjected to a discrete cosine transform to convert the energy to the cepstral domain, and the coefficients of the first few targets are retained to obtain the Mel frequency cepstral coefficients.

[0079] It should be noted that the Zero Crossing Rate (ZCR) measures the frequency at which an audio signal waveform crosses zero, typically reflecting the degree of audio oscillation. It can be used to distinguish between unvoiced sounds (high ZCR) and voiced sounds (low ZCR).

[0080] In this embodiment, the zero crossover rate can be extracted in the following way: traverse each sampling point of the second speech information, use the sign function to determine whether adjacent sampling points have different signs, count the total number of times the signal crosses the zero point N_zcr in the whole frame, and determine the total number of times N_zcr as the zero crossover rate.

[0081] In this embodiment, by performing first feature extraction on the second speech information, first feature information including pitch, fundamental frequency, formants, Mel frequency cepstral coefficients, and zero crossover rate can be obtained, thereby achieving comprehensive extraction of basic acoustic feature information.

[0082] S102-3: Extract the second feature from the second speech information to obtain the second feature information.

[0083] In this embodiment, the second feature information includes one or more of the fundamental frequency periodic perturbation parameters, amplitude perturbation parameters, and recurrence quantification analysis parameters.

[0084] It should be noted that the fundamental frequency periodic perturbation parameter (Jitter) is used to quantify the tiny fluctuations (perturbations) in the vocal cord vibration period. Teenagers typically have a higher proportion of jitter than adults, and this characteristic reflects the roughness of the sound.

[0085] In this embodiment, the fundamental frequency period perturbation parameter can be extracted in the following way: extract a series of fundamental frequency period values ​​Ti from multiple frames of the second speech information; calculate the absolute average value of the difference between consecutive fundamental frequency periods to obtain the absolute jitter; divide the absolute average value by the average fundamental frequency period of the second speech information to obtain the relative jitter; and determine the absolute jitter and / or relative jitter as the fundamental frequency period perturbation parameter.

[0086] Specifically, the absolute jitter can be calculated using the following formula: (5); Where Ti represents the period value corresponding to the i-th fundamental frequency period, and N is the number of fundamental frequency periods.

[0087] Specifically, the relative jitter can be calculated using the following formula: (6); It should be noted that the amplitude perturbation parameter (Shimmer) is used to quantify the periodic changes (perturbations) in the amplitude of vocal cord vibration, and it is also more pronounced in underage users. This feature reflects the hoarseness of the voice.

[0088] In this embodiment, the amplitude perturbation parameter can be extracted in the following way: extract the peak-to-peak amplitude Ai corresponding to each pitch period from the second speech information, calculate the absolute average value of the difference between consecutive period amplitudes to obtain the absolute perturbation value Shimmer (dB), and divide the absolute perturbation value by the average amplitude of the second speech information to obtain the relative perturbation value Shimmer (relative).

[0089] Specifically, the absolute disturbance value Shimmer (dB) can be calculated using the following formula: (7); Where Ai represents the peak-to-peak amplitude corresponding to the i-th fundamental frequency period, and N is the number of fundamental frequency periods.

[0090] Specifically, the relative disturbance value Shimmer can be calculated using the following formula: (8); It should be noted that Recurrence Quantification Analysis (RQA) is a nonlinear dynamics method used to analyze the complexity and stability of speech signals, revealing differences in vocal cord dynamic maturity.

[0091] In this embodiment, the recurrence quantification analysis parameters can be extracted as follows: the phase space of the second speech information is reconstructed using the delay coordinate method to obtain the phase space vector Yi=[xi,x(i+τ), x_(i+2τ), ..., x(i+(m-1)τ)], where m is the embedding dimension, τ is the delay time, (i=1, 2, ..., n′), n′=n (m 1) τ; Calculate the Euclidean distance between all phase space vector pairs and construct a binary recurrence matrix RM. If the distance between vectors Yi and Yj is ∥Yi-Yj∥ < distance threshold ε, then RM[i,j]=1, otherwise it is 0. Based on the recurrence matrix RM, calculate the recurrence rate (RR)=(ΣΣRM[i,j]for i≠j) / (n'*(n'-1)). This recurrence rate is defined as the ratio of the total number of recurrence points on the off-diagonal side of the recurrence matrix to the total number of all possible off-diagonal points. This recurrence rate (RR) can be determined as a parameter for recurrence quantification analysis.

[0092] In this embodiment, by preprocessing the first speech information and performing first feature extraction and second feature extraction on the preprocessed second speech information, the extracted first and second feature information can describe the user's physical and physiological attributes from different dimensions (time domain, frequency domain, cepstral domain, nonlinear dynamics), providing a solid data foundation for subsequent recognition, thereby ensuring the accuracy and comprehensiveness of user recognition.

[0093] In one feasible implementation, the step of determining the target user's target classification result based on the first feature information and the second feature information may specifically include the following sub-steps: S103-1: Identify the first feature information to obtain the first identification information.

[0094] In this embodiment, the first identification information represents the identification information of the user type identified based on the first feature information. For example, when the user type includes minors and adults, the first identification information can represent the confidence level that the target user is a minor.

[0095] In this embodiment, pitch, fundamental frequency, formant, Mel frequency cepstral coefficient, and zero crossover rate can be comprehensively identified to obtain the first identification information of the target user.

[0096] In this embodiment, the first feature information can be input into the first recognition model to obtain the first recognition information output by the first recognition model. The first recognition model can be trained based on models such as random forest, gradient boosting tree, Naive Bayes, and deep neural networks.

[0097] For example, when the first recognition model is trained based on a random forest model, statistical analysis can first be performed on pitch, fundamental frequency, formants, Mel frequency cepstral coefficients, and zero crossover rate to obtain a first feature vector. This first feature vector is then input into the pre-trained random forest model. The input feature vector starts from the root node of each decision tree, judges the feature values, and finally falls into a leaf node. Each decision tree outputs an independent judgment result, which is used to indicate whether the target user is a minor or an adult. The random forest model summarizes the voting results of all decision trees, and finally calculates the first recognition information P1 = (number of trees that voted that the user is a minor) / (total number of trees) based on the voting results.

[0098] In this embodiment, the random forest model, an ensemble learning algorithm, improves the accuracy and generalization ability of the model by combining a majority of decision trees. It can effectively handle high-dimensional features and is not prone to overfitting. To improve the model's recognition performance, the number of trees in each random forest, n_estimators, can be set to 300, the maximum depth of the trees, max_depth, can be limited to 18, the minimum number of samples per leaf node, min_samples_leaf, can be limited to 10, and the maximum number of features when splitting a tree, max_features, can be limited to sqrt.

[0099] In this embodiment, by analyzing and processing the first feature information through the first recognition model, the first recognition information can be quickly and accurately identified.

[0100] In this embodiment, when the user type includes minors and adults, the first recognition model can be trained based on a first training dataset, which includes a first minor speech dataset and an adult speech dataset. The first minor speech dataset represents a dataset composed of speech information from minors in a first age group from various provinces and cities across the country; the adult speech dataset represents a dataset composed of speech information from adults in a second age group from various provinces and cities across the country. The first age group can be [3, 18), meaning the first minor speech dataset contains the speech of minors in each age group capable of voice interaction.

[0101] In this embodiment, for any speech information in the first training dataset, a uniform sampling rate, such as 16kHz, can be used for recording to ensure high speech quality. Simultaneously, a relatively quiet environment is selected to reduce background noise interference with classification. After acquiring each speech information, each speech information is labeled as either "minor" or "adult," and its age information is also labeled.

[0102] In this embodiment, by providing a high-quality first training dataset, the training effect of the first recognition model can be effectively improved, thereby improving the recognition effect for users of all ages.

[0103] In this embodiment, after training the first recognition model is completed, the first recognition model can be converted into an Android-compatible format, such as saving it in a .pkl file. The trained Python model can be directly called on Android using the lightweight Python interpreter (Chaquopy). The model file can be imported into the project using Chaquopy, and the first recognition model can be called for inference.

[0104] S103-2: Identify the second feature information to obtain the second identification information.

[0105] In this embodiment, the second identification information represents user type information identified based on the second feature information. For example, when the user type includes minors and adults, the second identification information can represent the confidence level that the target user is a minor.

[0106] In this embodiment, the fundamental frequency period disturbance parameters, amplitude disturbance parameters, and recurrence quantification analysis parameters can be comprehensively identified to obtain the second identification information of the target user.

[0107] In this embodiment, the second feature information can be input into the second recognition model to obtain the second recognition information output by the second recognition model. The second recognition model can be trained based on models such as support vector machines, logistic regression, K-nearest neighbor algorithm, Gaussian process classification, and deep neural networks.

[0108] In this embodiment, considering the boundary behavior and nonlinear separability of the second feature information in low-dimensional space, the support vector machine can effectively adapt to the feature distribution through kernel functions such as RBF and has excellent small sample generalization ability. Therefore, for the second feature information, the support vector machine can be used as the main classification model to model the user age group discrimination.

[0109] For example, when the second recognition model is trained based on a support vector machine (SVM), statistical analysis can first be performed on the fundamental frequency period perturbation parameters, amplitude perturbation parameters, and recurrence quantization analysis parameters to obtain a second feature vector. This second feature vector is then input into a pre-trained SVM. The SVM uses a radial basis function (RBF) as its kernel function, which maps the features to a higher-dimensional space to find the optimal classification hyperplane and outputs the signed distance from the sample point to the classification hyperplane. By setting the probability output of the SVM to the True parameter, an additional Platt scaling process can be performed after the model training is complete. This involves using a sigmoid function to calibrate and map the distance values ​​output by the SVM to the [0, 1] interval, thereby outputting the second recognition information P2. This second recognition information P2 represents the confidence level that the target user is a minor.

[0110] In this embodiment, to improve the model's recognition performance, the penalty coefficient (C) can be set to 10. The penalty coefficient (C) is used to control the model's fault tolerance and maximize the balance interval. The larger the C value, the stronger the penalty for classification errors, but the easier it is to overfit. The Gaussian kernel width parameter (gamma) can be set to 'auto' or 0.01. The Gaussian kernel width parameter (gamma) represents the width of the Gaussian kernel and affects the flexibility of the discrimination boundary. It can be obtained through cross-validation optimization. Other parameters such as the maximum number of iterations (max_iter) and the convergence tolerance (tol) can be set to default values ​​to ensure efficient convergence of the algorithm.

[0111] In this embodiment, by analyzing and processing the second feature information through the second recognition model, the second recognition information can be quickly and accurately identified.

[0112] In this embodiment, when the user type includes minors and adults, the second recognition model can be trained based on the second training dataset, which is obtained in the following way: extracting the second minor voice dataset corresponding to minors in the target age group from the first minor voice dataset; and obtaining the second training dataset based on the second minor voice dataset and the adult voice dataset.

[0113] In this embodiment, the target age range can be an age range representing the voice change period, for example, the target age range can be set to [12, 18). That is, data of children under 12 years old can be removed from the first minor speech dataset to obtain the second minor speech dataset, and combined with the adult speech dataset to obtain the second training dataset.

[0114] In this embodiment, by extracting the second minor voice dataset corresponding to minors in the target age group from the first minor voice dataset as the training sample of the second recognition model, the system's discrimination ability in complex scenarios such as user boundary samples, voice change period, and falsetto imitation can be effectively enhanced.

[0115] In this embodiment, after the second recognition model is trained, it can be saved as a .pkl file or a model weight / parameter file. During deployment, the second recognition model can be inferred by calling the Python module in the Android application through Chaquopy. Alternatively, the LibSVM porting package for the Java platform can be used to migrate the model parameters to Java and perform direct inference locally using the LibSVM Java interface, achieving a more efficient edge response.

[0116] S103-3: Based on the first identification information and the second identification information, the target user's target classification result is obtained.

[0117] In this embodiment, the first identification information and the second identification information are used to characterize the confidence level that the target user is a minor. The first identification information is obtained from identifying first feature information based on a fundamental acoustic dimension, and the second identification information is obtained from identifying second feature information based on a medical acoustic dimension. Therefore, by comprehensively considering both the first and second identification information, the accuracy and robustness of the judgment can be improved.

[0118] In a specific implementation, the step of obtaining the target classification result of the target user based on the first identification information and the second identification information may include: performing a weighted calculation on the first identification information and the second identification information to obtain the target confidence score; if the target confidence score is greater than the confidence score threshold, determining the target user's target classification result as the first classification result, which indicates that the target user is a minor; if the target confidence score is less than or equal to the confidence score threshold, determining the target user's target classification result as the second classification result, which indicates that the target user is an adult.

[0119] In this embodiment, the target confidence level can be obtained by weighting the first identification information and the second identification information according to the following formula: P_final=α×P1+(1-α)×P2(9); Where P_final represents the target confidence level, P1 represents the first identification information, P2 represents the second identification information, and α represents the weight coefficient.

[0120] In this embodiment, the weighting coefficients can be adjusted according to actual needs to dynamically adapt to different users and noise environments, thereby improving the system's robustness and universality. For example, the weighting coefficients can be adjusted based on the age distribution of family members. For instance, if there are more children in the family going through puberty, the weighting coefficients can be reduced, thereby increasing the fusion weight of the second recognition information to improve the recognition accuracy for users going through puberty.

[0121] In this embodiment, by weighting the first and second identification information, complementary identification performance for different age groups can be achieved, significantly improving the model's classification performance in complex scenarios (such as puberty, noisy environments, and imitation attacks). When the voices of minors and adults overlap in basic acoustic features, medical acoustic features can provide more granular physiological differentiation criteria, thereby reducing the risk of misclassification.

[0122] In this embodiment, after training the first recognition model and / or the second recognition model is completed, the first recognition model and / or the second recognition model can be validated using a validation dataset. For example, a decimal cross-validation method can be used for validation.

[0123] In this implementation, the validation dataset can be divided into 10 equal parts, with 1 part used for validation and the remaining 9 parts used for training. The cross-validation process is repeated 10 times, changing the subset of the dataset as the validation set each time to ensure all data undergoes both training and validation. Based on the validation set results, target performance metrics are calculated, including accuracy and recall. Accuracy evaluates the proportion of correctly predicted samples out of the total sample, while recall, specifically for minors, represents the proportion of correctly identified minors out of the total number of real minors, measuring the risk of the system missing minors. Table 3 shows the statistical results of the ten-part cross-validation.

[0124] Table 3 Statistical results of decimal cross-validation

[0125] According to the statistical results, the weighted fusion of the recognition results from the first and second recognition models, under ten-part cross-validation, achieved an average accuracy of 96.17%, with a recall rate of 95.71% for the minor category. These results demonstrate high accuracy in distinguishing between minors and adults, thus effectively supporting automatic content rating and access security control on smart terminals such as televisions.

[0126] In this embodiment, the step of pushing content to the target user based on the target classification result may specifically include: broadcasting the first classification result to the target application when the target classification result is the first classification result, and / or setting the system mode to minor mode, so that the target application can filter the content to be pushed when it hears the first classification result and / or detects that the system mode is minor mode, and push the filtered content to the target user.

[0127] In this embodiment, when the target classification result is the first classification result, i.e., the target user is a minor, to ensure resource access security, the target application can be controlled to filter the content to be pushed through two mechanisms: broadcasting the classification result or modifying the system mode. The content to be pushed refers to the original push content determined based on the target user's first voice information. For example, semantic recognition can be performed on the first voice information, and the content to be pushed can be determined from a preset multimedia resource library based on the semantic recognition result.

[0128] In this embodiment, the target application refers to one or more applications that currently provide multimedia resources.

[0129] In this implementation, when the target user is identified as a minor, the ACTION_TEEN_MODE broadcast can be triggered. All running target applications (such as video apps, game apps, and desktop launchers) that have registered to listen to this broadcast will immediately receive the event and enter the preset minor mode.

[0130] Specifically, when a video app detects the first category result, it can immediately switch from the full-featured main interface to a "minor mode" interface with a softer color scheme and recommended content consisting only of animations and educational videos, while hiding entry points such as "recharge" and "live streaming". For applications that are already open, it can immediately filter sensitive content in the currently displayed content, for example, automatically closing inappropriate tabs that are already open in the browser. When the desktop launcher detects the first category result, it can immediately hide application icons that are not suitable for minors (such as some social software and news apps).

[0131] In this implementation, considering that if an application is not started or is cleared in the background, it will not be able to receive this broadcast event, the system mode can also be set to minor mode, for example, by setting the corresponding KEY value of the system SettingsProvider to SETTINGS_TEEN_MODE.

[0132] In this implementation, by persistently writing SETTINGS_TEEN_MODE to the global SettingsProvider system service, a unified status flag can be set at the system level. This mode status will be maintained until the user identification result changes or the user manually modifies the system mode.

[0133] It's important to note that for any application on the TV, regardless of whether it was running when the broadcast was sent, once the application starts, it can query the SettingsProvider to obtain its current mode state and adjust its behavior accordingly. This effectively compensates for the shortcomings of the broadcast mechanism. Even if an application misses a broadcast (e.g., being cleared from the background), it can quickly enter the correct mode upon the next launch by querying the SettingsProvider, greatly enhancing the system's fault tolerance and the consistency of the user experience.

[0134] In this embodiment, by simultaneously employing both broadcast and system attribute mechanisms when the target user is detected to be a minor, it is possible to effectively ensure that the target application enters the minor mode immediately upon startup or runtime, thereby avoiding the display of inappropriate content and achieving content access security and hierarchical management for family members.

[0135] To facilitate better implementation of the content push method of this application, this application also provides a content push device based on the above-described content push method. The meanings of the terms used are the same as in the content push method described above, and specific implementation details can be found in the descriptions of the method embodiments.

[0136] Based on the same inventive concept, and referring to Figure 2 This application provides a content push device 200, which includes: Information acquisition module 201 is used to acquire the first voice information of the target user; Feature extraction module 202 is used to extract features from the first speech information to obtain first feature information and second feature information; The result determination module 203 is used to determine the target classification result of the target user based on the first feature information and the second feature information; The content push module 204 is used to push content to target users based on the target classification results.

[0137] In one embodiment, the feature extraction module 202 includes: The preprocessing submodule is used to preprocess the first speech information to obtain the second speech information; the preprocessing includes one or more of the following: noise reduction processing, normalization processing, frame segmentation processing, and windowing processing. The first feature extraction submodule is used to perform first feature extraction on the second speech information to obtain first feature information; the first feature information includes one or more of pitch, fundamental frequency, formants, Mel frequency cepstral coefficients and zero crossover rate. The second feature extraction submodule is used to extract second features from the second speech information to obtain second feature information; the second feature information includes one or more of the fundamental frequency period perturbation parameters, amplitude perturbation parameters, and recurrence quantification analysis parameters.

[0138] In one embodiment, the result determination module 203 includes: The first identification submodule is used to identify the first feature information to obtain the first identification information; The second identification submodule is used to identify the second feature information to obtain the second identification information; The result determination submodule is used to obtain the target classification result of the target user based on the first identification information and the second identification information.

[0139] In one embodiment, the first recognition submodule is specifically used to input the first feature information into the first recognition model to obtain the first recognition information output by the first recognition model; The second recognition submodule is specifically used to input the second feature information into the second recognition model and obtain the second recognition information output by the second recognition model.

[0140] In one embodiment, the first identification information and the second identification information are used to characterize the confidence level that the target user is a minor; The first recognition model is trained based on the first training dataset, which includes the first minor speech dataset and the first adult speech dataset. The second recognition model is trained based on a second training dataset, which is obtained in the following way: In the first minor voice dataset, extract the second minor voice dataset corresponding to minors in the target age group; A second training dataset was obtained based on the second minor speech dataset and the adult speech dataset.

[0141] In one embodiment, the result determination submodule includes: The weighted calculation unit is used to perform weighted calculation on the first identification information and the second identification information to obtain the target confidence level; The first classification unit is used to determine the target user's target classification result as the first classification result when the target confidence score is greater than the confidence threshold; the first classification result indicates that the target user is a minor. The second classification unit is used to determine the target user's target classification result as the second classification result when the target confidence level is less than or equal to the confidence level threshold; the second classification result indicates that the target user is an adult.

[0142] In one embodiment, the content push module 204 includes: The first content push submodule is used to broadcast the first category result to the target application when the target category result is the first category result, and / or set the system mode to minor mode, so that the target application can filter the content to be pushed when it hears the first category result and / or detects that the system mode is minor mode, and push the filtered content to the target user; wherein, the first category result indicates that the target user is a minor.

[0143] By employing the technical solution of this application embodiment, multi-dimensional feature extraction is performed on the first voice information of the target user to obtain first feature information and second feature information. The first feature information and second feature information are then comprehensively analyzed, which can significantly improve the accuracy of the target classification results. In turn, appropriate content can be pushed according to the target classification results, thereby improving the user experience and effectively ensuring the content access security of family members.

[0144] Specific limitations regarding the content push device 200 can be found in the limitations of the content push method described above, and will not be repeated here. Each module in the aforementioned content push device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0145] In addition, this application also provides an electronic device, such as Figure 3 As shown, it illustrates the structural diagram of the electronic device involved in this application, specifically: The electronic device may include components such as a processor 301 with one or more processing cores and a memory 302 with one or more computer-readable storage media. Those skilled in the art will understand that... Figure 3 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 301 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in the memory 302, and by calling data stored in the memory 302, thereby providing overall monitoring of the electronic device. Optionally, the processor 301 may include one or more processing cores; preferably, the processor 301 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 301.

[0146] The memory 302 can be used to store software programs and modules. The processor 301 executes various functional applications and data processing by running the software programs and modules stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 302 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.

[0147] In one feasible implementation, the electronic device further includes a power supply 303 that supplies power to the various components. Preferably, the power supply 303 can be logically connected to the processor 301 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 303 may also include one or more DC or AC power supplies, recharging systems, power equipment debugging circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0148] In one feasible implementation, the electronic device may further include an input unit 304, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0149] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 301 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 302 according to the following instructions, and the processor 301 runs the applications stored in the memory 302, thereby implementing the steps in any of the content push methods provided in the embodiments of this application.

[0150] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0151] In one feasible implementation, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the methods described in any embodiment of this application.

[0152] In one feasible implementation, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the methods described in any embodiment of this application.

[0153] In one feasible implementation, a computer program product is also proposed, comprising a computer program or instructions that, when executed by a processor, implement the methods described in any embodiment of this application.

[0154] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0155] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0156] To this end, this application provides a computer-readable storage medium storing a computer program that can be loaded by a processor to execute the steps in any of the content delivery methods provided in this application.

[0157] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0158] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0159] Since the instructions stored in the computer-readable storage medium can execute the steps of any of the content push methods provided in this application, the beneficial effects that any of the content push methods provided in this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0160] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0161] The above provides a detailed description of a content push method, apparatus, electronic device, and computer-readable storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A content push method characterized by, The method comprises: obtaining first voice information of a target user; performing feature extraction on the first voice information to obtain first feature information and second feature information; determining a target classification result of the target user based on the first feature information and the second feature information; performing content pushing on the target user based on the target classification result.

2. The content push method according to claim 1, characterized by, The feature extraction on the first voice information to obtain the first feature information and the second feature information comprises: performing preprocessing on the first voice information to obtain second voice information; the preprocessing comprises one or more of denoising processing, normalization processing, frame processing and windowing processing; performing first feature extraction on the second voice information to obtain the first feature information; the first feature information comprises one or more of pitch, fundamental frequency, formant, mel-frequency cepstral coefficient and zero-crossing rate; performing second feature extraction on the second voice information to obtain the second feature information; the second feature information comprises one or more of fundamental frequency period perturbation parameter, amplitude perturbation parameter and recurrence quantification analysis parameter.

3. The content push method of claim 1, wherein, The determination of the target classification result of the target user based on the first feature information and the second feature information comprises: performing identification on the first feature information to obtain first identification information; performing identification on the second feature information to obtain second identification information; obtaining the target classification result of the target user based on the first identification information and the second identification information.

4. The content push method according to claim 3, characterized by, The identification on the first feature information to obtain the first identification information comprises: inputting the first feature information into a first identification model to obtain the first identification information output by the first identification model; The identification on the second feature information to obtain the second identification information comprises: inputting the second feature information into a second identification model to obtain the second identification information output by the second identification model.

5. The content push method of claim 4, wherein, The first identification information and the second identification information are respectively used to represent the confidence degree of the target user being a minor; The first identification model is obtained based on a first training data set, and the first training data set comprises a first minor voice data set and an adult voice data set; The second identification model is obtained based on a second training data set, and the second training data set is obtained in the following manner: extracting a second minor voice data set corresponding to minors in a target age range from the first minor voice data set; obtaining the second training data set based on the second minor voice data set and the adult voice data set.

6. The content push method of claim 3, wherein, The obtaining of the target classification result of the target user based on the first identification information and the second identification information comprises: performing weighted calculation on the first identification information and the second identification information to obtain a target confidence degree; determining the target classification result of the target user as a first classification result in a case where the target confidence degree is greater than a confidence degree threshold; the first classification result represents that the target user is a minor. In a case where the target confidence is less than or equal to the confidence threshold, a target classification result of the target user is determined as a second classification result; the second classification result represents that the target user is an adult.

7. The content push method of claim 1, wherein, The content pushing to the target user based on the target classification result comprises: In a case where the target classification result is the first classification result, the first classification result is broadcast to a target application, and / or a system mode is set as a minor mode, so that the target application filters to-be-pushed content and pushes the filtered to-be-pushed content to the target user in a case where the first classification result is listened to and / or the system mode is detected as the minor mode; the first classification result represents that the target user is a minor.

8. A content pusher device, characterized by, The apparatus comprises: an information acquisition module configured to acquire first voice information of a target user; a feature extraction module configured to perform feature extraction on the first voice information to obtain first feature information and second feature information; a result determination module configured to determine a target classification result of the target user based on the first feature information and the second feature information; a content pushing module configured to push content to the target user based on the target classification result.

9. An electronic device, comprising: The computer program is stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps in the content pushing method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps in the content pushing method according to any one of claims 1 to 7.

Images