User attribute identification method, device, electronic device and computer readable medium

Abstract

The present disclosure relates to a method, device, electronic device and computer-readable medium for identifying user attributes, and belongs to the field of speech recognition technology. The method includes: obtaining call voice data to be identified, and extracting a feature vector corresponding to the call voice data; inputting the feature vector into a pre-trained outbound user classification model to obtain the gender attribute of the outbound user in the call voice data; determining a target answering user classification model from two pre-trained answering user classification models according to the gender attribute of the outbound user; inputting the feature vector into the target answering user classification model to obtain the gender attribute of the answering user in the call voice data. The present disclosure can determine the role type and gender attribute of the two parties in a call based on call voice data by means of feature extraction and pattern matching.

Description

Technical Field

[0001] The present disclosure relates to the technical field of speech recognition, and in particular to a method for identifying user attributes, a device for identifying user attributes, an electronic device, and a computer-readable medium. Background Art

[0002] Telecommunications operators have a huge amount of user data, which is of great value. Using user data can make a comprehensive portrait of users. However, since the accuracy of some user attributes cannot be guaranteed, the user portrait may be inaccurate, which may affect the success rate of telemarketing for telecommunications operators.

[0003] In view of this, there is an urgent need in the art for a method that can accurately identify user attributes.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The purpose of the present disclosure is to provide a method for identifying user attributes, a device for identifying user attributes, an electronic device and a computer-readable medium, thereby improving the accuracy of identifying user attributes at least to a certain extent.

[0006] According to a first aspect of the present disclosure, a method for identifying user attributes is provided, comprising:

[0007] Acquire call voice data to be recognized, and extract feature vectors corresponding to the call voice data;

[0008] Inputting the feature vector into a pre-trained outbound caller classification model to obtain the gender attribute of the outbound caller in the call voice data;

[0009] Determine a target incoming call user classification model from two pre-trained incoming call user classification models according to the gender attribute of the outgoing call user;

[0010] The feature vector is input into the target answering user classification model to obtain the gender attribute of the answering user in the call voice data.

[0011] In an exemplary embodiment of the present disclosure, the extracting a feature vector corresponding to the call voice data includes:

[0012] Annotating the attributes of each voice frame in the call voice data by using a pre-trained voice basic model, wherein the attributes of the voice frame include male voice, female voice and silence;

[0013] Dividing the call voice data into a plurality of voice data segments according to the position of the voice frame with the attribute of silence;

[0014] Determining the ratio of male voice frames or female voice frames in each of the voice data segments according to the attributes of each of the voice frames;

[0015] Obtaining a frame ratio distribution statistic of the call voice data according to the male voice frame ratio or the female voice frame ratio in each of the voice data segments;

[0016] Determine the fundamental frequency distribution statistics of the call voice data according to the fundamental frequency corresponding to each of the voice frames;

[0017] A feature vector corresponding to the call voice data is obtained according to the frame ratio distribution statistics and the fundamental frequency distribution statistics.

[0018] In an exemplary embodiment of the present disclosure, obtaining the frame ratio distribution statistic of the call voice data according to the male voice frame ratio or the female voice frame ratio in each of the voice data segments includes:

[0019] Obtaining an array of male voice frame ratios of the call voice data according to the male voice frame ratios in each of the voice data segments, or obtaining an array of female voice frame ratios of the call voice data according to the female voice frame ratios in each of the voice data segments;

[0020] Acquire a plurality of preset frame ratio intervals, and obtain male voice frame ratio distribution statistics according to the male voice frame ratio array and the frame ratio interval, or obtain female voice frame ratio distribution statistics according to the female voice frame ratio array and the frame ratio interval;

[0021] The male voice frame ratio distribution statistic or the female voice frame ratio distribution statistic is used as the frame ratio distribution statistic of the call voice data.

[0022] In an exemplary embodiment of the present disclosure, determining the fundamental frequency distribution statistic corresponding to the call voice data according to the fundamental frequency of each of the voice frames includes:

[0023] Removing noise frames from the call voice data according to a preset effective fundamental frequency interval to obtain effective voice frames from the call voice data;

[0024] Obtaining a pitch frequency array corresponding to the call voice data according to the pitch frequencies of each of the valid voice frames;

[0025] A plurality of preset pitch frequency intervals are obtained, and a pitch frequency distribution statistic corresponding to the call voice data is obtained according to the pitch frequency array and the pitch frequency interval.

[0026] In an exemplary embodiment of the present disclosure, the method for training the basic speech model includes:

[0027] Acquire historical call voice data, and obtain male voice training samples, female voice training samples, and silent training samples according to the attributes of the voice frames in the historical call voice data;

[0028] Corresponding speech features are extracted from the male voice training samples, the female voice training samples and the silent training samples respectively, and a male voice basic model, a female voice basic model and a silent basic model are obtained through the speech feature training.

[0029] In an exemplary embodiment of the present disclosure, the training method of the outbound user classification model includes:

[0030] Acquire the first voice training data corresponding to the outbound user classification model, and acquire the outbound call attributes corresponding to each of the first voice training data;

[0031] Determine the feature vectors corresponding to each of the first voice training data, and obtain an outbound call feature vector group according to the feature vectors corresponding to each of the first voice training data;

[0032] The outbound call user classification model is obtained by training according to the outbound call attributes corresponding to the first voice training data, the outbound call feature vector group and a preset classification algorithm.

[0033] In an exemplary embodiment of the present disclosure, the training method of the answering user classification model includes:

[0034] Acquire the second voice training data corresponding to the answering user classification model, and acquire the answering attributes corresponding to each of the second voice training data;

[0035] Determine the feature vectors corresponding to each of the second voice training data, and obtain a listening feature vector group according to the feature vectors corresponding to each of the second voice training data;

[0036] The outbound call user classification model is used to divide the answering feature vector group into a first answering feature vector group and a second answering feature vector group according to outbound call attributes;

[0037] A first answering user classification model is obtained by training according to the answering attributes corresponding to the second voice training data, the first answering feature vector group, and a preset classification algorithm;

[0038] The second answering user classification model is obtained by training according to the answering attributes corresponding to the second voice training data, the second answering feature vector group and a preset classification algorithm.

[0039] According to a second aspect of the present disclosure, there is provided a device for identifying user attributes, comprising:

[0040] A feature vector acquisition module, used to acquire the call voice data to be recognized and extract the feature vector corresponding to the call voice data;

[0041] An outbound call attribute determination module, used for inputting the feature vector into a pre-trained outbound call user classification model to obtain the gender attribute of the outbound call user in the call voice data;

[0042] A call answering model determination module, used to determine a target call answering user classification model from two pre-trained call answering user classification models according to the gender attribute of the outbound call user;

[0043] The answering attribute determination module is used to input the feature vector into the target answering user classification model to obtain the gender attribute of the answering user in the call voice data.

[0044] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned user attribute identification methods by executing the executable instructions.

[0045] According to a fourth aspect of the present disclosure, a computer-readable medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the method for identifying user attributes described in any one of the above is implemented.

[0046] The exemplary embodiments of the present disclosure may have the following beneficial effects:

[0047] In the method for identifying user attributes in the example implementation of the present disclosure, the gender attribute of the outbound user in the call voice data is obtained by extracting the feature vector corresponding to the call voice data to be identified, and inputting the feature vector into the outbound user classification model, and then matching the corresponding answering user classification model according to the identification result of the outbound user's gender attribute, and then judging the gender attribute of the answering user according to the answering user classification model. The method for identifying user attributes in the example implementation of the present disclosure, on the one hand, can accurately judge the role type of the communicating parties and identify the gender attribute of each communicating role by means of feature extraction and pattern matching, solve the problem of inaccurate user key labels in user portraits, and provide reliable data support for upper-level data applications; on the other hand, by associating the user's gender attribute with the user ID, the user's basic data can be dynamically maintained, and the dynamic update of user data can be realized.

[0048] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0050] Figure 1 A schematic diagram showing a flow chart of a method for identifying user attributes according to an exemplary embodiment of the present disclosure;

[0051] Figure 2 A schematic diagram showing a process of extracting a feature vector corresponding to call voice data according to an exemplary embodiment of the present disclosure;

[0052] Figure 3 A flowchart of a method for training a basic speech model according to an exemplary embodiment of the present disclosure is shown;

[0053] Figure 4 A schematic diagram showing a flow chart of determining a frame ratio distribution statistic of call voice data according to an exemplary embodiment of the present disclosure;

[0054] Figure 5 A schematic diagram showing a flow chart of determining a fundamental frequency distribution statistic of call speech data according to an exemplary embodiment of the present disclosure;

[0055] Figure 6 A flow chart showing a method for training an outbound user classification model according to an exemplary embodiment of the present disclosure;

[0056] Figure 7 A flow chart showing a method for training a call-answering user classification model according to an exemplary embodiment of the present disclosure is provided;

[0057] Figure 8 A schematic diagram showing a flow chart of a method for identifying user attributes according to a specific implementation of the present disclosure;

[0058] Fig. 9 The overall architecture diagram of the classification model according to a specific implementation of the present disclosure is shown;

[0059] Fig.10 An architectural diagram of a distributed service engine according to a specific embodiment of the present disclosure is shown;

[0060] Fig.11 A block diagram showing an apparatus for identifying user attributes according to an exemplary embodiment of the present disclosure;

[0061] Fig.12A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0062] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0063] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0064] In scenarios such as maintaining user relationships and recommending new products, communication operators often need to use basic user data for analysis and modeling, and recommend products to different users based on user portraits. However, for operator users, there may be situations where the user uses his or her ID card to apply for a phone card that is not used by the user himself or herself, so the user information recorded in the database based on the ID card may not be accurate. In addition, when communication operators manage user data, there may be defects in management methods or management technologies in various links of data transmission, the quality of collected data is not good enough, and the accuracy of some user attributes cannot be guaranteed, which may cause inaccurate user portraits and affect the success rate of communication operators in telephone marketing or product recommendations.

[0065] Based on the above problems, this exemplary embodiment first provides a method for identifying user attributes. Figure 1 As shown, the above method for identifying user attributes may include the following steps:

[0066] Step S110: Acquire the call voice data to be recognized, and extract the feature vector corresponding to the call voice data.

[0067] Step S120: Input the feature vector into a pre-trained outbound user classification model to obtain the gender attribute of the outbound user in the call voice data.

[0068] Step S130: Determine a target incoming call user classification model from two pre-trained incoming call user classification models according to the gender attribute of the outgoing call user.

[0069] Step S140: Input the feature vector into the target answering user classification model to obtain the gender attribute of the answering user in the call voice data.

[0070] In the method for identifying user attributes in the example implementation of the present disclosure, the gender attribute of the outbound user in the call voice data is obtained by extracting the feature vector corresponding to the call voice data to be identified, and inputting the feature vector into the outbound user classification model, and then matching the corresponding answering user classification model according to the identification result of the outbound user's gender attribute, and then judging the gender attribute of the answering user according to the answering user classification model. The method for identifying user attributes in the example implementation of the present disclosure, on the one hand, can accurately judge the role type of the communicating parties and identify the gender attribute of each communicating role by means of feature extraction and pattern matching, solve the problem of inaccurate user key labels in user portraits, and provide reliable data support for upper-level data applications; on the other hand, by associating the user's gender attribute with the user ID, the user's basic data can be dynamically maintained, and the dynamic update of user data can be realized.

[0071] The method for identifying user attributes in the example implementation of the present disclosure can be applied to the outbound marketing call scenario of a communication operator and has a certain degree of versatility. Compared with the solution that only identifies the user attributes of a single party, in the specific field of outbound voice voiceprint recognition, based on the marketing call recordings of marketing personnel and users, by adding a post-algorithm to solve the problem of identifying the roles and genders of both callers, it is possible to achieve gender discrimination of both the outbound caller and the answering user, reflecting the application value of voiceprint recognition in this field.

[0072] Next, combine Figures 2 to 7 The above steps of this exemplary embodiment are described in more detail.

[0073] In step S110, the call voice data to be recognized is obtained, and the feature vector corresponding to the call voice data is extracted.

[0074] In this example implementation, the call voice data to be identified can be obtained by acquiring corresponding telephone recordings, such as outbound marketing recordings of operators, etc. The feature vector extracted based on the call voice data is a multi-dimensional feature vector.

[0075] In this example implementation, Figure 2 As shown, extracting the feature vector corresponding to the call voice data may specifically include the following steps:

[0076] Step S210: Label the attributes of each speech frame in the call speech data using a pre-trained basic speech model.

[0077] In this example implementation, the attributes of the speech frame include male voice, female voice and silence, and the speech basic model includes a male voice basic model, a female voice basic model and a silence basic model.

[0078] In this example implementation, Figure 3 As shown, the training method of the speech basic model may specifically include the following steps:

[0079] Step S310: Acquire historical call voice data, and obtain male voice training samples, female voice training samples and silent training samples according to the attributes of the voice frames in the historical call voice data.

[0080] By collecting historical call voice data, such as historical outbound marketing voice data, etc., and then marking and organizing them to form male voice training samples representing male voices, female voice training samples representing female voices, and silent training samples representing silent voices.

[0081] Specifically, multiple data frames can be extracted from the historical call voice data and labeled respectively, with the data frames labeled as male voice being used as male voice training samples, the voice data frames labeled as female voice being used as female voice training samples, and the data frames labeled as silent voice being used as silent voice training samples.

[0082] Step S320. Extract corresponding speech features from the male voice training samples, the female voice training samples and the silent training samples respectively, and obtain a male voice basic model, a female voice basic model and a silent basic model through speech feature training.

[0083] In this example implementation, corresponding speech features such as MFCC (Mel-scale Frequency Cepstral Coefficients), resonance peak frequency, and fundamental frequency F0 can be extracted from male voice training samples, female voice training samples, and silence training samples, respectively, and then a male voice basic model representing male voice, a female voice basic model representing female voice, and a silence basic model representing silence can be obtained through the above-mentioned speech feature training.

[0084] The male voice basic model, female voice basic model and silent voice basic model can be used to analyze and judge the call voice data to be recognized, and each voice frame in the call voice data can be labeled according to the attributes of "male voice, female voice, silent voice".

[0085] For example, a call voice data to be recognized includes 7 voice frames: A, B, C, D, E, F, G. By judging these voice frames separately according to the above-mentioned basic voice model, the following marking results can be obtained: A, B, G are marked as male voices, D and F are marked as female voices, and C and E are marked as silent.

[0086] Step S220: Divide the call voice data into a plurality of voice data segments according to the positions of the voice frames with the attribute of silence.

[0087] After marking the attributes of each voice frame in the call voice data, the call voice data is segmented according to the positions of the voice frames with the attribute of "silence" to obtain voice data segments.

[0088] For example, according to the location of voice frames C and E with the attribute of "silent", the above call voice data can be divided into three voice data segments, the first voice data segment is A and B, the second voice data segment is D, and the third voice data segment is F and G.

[0089] Step S230: Determine the male voice frame ratio or the female voice frame ratio in each voice data segment according to the attributes of each voice frame.

[0090] According to the attributes of each speech frame, the male voice frame ratio or female voice frame ratio in each speech data segment can be calculated. The male voice frame ratio refers to the ratio of the speech frames marked as "male voice" in each speech data segment to the total number of speech frames in the speech data segment, and the female voice frame ratio is the same.

[0091] For example, in the above call voice data, the male voice frame ratio of the first voice data segment A and B is 1, the male voice frame ratio of the second voice data segment D is 0, and the male voice frame ratio of the third voice data segment F and G is 0.5. Alternatively, the female voice frame ratio of each voice data segment can be calculated by the same method.

[0092] Step S240: Obtain a frame ratio distribution statistic of the call voice data according to the male voice frame ratio or the female voice frame ratio in each voice data segment.

[0093] In this example implementation, Figure 4 As shown, obtaining the frame ratio distribution statistics of the call voice data according to the male voice frame ratio or the female voice frame ratio in each voice data segment may specifically include the following steps:

[0094] Step S410. Obtain a male voice frame ratio array of the call voice data according to the male voice frame ratio in each voice data segment, or obtain a female voice frame ratio array of the call voice data according to the female voice frame ratio in each voice data segment.

[0095] The male voice frame ratio array is an array consisting of the male voice frame ratios in each voice data segment, and the female voice frame ratio array is an array consisting of the female voice frame ratios in each voice data segment. For example, the male voice frame ratio array of the above call voice data is (1, 0, 0.5).

[0096] Step S420. Obtain multiple preset frame ratio ranges, and obtain male voice frame ratio distribution statistics based on the male voice frame ratio array and the frame ratio range, or obtain female voice frame ratio distribution statistics based on the female voice frame ratio array and the frame ratio range.

[0097] By counting the number of data in the male voice frame ratio array that fall within a plurality of different and equally wide frame ratio intervals, the male voice frame ratio distribution statistics can be obtained. The female voice frame ratio distribution statistics are similar.

[0098] For example, the male voice frame ratio array is (1, 0, 0.5), and the preset multiple frame ratio intervals are [0-0.25), [0.25-0.5), [0.5-0.75), and [0.75-1]. In the above male voice frame ratio array, the male voice frame ratio 1 of the first voice data segment is included in the range of [0.75-1], the male voice frame ratio 0 of the second voice data segment is included in the range of [0-0.25), and the male voice frame ratio 0.5 of the third voice data segment is included in the range of [0.5-0.75]. Therefore, the male voice frame ratio distribution statistic can be obtained as (1, 0, 1, 1).

[0099] Step S430: Using the male voice frame ratio distribution statistic or the female voice frame ratio distribution statistic as the frame ratio distribution statistic of the call voice data.

[0100] Finally, one of the male voice frame ratio distribution statistics or the female voice frame ratio distribution statistics is used as the frame ratio distribution statistics of the call voice data.

[0101] Step S250: Determine the fundamental frequency distribution statistics of the call voice data according to the fundamental frequency corresponding to each voice frame.

[0102] In this example implementation, Figure 5 As shown, determining the fundamental frequency distribution statistics of the call voice data according to the fundamental frequency corresponding to each voice frame may specifically include the following steps:

[0103] Step S510: Remove noise frames in the call voice data according to a preset effective fundamental frequency interval to obtain effective voice frames in the call voice data.

[0104] First, noise frames in the call voice data are removed according to a preset effective fundamental frequency range. For example, voice frames with a fundamental frequency outside 50HZ-500HZ can be removed as noise frames, and voice frames with a fundamental frequency between 50HZ-500HZ are retained as effective voice frames.

[0105] Step S520: Obtain a pitch frequency array corresponding to the call voice data according to the pitch frequencies of each valid voice frame.

[0106] Assuming that there are 5 valid voice frames in a call voice data, the fundamental frequency of each valid voice frame is obtained respectively, and the fundamental frequency array corresponding to the call voice data is obtained as [90, 120, 220, 390, 440].

[0107] Step S530: Acquire a plurality of preset pitch frequency intervals, and obtain the pitch frequency distribution statistics corresponding to the call voice data according to the pitch frequency array and the pitch frequency intervals.

[0108] The preset multiple fundamental frequency interval ranges can be determined according to the preset effective fundamental frequency interval. For example, the effective fundamental frequency interval 50HZ-500HZ can be divided into several equal parts, and then the frequency of the fundamental frequency falling in each interval is counted to obtain the fundamental frequency distribution statistics corresponding to the call voice data.

[0109] Assuming that the preset multiple fundamental frequency intervals are [50-100), [100-150), [150-200), [200-250), [250-300), [300-350), [350-400), [400-450), [450-500], therefore, for the fundamental frequency array [90, 120, 220, 390, 440], the fundamental frequency distribution statistic obtained is (1, 1, 0, 1, 0, 0, 1, 1, 0), which is a 9-dimensional eigenvector.

[0110] Step S260: Obtain a feature vector corresponding to the call voice data according to the frame ratio distribution statistics and the fundamental frequency distribution statistics.

[0111] By concatenating the frame ratio distribution statistics and the pitch frequency distribution statistics, a multi-dimensional feature vector for characterizing the call voice data can be obtained.

[0112] In actual operation, the above frame ratio interval range and fundamental frequency interval range can be set differently as needed, for example, they can be divided into finer interval ranges according to finer granularity, so as to achieve better feature extraction effect. The above frame ratio interval range is only for illustration and is not specifically limited in this example implementation.

[0113] In this example implementation, the call voice data to be identified is converted into a multi-dimensional feature vector based on multiple measurement factors. Therefore, other voice identification functions can also be completed based on the multi-dimensional feature vector to achieve the expansion of more business functions.

[0114] In step S120, the feature vector is input into a pre-trained outbound user classification model to obtain the gender attribute of the outbound user in the call voice data.

[0115] After extracting the feature vector corresponding to the call voice data, the gender attribute of the outbound user in the call voice data can be obtained by inputting the feature vector into a pre-trained outbound user classification model. The gender attribute may include male and female.

[0116] In this example implementation, the principle of model establishment is: the vector and the corresponding true label are sent to the classification algorithm for model training. The algorithm continuously adjusts its internal parameters until the classification result of the model is close to the true label. At this time, the optimal correlation model between the feature vector and the label is obtained. When a new vector is sent to the model, the model will estimate the most relevant label based on the vector. In this example implementation, by using traditional machine learning methods, no large-scale hardware support is required. After multiple analyses combined with historical data, a certain degree of accuracy can be achieved.

[0117] In this example implementation, Figure 6 As shown, the training method of the outbound user classification model may specifically include the following steps:

[0118] Step S610: Acquire first voice training data corresponding to the outbound user classification model, and acquire outbound call attributes corresponding to each first voice training data.

[0119] A batch of voice data is collected as the first voice training data corresponding to the outbound user classification model, wherein each data in the first voice training data is labeled with two layers of labels, the first layer of labels is the outbound attribute, and the second layer of labels is the answering attribute. For example, in the first layer of labels, the label of the outbound caller who is male is "M-", and the label of the outbound caller who is female is "F-", and the first layer of label set is recorded as label set L1; in the second layer of labels, the label of the outbound caller who is male and the answering party who is male is "MM", the label of the outbound caller who is male and the answering party who is female is "MF", the label of the outbound caller who is female and the answering party who is male is "FM", and the label of the outbound caller who is female and the answering party who is female is "FF", and the second layer of label set is recorded as label set L2.

[0120] When training the outbound user classification model, only the first layer of labels, that is, the outbound attribute corresponding to each voice training data, needs to be obtained, and the outbound user classification model is trained through the label set L1.

[0121] Step S620: Determine the feature vectors corresponding to each first voice training data, and obtain an outbound call feature vector group according to the feature vectors corresponding to each first voice training data.

[0122] Method for extracting feature vectors corresponding to speech training data Figure 2 The method for extracting the feature vector corresponding to the call voice data is the same as that in the above method, which will not be described in detail here. Each voice in the batch of voice training data is converted into a feature vector of fixed length to obtain an outbound call feature vector group, which is recorded as a multi-dimensional feature vector group a.

[0123] Step S630: Obtain an outbound user classification model by training according to the outbound call attributes, the outbound call feature vector group and the preset classification algorithm corresponding to the first voice training data.

[0124] A classification model for the gender attributes of outbound users is established by using a preset classification algorithm and a multidimensional feature vector group a, combined with the label set L1 of the first layer of labels. That is, a classification model for identifying the gender attributes of outbound users is trained, which is denoted as model A. Model A can be used as a basic model for classifying the gender attributes of outbound users using input voice feature vectors.

[0125] Among them, the preset classification algorithm can be one or more of xgboost (eXtreme Gradient Boosting), lightGBM (Light Gradient Boosting Machine, distributed gradient boosting machine), randomforest (random forest), catboost (Category Boosting, gradient boosting algorithm library), or other types of classification algorithms, which are not specifically limited in this example implementation.

[0126] In step S130, a target incoming call user classification model is determined from two pre-trained incoming call user classification models according to the gender attribute of the outgoing call user.

[0127] In this example implementation, based on the gender attribute of the outbound user, one of the two pre-trained answering user classification models is selected as the target answering user classification model. The answering user classification model may include a first answering user classification model and a second answering user classification model, the first answering user classification model is used to identify the gender attribute of the answering user corresponding to a male outbound user, and the second answering user classification model is used to identify the gender attribute of the answering user corresponding to a female outbound user.

[0128] In this example implementation, Figure 7 As shown, the training method of the answering user classification model may specifically include the following steps:

[0129] Step S710: Obtain the second voice training data corresponding to the answering user classification model, and obtain the answering attributes corresponding to each second voice training data.

[0130] A batch of new voice data is collected as the second voice training data corresponding to the answering user classification model. Each data in the second voice training data is marked with two layers of labels, the first layer of labels are outbound call attributes, and the second layer of labels are answering attributes. The specific data labels are the same as the labels in step S610. The first layer of label set is recorded as label set L1, and the second layer of label set is recorded as label set L2. When training the answering user classification model, it is necessary to obtain the second layer of labels, that is, the answering attributes corresponding to each voice training data, and train the answering user classification model through the label set L2.

[0131] Step S720: Determine the feature vectors corresponding to each second voice training data, and obtain a listening feature vector group according to the feature vectors corresponding to each second voice training data.

[0132] Method for extracting feature vectors corresponding to speech training data Figure 2 The method for extracting the feature vector corresponding to the call voice data is the same as that in the above method, which will not be repeated here. Each voice in the batch of voice training data is converted into a feature vector of fixed length to obtain a call feature vector group, which can be recorded as a multi-dimensional feature vector group b.

[0133] Step S730: Using the outbound call user classification model, the answering feature vector group is divided into a first answering feature vector group and a second answering feature vector group according to the outbound call attribute.

[0134] The outbound call user classification model is used to divide the answering feature vector group into two categories according to the outbound call attributes, wherein the first answering feature vector group includes male calling male and male calling female, and the second answering feature vector group includes female calling male and female calling female.

[0135] Step S740: Obtain a first answering user classification model through training according to the answering attributes corresponding to the second voice training data, the first answering feature vector group, and a preset classification algorithm.

[0136] Model training is performed based on the answering attributes corresponding to the second voice training data, the first answering feature vector group in the multidimensional feature vector group b, and the preset classification algorithm to obtain the first answering user classification model B1 for identifying the gender attribute of the answering user in the group where the outgoing caller is male "M-".

[0137] Step S750: Obtain a second answering user classification model through training according to the answering attributes corresponding to the second voice training data, the second answering feature vector group, and a preset classification algorithm.

[0138] Model training is performed based on the answering attributes corresponding to the second voice training data, the second answering feature vector group in the multidimensional feature vector group b, and the preset classification algorithm to obtain a second answering user classification model B2 for identifying the gender attribute of the answering user in the group where the outgoing caller is female "F-".

[0139] In step S140, the feature vector is input into a target answering user classification model to obtain the gender attribute of the answering user in the call voice data.

[0140] Finally, the feature vector corresponding to the call voice data is input into the selected target answering user classification model to obtain the gender attribute of the answering user in the call voice data. At this point, the gender attribute of the outbound user and the gender attribute of the answering user can be identified.

[0141] like Figure 8 The figure is a complete flow chart of a method for identifying user attributes in a specific implementation of the present disclosure, which is an example of the above steps in this example implementation. The specific steps of the flow chart are as follows:

[0142] Step S802: Collect historical outbound voice data.

[0143] Step S804: Mark the historical outbound voice data.

[0144] The data frames in the historical outbound voice data are labeled to obtain male voice training samples, female voice training samples and silent training samples.

[0145] Step S806: Extract speech features.

[0146] The corresponding MFCC, differential MFCC, second-order differential MFCC, fundamental frequency F0, first formant frequency F1, second formant frequency F2, third formant frequency F3, fourth formant frequency F4 and other speech features are extracted from the male voice training samples, female voice training samples and silent training samples respectively.

[0147] Step S808: Build a basic model library.

[0148] Through the above speech feature training, a male voice basic model representing male voice, a female voice basic model representing female voice, and a silent voice basic model representing silent voice are obtained.

[0149] Step S810: Obtain outbound voice training data.

[0150] Step S812: Perform matching annotation based on the basic model library.

[0151] Match and annotate the outbound voice training data based on the basic model library.

[0152] Step S814: Speech segmentation.

[0153] The outbound voice training data is segmented according to the positions of the voice frames with the attribute of "silence".

[0154] Step S816: Obtain the speech frame ratio.

[0155] The male voice frame ratio or the female voice frame ratio in each speech data segment is determined.

[0156] Step S818: Determine the frame ratio distribution statistic.

[0157] The male voice frame ratio distribution statistics are obtained according to the male voice frame ratio array and the preset frame ratio interval range, and the female voice frame ratio distribution statistics are obtained according to the female voice frame ratio array and the preset frame ratio interval range.

[0158] Step S820: Obtain outbound voice training data.

[0159] Step S822: Speech preprocessing.

[0160] Noise frames are removed according to the preset effective fundamental frequency range.

[0161] Step S824: Extract speech features.

[0162] Extract the fundamental frequency F0 corresponding to the outbound voice training data.

[0163] Step S826: Determine the fundamental frequency distribution statistics.

[0164] The fundamental frequency distribution statistics corresponding to the call voice data are obtained according to the fundamental frequency array and a plurality of preset fundamental frequency interval ranges.

[0165] Step S828. Obtain a multi-dimensional feature vector.

[0166] The frame ratio distribution statistics and the pitch frequency distribution statistics are concatenated to obtain a multi-dimensional feature vector.

[0167] Step S830: Training the classification model.

[0168] Train the outbound user classification model A and the incoming user classification models B1 and B2.

[0169] Step S832: Extract the feature vector of the speech to be recognized.

[0170] Extract the multi-dimensional feature vector corresponding to the speech to be recognized.

[0171] Step S834: Use the classification model to identify the user's gender.

[0172] The gender of the outgoing user is identified through model A, and based on the gender of the outgoing user, model B1 or model B2 is selected to identify the gender of the answering user.

[0173] Step S836: Obtain recognition results.

[0174] The above steps S802 to S830 are offline training steps, and steps S832 to S836 are online recognition steps.

[0175] like Fig. 9 The figure shows the overall architecture of the classification model in a specific implementation of the present invention. The feature vector corresponding to the voice to be recognized is input into model A to determine the gender of the outbound user. Then, model B1 or model B2 is selected according to the gender of the outbound user to recognize the gender of the answering user.

[0176] like Fig.10 The figure shows the architecture of the distributed service engine in a specific embodiment of the present disclosure. To facilitate the application of the model, the obtained basic speech model and classification model can be loaded into the online recognition service for online service. The online recognition method of the model is:

[0177] (1) The algorithm is packaged into an image. The trained model is packaged into a docker (development platform) image to form an independent microservice. The calculation results and speech index are stored in the database.

[0178] (2) Parallel processing. Divide into multiple services, design a sharding algorithm, start n service engines, and each engine processes voice data with an index of the remainder of N / n, where N is the total amount of unprocessed data, thereby improving processing efficiency.

[0179] In this example implementation, the model algorithm is encapsulated and parallelized to enable it to process large-scale voice data, with portability and ease of use. In addition, since the solution is simple to deploy and highly portable, a lightweight deployment method can be used to achieve dynamic expansion of the recognition service.

[0180] It should be noted that although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0181] Furthermore, the present disclosure also provides a device for identifying user attributes. Fig.11As shown, the user attribute recognition device may include a feature vector acquisition module 1110, an outbound call attribute determination module 1120, an answering model determination module 1130, and an answering attribute determination module 1140. Among them:

[0182] The feature vector acquisition module 1110 may be used to acquire the call voice data to be recognized and extract the feature vector corresponding to the call voice data;

[0183] The outbound call attribute determination module 1120 may be used to input the feature vector into a pre-trained outbound call user classification model to obtain the gender attribute of the outbound call user in the call voice data;

[0184] The answering model determination module 1130 may be used to determine a target answering user classification model from two pre-trained answering user classification models according to the gender attribute of the outbound calling user;

[0185] The answering attribute determination module 1140 can be used to input the feature vector into the target answering user classification model to obtain the gender attribute of the answering user in the call voice data.

[0186] In some exemplary embodiments of the present disclosure, the feature vector acquisition module 1110 may include a speech frame attribute labeling unit, a speech segment division unit, a speech frame ratio determination unit, a frame ratio distribution determination unit, a fundamental frequency distribution determination unit, and a feature vector determination unit. Among them:

[0187] The speech frame attribute labeling unit may be used to label the attributes of each speech frame in the call speech data by using a pre-trained speech basic model, wherein the attributes of the speech frame include male voice, female voice and silence;

[0188] The voice segment division unit may be used to divide the call voice data into a plurality of voice data segments according to the positions of the voice frames with the attribute of silence;

[0189] The speech frame ratio determination unit may be used to determine the male voice frame ratio or the female voice frame ratio in each speech data segment according to the attributes of each speech frame;

[0190] The frame ratio distribution determination unit may be used to obtain a frame ratio distribution statistic of the call voice data according to the male voice frame ratio or the female voice frame ratio in each voice data segment;

[0191] The fundamental frequency distribution determination unit may be used to determine the fundamental frequency distribution statistics of the call voice data according to the fundamental frequency corresponding to each voice frame;

[0192] The feature vector determination unit may be configured to obtain a feature vector corresponding to the call voice data according to the frame ratio distribution statistics and the fundamental frequency distribution statistics.

[0193] In some exemplary embodiments of the present disclosure, the frame ratio distribution determination unit may include a speech frame ratio array determination unit, a male and female voice frame ratio distribution determination unit, and a frame ratio distribution statistics determination unit.

[0194] The voice frame ratio array determination unit may be used to obtain a male voice frame ratio array of the call voice data according to the male voice frame ratio in each voice data segment, or to obtain a female voice frame ratio array of the call voice data according to the female voice frame ratio in each voice data segment;

[0195] The male-female voice frame ratio distribution determination unit can be used to obtain a plurality of preset frame ratio intervals, and obtain male voice frame ratio distribution statistics according to the male voice frame ratio array and the frame ratio interval, or obtain female voice frame ratio distribution statistics according to the female voice frame ratio array and the frame ratio interval;

[0196] The frame ratio distribution statistic determining unit may be configured to use the male voice frame ratio distribution statistic or the female voice frame ratio distribution statistic as the frame ratio distribution statistic of the call speech data.

[0197] In some exemplary embodiments of the present disclosure, the pitch frequency distribution determining unit may include a noise frame removing unit, a pitch frequency array determining unit, and a pitch frequency distribution statistics determining unit.

[0198] The noise frame removal unit may be used to remove noise frames in the call voice data according to a preset effective fundamental frequency interval to obtain effective voice frames in the call voice data;

[0199] The fundamental frequency array determination unit may be used to obtain the fundamental frequency array corresponding to the call voice data according to the fundamental frequency of each valid voice frame;

[0200] The fundamental frequency distribution statistics determination unit can be used to obtain a plurality of preset fundamental frequency intervals, and obtain the fundamental frequency distribution statistics corresponding to the call voice data according to the fundamental frequency array and the fundamental frequency interval.

[0201] In some exemplary embodiments of the present disclosure, a user attribute recognition device provided by the present disclosure may further include a speech basic model training module, and the speech basic model training module may include a training sample acquisition unit and a basic model training unit.

[0202] The training sample acquisition unit can be used to acquire historical call voice data, and obtain male voice training samples, female voice training samples and silent training samples according to the attributes of the voice frames in the historical call voice data;

[0203] The basic model training unit can be used to extract corresponding speech features from male voice training samples, female voice training samples and silent training samples respectively, and obtain male voice basic model, female voice basic model and silent basic model through speech feature training.

[0204] In some exemplary embodiments of the present disclosure, a user attribute recognition device provided by the present disclosure may further include an outbound user classification model training module, which may include a first voice training data acquisition unit, an outbound feature vector group determination unit, and an outbound user classification model training unit. Wherein:

[0205] The first voice training data acquisition unit may be used to acquire first voice training data corresponding to the outbound user classification model, and acquire outbound call attributes corresponding to each first voice training data;

[0206] The outbound call feature vector group determination unit may be used to determine the feature vectors corresponding to each first voice training data, and obtain the outbound call feature vector group according to the feature vectors corresponding to each first voice training data;

[0207] The outbound user classification model training unit can be used to train an outbound user classification model according to the outbound call attributes, the outbound call feature vector group and the preset classification algorithm corresponding to the first voice training data.

[0208] In some exemplary embodiments of the present disclosure, a user attribute recognition device provided by the present disclosure may further include an answering user classification model training module, which may include a second voice training data acquisition unit, an answering feature vector group determination unit, an answering feature vector group division unit, a first answering user classification model training unit, and a second answering user classification model training unit. Wherein:

[0209] The second voice training data acquisition unit may be used to acquire the second voice training data corresponding to the answering user classification model, and acquire the answering attributes corresponding to each second voice training data;

[0210] The answering feature vector group determination unit may be used to determine the feature vectors corresponding to each second voice training data, and obtain the answering feature vector group according to the feature vectors corresponding to each second voice training data;

[0211] The answering feature vector group division unit may be used to divide the answering feature vector group into a first answering feature vector group and a second answering feature vector group according to the outbound call attribute through the outbound call user classification model;

[0212] The first answering user classification model training unit can be used to train and obtain the first answering user classification model according to the answering attribute corresponding to the second voice training data, the first answering feature vector group and the preset classification algorithm;

[0213] The second answering user classification model training unit can be used to obtain the second answering user classification model through training according to the answering attributes corresponding to the second voice training data, the second answering feature vector group and the preset classification algorithm.

[0214] The specific details of each module / unit in the above-mentioned user attribute identification device have been described in detail in the corresponding method embodiment part, and will not be repeated here.

[0215] Fig.12 A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present invention is shown.

[0216] It should be noted that Fig.12 The computer system 1200 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0217] like Fig.12 As shown, the computer system 1200 includes a central processing unit (CPU) 1201, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1202 or a program loaded from a storage part 1208 into a random access memory (RAM) 1203. Various programs and data required for system operation are also stored in the RAM 1203. The CPU 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0218] The following components are connected to the I / O interface 1205: an input section 1206 including a keyboard, a mouse, etc.; an output section 1207 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card such as a LAN card, a modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the I / O interface 1205 as needed. A removable medium 1211, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1210 as needed, so that a computer program read therefrom is installed into the storage section 1208 as needed.

[0219] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication section 1209, and / or installed from a removable medium 1211. When the computer program is executed by a central processing unit (CPU) 1201, various functions defined in the system of the present application are executed.

[0220] It should be noted that the computer-readable medium shown in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM 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 of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0221] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0222] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the above embodiment.

[0223] It should be noted that, although several modules of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules described above can be concretized in one module. Conversely, the features and functions of one module described above can be further divided into multiple modules to be concretized.

[0224] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modifications, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure.

[0225] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for identifying user attributes, characterized in that: include: Acquire call voice data to be recognized, and extract a feature vector corresponding to the call voice data, wherein the feature vector is a multidimensional feature vector obtained according to frame ratio distribution statistics and fundamental frequency distribution statistics of voice frames with different attributes in the call voice data, and the attributes of the voice frame include male voice, female voice and silence; Input the feature vector into a pre-trained outbound caller classification model to obtain the gender attribute of the outbound caller in the call voice data, wherein each piece of data in the first voice training data used to train the outbound caller classification model contains two layers of labels, the first layer of labels is the outbound call attribute, and the second layer of labels is the answering attribute, and the outbound caller classification model is obtained by training the first layer of labels; Determine a target answering user classification model from two pre-trained answering user classification models according to the gender attribute of the outbound user, the answering user classification model comprising a first answering user classification model and a second answering user classification model, the first answering user classification model being used to identify the gender attribute of the answering user corresponding to a male outbound user, and the second answering user classification model being used to identify the gender attribute of the answering user corresponding to a female outbound user; The feature vector is input into the target answering user classification model to obtain the gender attribute of the answering user in the call voice data.

2. The method for identifying user attributes according to claim 1, characterized in that: The extracting the feature vector corresponding to the call voice data includes: Annotating the attributes of each speech frame in the call speech data by using a pre-trained speech basic model; Dividing the call voice data into a plurality of voice data segments according to the position of the voice frame with the attribute of silence; Determining the ratio of male voice frames or female voice frames in each of the voice data segments according to the attributes of each of the voice frames; Obtaining a frame ratio distribution statistic of the call voice data according to the male voice frame ratio or the female voice frame ratio in each of the voice data segments; Determine the fundamental frequency distribution statistics of the call voice data according to the fundamental frequency corresponding to each of the voice frames; A feature vector corresponding to the call voice data is obtained according to the frame ratio distribution statistics and the fundamental frequency distribution statistics.

3. The method for identifying user attributes according to claim 2, characterized in that: The obtaining of the frame ratio distribution statistics of the call voice data according to the male voice frame ratio or the female voice frame ratio in each of the voice data segments includes: Obtaining an array of male voice frame ratios of the call voice data according to the male voice frame ratios in each of the voice data segments, or obtaining an array of female voice frame ratios of the call voice data according to the female voice frame ratios in each of the voice data segments; Acquire a plurality of preset frame ratio intervals, and obtain male voice frame ratio distribution statistics according to the male voice frame ratio array and the frame ratio interval, or obtain female voice frame ratio distribution statistics according to the female voice frame ratio array and the frame ratio interval; The male voice frame ratio distribution statistic or the female voice frame ratio distribution statistic is used as the frame ratio distribution statistic of the call voice data.

4. The method for identifying user attributes according to claim 2, characterized in that: The determining of the fundamental frequency distribution statistics of the call voice data according to the fundamental frequency corresponding to each of the voice frames includes: Removing noise frames from the call voice data according to a preset effective fundamental frequency interval to obtain effective voice frames from the call voice data; Obtaining a pitch frequency array corresponding to the call voice data according to the pitch frequencies of each of the valid voice frames; A plurality of preset pitch frequency intervals are obtained, and a pitch frequency distribution statistic corresponding to the call voice data is obtained according to the pitch frequency array and the pitch frequency interval.

5. The method for identifying user attributes according to claim 2, characterized in that: The training method of the speech basic model includes: Acquire historical call voice data, and obtain male voice training samples, female voice training samples, and silent training samples according to the attributes of the voice frames in the historical call voice data; Corresponding speech features are extracted from the male voice training samples, the female voice training samples and the silent training samples respectively, and a male voice basic model, a female voice basic model and a silent basic model are obtained through the speech feature training.

6. The method for identifying user attributes according to claim 1, characterized in that: The training method of the outbound user classification model includes: Acquire the first voice training data corresponding to the outbound user classification model, and acquire the outbound call attributes corresponding to each of the first voice training data; Determine the feature vectors corresponding to each of the first voice training data, and obtain an outbound call feature vector group according to the feature vectors corresponding to each of the first voice training data; The outbound call user classification model is obtained by training according to the outbound call attributes corresponding to the first voice training data, the outbound call feature vector group and a preset classification algorithm.

7. The method for identifying user attributes according to claim 1, characterized in that: The training method of the answering user classification model includes: Acquire the second voice training data corresponding to the answering user classification model, and acquire the answering attributes corresponding to each of the second voice training data; Determine the feature vectors corresponding to each of the second voice training data, and obtain a listening feature vector group according to the feature vectors corresponding to each of the second voice training data; The outbound call user classification model is used to divide the answering feature vector group into a first answering feature vector group and a second answering feature vector group according to outbound call attributes; A first answering user classification model is obtained by training according to the answering attributes corresponding to the second voice training data, the first answering feature vector group, and a preset classification algorithm; The second answering user classification model is obtained by training according to the answering attributes corresponding to the second voice training data, the second answering feature vector group and a preset classification algorithm.

8. A user attribute identification device, characterized in that: include: A feature vector acquisition module, used to acquire call voice data to be recognized, and extract a feature vector corresponding to the call voice data, wherein the feature vector is a multidimensional feature vector obtained according to the frame ratio distribution statistics and the fundamental frequency distribution statistics of the voice frames with different attributes in the call voice data, and the attributes of the voice frame include male voice, female voice and silence; An outbound call attribute determination module, used for inputting the feature vector into a pre-trained outbound call user classification model to obtain the gender attribute of the outbound call user in the call voice data, wherein each piece of data in the first voice training data used for training the outbound call user classification model contains two layers of labels, the first layer of labels is the outbound call attribute, and the second layer of labels is the answering attribute, and the outbound call user classification model is obtained by training the first layer of labels; An answering model determination module is used to determine a target answering user classification model from two pre-trained answering user classification models according to the gender attribute of the outbound user, the answering user classification model comprising a first answering user classification model and a second answering user classification model, the first answering user classification model is used to identify the gender attribute of the answering user corresponding to a male outbound user, and the second answering user classification model is used to identify the gender attribute of the answering user corresponding to a female outbound user; The answering attribute determination module is used to input the feature vector into the target answering user classification model to obtain the gender attribute of the answering user in the call voice data.

9. An electronic device, characterized in that: include: processor; as well as A memory for storing one or more programs, which, when executed by the processor, enables the processor to implement the method for identifying user attributes as described in any one of claims 1 to 7.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for identifying user attributes according to any one of claims 1 to 7 is implemented.

Images