CN111798857A - Information identification method and device, electronic equipment and storage medium - Google Patents

Abstract

The application provides an information identification method, an information identification device, an electronic device and a storage medium, wherein the method comprises the following steps: acquiring audio to be identified; determining a feature vector representing acoustic features of the audio to be recognized; obtaining a service label corresponding to the audio to be recognized based on the feature vector and a pre-trained information recognition model; and determining the service information matched with the service label according to the service label. According to the service tag identification method and the service tag identification device, the service tag corresponding to the audio to be identified is identified through the information identification model, and then the service information is matched for the user according to the service tag, so that more suitable service is provided for the user under the condition that the user does not sense, the user can enjoy the service without bringing burden to the user, and the user experience is improved.

Description

An information identification method, device, electronic device and storage medium

technical field

The present application relates to the field of Internet technologies, and in particular, to an information identification method, apparatus, electronic device, and storage medium.

Background technique

With the development of science and technology, Internet technology has penetrated into all aspects of people's lives, such as taxis, shopping, takeout orders, etc. through the Internet. However, the information on the Internet is increasing explosively, and the frequency of information updates is getting faster and faster. , which makes all kinds of information exist on the network. When a user purchases a service through the network, he needs to browse a large amount of information to find the desired service, which will undoubtedly reduce the user experience. Therefore, for service providers, how to provide users with more suitable services without burdening users is an urgent problem to be solved.

SUMMARY OF THE INVENTION

In view of this, the purpose of the embodiments of the present application is to provide an information identification method, apparatus, electronic device and storage medium, which can provide more suitable services and improve user experience without the user's perception.

In a first aspect, an embodiment of the present application provides an information recognition method, the method includes: acquiring audio to be recognized; determining a feature vector representing an acoustic feature of the audio to be recognized; based on the feature vector and pre-trained information Identifying the model to obtain a service tag corresponding to the audio to be recognized; and determining service information matching the service tag according to the service tag.

In an optional embodiment, the service tag includes an age recognition result; the step of obtaining the service tag corresponding to the audio to be recognized based on the feature vector and a pre-trained information recognition model includes: The feature vector is input into the information recognition model, and the probability value of the feature vector in each preset age interval is obtained; according to the probability value of the feature vector in each preset age interval, the age recognition corresponding to the to-be-recognized audio is determined result.

In an optional embodiment, the information identification model includes a first network and a second network; the feature vector is input into the information identification model to obtain the probability value of the feature vector in each preset age interval. The steps include: inputting the feature vector into the information recognition model, and using the first network to perform feature extraction on the feature vector to obtain a first feature map; The feature map is input into the second network and classified to obtain the probability value of the feature vector in each preset age range.

In an optional embodiment, the first network includes a convolution layer and a statistical pooling layer; the feature vector is input into the information recognition model, and the feature vector is characterized by using the first network. The step of extracting and obtaining the first feature map includes: inputting the feature vector into the information recognition model, and using the convolution kernel of the convolution layer and the convolution kernel offset to perform convolution processing on the feature vector, obtaining a first output feature map; using the statistical pooling layer to perform statistical pooling processing on the first output feature map to obtain a first feature map.

In an optional implementation manner, the second network includes a fully connected layer and a multi-class logistic regression layer; the first feature map extracted from the first network is input to the second network and classified, The step of obtaining the probability value of the feature vector in each preset age interval includes: inputting the first feature map extracted from the first network into the fully connected layer, and using the fully connected layer to analyze the first feature Dimensionality reduction is performed on the graph to obtain a first vector; the first vector is input into the multi-class logistic regression layer to obtain the probability value of the feature vector in each preset age interval.

In an optional embodiment, the step of determining the age recognition result corresponding to the audio to be recognized according to the probability value of the feature vector in each preset age interval includes: placing the feature vector in each preset age range. It is assumed that the probability values of the age intervals are weighted and averaged to obtain the age recognition result corresponding to the audio to be recognized.

In an optional implementation manner, the server pre-stores a plurality of preset age intervals and the service content corresponding to each preset age interval; The steps include: determining a target age interval to which the age recognition result belongs from the plurality of preset age intervals according to the age recognition result; acquiring service content corresponding to the target age interval, and obtaining the service content corresponding to the service The tag matches the service information.

In an optional embodiment, the service label includes a gender recognition result; the step of obtaining the service label corresponding to the audio to be recognized based on the feature vector and a pre-trained information recognition model includes: The feature vector is input into the information recognition model, and the probability value of the feature vector in each preset gender category is obtained; the pre-trained hidden Markov model is used to constrain the probability value of the feature vector in each preset gender category, A gender recognition result corresponding to the audio to be recognized is obtained.

In an optional embodiment, the information identification model includes a third network and a fourth network; the feature vector is input into the information identification model to obtain the probability value of the feature vector in each preset gender category. The steps include: inputting the feature vector into the information recognition model, using the third network to perform feature extraction on the feature vector, and obtaining a second feature map; using the fourth network to extract the third network The outputted second feature map is classified to obtain the probability value of the feature vector in each preset gender category.

In an optional embodiment, the third network includes a convolution layer and a long-term and short-term memory layer; the feature vector is input into the information recognition model, and the feature vector is characterized by the third network. The step of extracting and obtaining the second feature map includes: inputting the feature vector into the information recognition model, and performing convolution processing on the feature vector by using the convolution kernel and the convolution kernel offset of the convolution layer, Obtain a second output feature map; use the long and short-term memory layer to capture sequence information of the second output feature map to obtain a second feature map.

In an optional implementation manner, the fourth network includes a fully connected layer and a multi-class logistic regression layer; the fourth network is used to classify the second feature map output by the third network to obtain the The step of determining the probability value of the feature vector in each preset gender category includes: inputting the second feature map output by the third network into the fully connected layer, and using the fully connected layer to reduce the dimension of the second feature map, Obtain a second vector; use the multi-class logistic regression layer to process the second vector to obtain the probability value of the feature vector in each preset gender category.

In an optional implementation manner, the server pre-stores multiple preset gender categories and service content corresponding to each preset gender category; The step includes: acquiring target service content corresponding to the gender identification result from the plurality of service contents according to the gender identification result, and obtaining service information matching the service tag.

In an optional embodiment, the service label includes an identity verification result; the step of obtaining the service label corresponding to the audio to be recognized based on the feature vector and a pre-trained information recognition model includes: obtaining the The standard audio corresponding to the audio to be recognized; determine the feature vector representing the acoustic features of the standard audio; input the feature vector corresponding to the audio to be recognized and the feature vector corresponding to the standard audio into the information recognition model to obtain the The similarity score between the to-be-recognized audio and the standard audio is determined; the identity verification result corresponding to the to-be-recognized audio is determined according to the similarity score, and the identity verification result includes a verification pass result or a verification failure result.

In an optional embodiment, the server pre-stores the identity verification result and the service content corresponding to the identity verification result; the step of determining the service information matching the service tag according to the service tag includes: When the identity verification result is a verification pass result, acquire the service content corresponding to the verification pass result as the service information matching the service tag; when the identity verification result is a verification failure result, acquire the verification The service content corresponding to the failure result is used as the service information matching the service tag.

In a second aspect, an embodiment of the present application further provides an information identification device, the device includes: an audio acquisition module, configured to acquire audio to be identified; a first execution module, configured to determine an acoustic feature characterizing the audio to be identified The second execution module is used to obtain the corresponding service tag of the audio to be identified based on the feature vector and the pre-trained information identification model; the third execution module is used to determine the corresponding service tag according to the service tag. The service tag matches the service information.

In a third aspect, an embodiment of the present application further provides an electronic device, including: a processor, a storage medium, and a bus, where the storage medium stores machine-readable instructions executable by the processor, and when the electronic device runs, The processor communicates with the storage medium through a bus, and the processor executes the machine-readable instructions to execute the steps of the above-mentioned information identification method at runtime.

In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the above-mentioned information identification method when the computer program is run by a processor.

The embodiment of the present application obtains the audio to be recognized of the user, determines a feature vector representing the acoustic feature of the audio to be recognized, and then uses a pre-trained information recognition model to identify the feature vector to obtain a service tag corresponding to the audio to be recognized, Finally, the service information matching the service tag is determined according to the obtained service tag, so as to provide the user with a more suitable service without the user's perception, so that the user can enjoy the service without burdening the user. user experience.

Another embodiment of the present application inputs the determined feature vector into a pre-trained information recognition model, identifies the probability value of the feature vector in each preset age range through the information recognition model, and then determines the corresponding audio frequency to be recognized according to the obtained probability value That is, the service tag of the audio to be recognized is obtained as the age recognition result, so that the service information matching the age of the user can be further determined according to the age recognition result.

Another embodiment of the present application inputs the determined feature vector into a pre-trained information identification model, identifies the probability value of the feature vector in each preset gender category through the information identification model, and then uses the pre-trained hidden Markov model to obtain The probability value is constrained to determine a more accurate gender recognition result, that is, the service label of the audio to be recognized is the gender recognition result, so that the service information that matches the user's gender can be further determined according to the gender recognition result. .

Another embodiment of the present application obtains the standard audio corresponding to the audio to be recognized, determines a feature vector representing the acoustic features of the standard audio, and then inputs the feature vector corresponding to the audio to be recognized and the feature vector corresponding to the standard audio for information identification. model, obtain the similarity score of the audio to be recognized and the standard audio, and finally determine the authentication result according to the similarity score, that is, get the service tag of the audio to be recognized as the authentication result, so that it can be further based on the authentication result. Determine the service information that matches the user's identity.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present application more clearly, the following drawings will briefly introduce the drawings that need to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore do not It should be regarded as a limitation of the scope, and for those of ordinary skill in the art, other related drawings can also be obtained according to these drawings without any creative effort.

FIG. 1 shows an architecture diagram of an information identification system provided by an embodiment of the present application;

2 shows a flowchart of an information identification method provided by an embodiment of the present application;

Fig. 3 is a sub-step flowchart of step S103 shown in Fig. 2;

Fig. 4 is another sub-step flowchart of step S103 shown in Fig. 2;

Fig. 5 is another sub-step flowchart of step S103 shown in Fig. 2;

FIG. 6 shows a schematic diagram of an information identification device provided by an embodiment of the present application;

FIG. 7 shows a structural block diagram of an electronic device provided by an embodiment of the present application.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. The drawings are only for the purpose of illustration and description, and are not used to limit the protection scope of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented in accordance with some embodiments of the application. It should be understood that the operations of the flowcharts may be performed out of order and that steps without logical context may be performed in reverse order or concurrently. In addition, those skilled in the art can add one or more other operations to the flowchart, and can also remove one or more operations from the flowchart under the guidance of the content of the present application.

In addition, the described embodiments are only some of the embodiments of the present application, but not all of the embodiments. The components of the embodiments of the present application generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Thus, the following detailed description of the embodiments of the application provided in the accompanying drawings is not intended to limit the scope of the application as claimed, but is merely representative of selected embodiments of the application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared later, but does not exclude the addition of other features.

FIG. 1 is a block diagram of an information identification system 100 according to some embodiments of the present application. For example, the information identification system 100 may be an online transportation service platform for transportation services such as taxis, ride-hailing services, express cars, carpooling, bus services, driver rentals, shuttle services, bike sharing, or any combination thereof ; It can also be an online delivery service platform for delivery services such as takeaway delivery, fresh food delivery, intra-city delivery, logistics delivery, or any combination thereof; it can also be used for online supermarkets, online malls, and online bookstores. type of shopping services, or any combination of e-commerce service platforms. The information identification system 100 may include one or more of a server 110 , a network 120 , a client 130 and a database 140 , and the server 110 may include a processor for executing instruction operations.

In some embodiments, server 110 may include a processor. The processor may process information and/or data related to the service request to perform one or more functions described herein. In some embodiments, a processor may include one or more processing cores (eg, a single-core processor (S) or a multi-core processor (S)). By way of example only, the processor may include a Central Processing Unit (CPU), an Application Specific Integrated Circuit (ASIC), an Application Specific Instruction-set Processor (ASIP), a graphics processing unit ( Graphics Processing Unit (GPU), Physical Processing Unit (PPU), Digital Signal Processor (DSP), Field Programmable Gate Array (FPGA), Programmable Logic Device (Programmable Logic Device) Logic Device, PLD), controller, microcontroller unit, Reduced Instruction Set Computing (RISC), or microprocessor, etc., or any combination thereof.

In some embodiments, the device type corresponding to the client 130 may be a mobile device, such as a smart home device, a wearable device, a smart mobile device, a virtual reality device, or an augmented reality device, etc. It may also be a tablet computer, a laptop Upper computer, or built-in equipment in motor vehicles, etc.

In some embodiments, database 140 may be connected to network 120 to communicate with one or more components in information identification system 100 (eg, server 110, client 130, etc.). One or more components in information identification system 100 may access data or instructions stored in database 140 via network 120 . In some embodiments, database 140 may be directly connected to one or more components in information identification system 100 , or database 140 may be part of server 110 .

For ease of understanding, the following embodiments mainly describe the application scenario of "online car-hailing" as an example. For those skilled in the art, the general principles defined herein may be applied to other embodiments and application scenarios without departing from the spirit and scope of the present application. Although this application is mainly described around the information identification of taxis, it should be understood that this is only an exemplary embodiment.

Please refer to FIG. 2. FIG. 2 shows a flowchart of an information identification method provided by an embodiment of the present application. The method can be executed by the server 110 in the information identification system 100, and includes the following steps:

Step S101, acquiring the audio to be recognized.

In an embodiment of the present application, according to different application scenarios of the information identification method, the audio to be identified can be obtained in different ways. For example, if the information identification method is applied to the online car-hailing scene, passengers and drivers can be obtained through the operator. The audio of the telephone communication between the two parties can also be obtained from the audio sent by the passenger and the driver through the taxi application. If the information of the passenger needs to be identified, the passenger audio can be extracted from the obtained audio to obtain the audio to be identified; When the driver performs information identification, the driver audio can be extracted from the acquired audio to obtain the audio to be recognized.

In another embodiment of the present application, since the information identification model is easily disturbed by channel changes, for example, the same speaker is easily identified as two persons by the information identification model under the telephone channel and the mobile wireless channel. In the audio process, this problem can be overcome by introducing channel compensation technology.

That is to say, the user's audio in different channels can be collected, and each channel has an independent channel model established in advance. Based on the channel model, through channel mapping, algorithms such as Maximum Likelihood Linear Regression (MLLR) are used. By mapping the audio of different channels to the same standard feature space, the to-be-recognized audio after channel compensation can be obtained, thereby realizing the recognition of the same speaker in multiple channels. For example, the audio of the user in the telephone channel and the mobile wireless channel can be collected. Specifically, the audio of the telephone communication between the passenger and the driver obtained through the operator is the audio of the telephone channel. The obtained passenger and driver through the taxi application The transmitted audio is the audio of the mobile wireless channel. At the same time, a large amount of statistical data can be collected for the telephone channel and the mobile wireless channel in advance to establish a channel model, and then the channel model is used to establish a mapping for the telephone channel and the mobile wireless channel, so that The audio to be identified can be obtained by mapping the audio of the telephone channel and the mobile wireless channel to the same standard feature space.

Step S102, determining a feature vector representing the acoustic feature of the audio to be recognized.

In an embodiment of the present application, the acoustic feature includes at least one of the Mel-frequency cepstral coefficient MFCC feature and the bottleneck feature BNF feature, and the feature vector representing the acoustic feature of the audio to be identified may be at least one of the MFCC vector and the BNF vector. one.

As an embodiment, the method for obtaining the MFCC vector of the audio to be recognized may include the following steps: firstly, pre-emphasizing the audio to be recognized through a high-pass filter to obtain a pre-emphasized audio signal, thereby improving the audio frequency to be recognized. The high-frequency part makes its spectrum flat; second, the pre-emphasized audio to be recognized is divided into frames and windows. Specifically, the audio to be recognized is first divided into multiple frames, and each frame contains the audio to be recognized. There are N sampling points of , and there is an overlapping area between two adjacent frames, and then each frame is substituted into the window function, and the value outside the window is set to 0; Fourier transform to obtain the frequency spectrum of the audio to be recognized; fourthly, use the Mel filter to filter the audio to be recognized after the fast Fourier transform; fifthly, to filter the audio to be recognized after filtering by the Mel filter; Logarithmic energy parameter operation; sixth, perform discrete cosine transform on the to-be-identified audio after the logarithmic energy parameter operation, to obtain the MFCC vector of the to-be-identified audio.

In addition, the method for obtaining the BNF vector of the audio to be recognized is similar to the above-mentioned method for obtaining the MFCC vector of the audio to be recognized, and will not be repeated here.

Step S103, based on the feature vector and the pre-trained information identification model, obtain a service tag corresponding to the audio to be identified.

In an embodiment of the present application, after determining the feature vector representing the acoustic features of the audio to be recognized, the feature vector can be input into a pre-trained information recognition model, and the information recognition model can be used to obtain a service tag corresponding to the audio to be recognized. It can include at least one of age recognition results, gender recognition results, and identity verification results, that is, based on the feature vector of the user's audio to be recognized, the information recognition model can be used to recognize the user's age and gender, and can identify the user's age and gender. The actual identity of the audio is verified, and the process of obtaining the service tag corresponding to the audio to be recognized is described in detail below.

In some embodiments, when the service tag includes the age recognition result, please refer to FIG. 3 , the process of obtaining the age recognition result based on the feature vector and the information recognition model may include sub-steps S1031 to S1032, which are described in detail below:

Sub-step S1031, the feature vector is input into the information identification model, and the probability value of the feature vector in each preset age interval is obtained.

In one embodiment, when the service tag includes an age recognition result, an information recognition model may be used to perform age recognition on the audio to be recognized. In this case, the information recognition model may include a Gaussian Mixed Model (GMM), a neural network model, and the like, or a combination thereof; the neural network model may include BP (Back Propagation Neural Network, back propagation) neural network, convolutional neural network (Convolutional Neural Networks, CNN), recurrent neural network (Recurrent Neural Network, RNN), long and short-term memory network ( Long ShortTerm Memory Network, LSTM) and so on.

As an implementation manner, the information identification model may include a first network and a second network, the first network is used for feature extraction of the feature vector, the second network is used for outputting the probability value of the feature vector in each preset age interval, and the prediction It is assumed that the age category can be flexibly set by the user according to the actual situation, for example, three age ranges of 10-20 years old, 20-30 years old, and 30-40 years old are preset, which are not limited here. At this time, the process of inputting the feature vector into the information recognition model to obtain the probability value of the feature vector in each preset age range specifically includes:

The first step is to input the feature vector into the information recognition model, and use the first network to perform feature extraction on the feature vector to obtain a first feature map.

In one embodiment, after the feature vector is input into the information recognition model, the first network of the information recognition model may perform feature extraction on the feature vector, and the first network may include a convolutional layer, a convolutional layer+pooling layer (Pooling layer), convolution layer + pooling layer + random drop layer (Dropout layer), fully connected layer (Fully Connected layer, FC layer), etc. The specific settings of the first network can be flexibly adjusted by the user according to the actual situation. This will not be repeated here.

As an embodiment, the first network may include a convolution layer and a statistical pooling layer. In this case, the feature vector is input into the information recognition model, and the first network is used to perform feature extraction on the feature vector to obtain the first feature map. Can include:

First, the feature vector is input into the information recognition model, and the feature vector is convolved with the convolution kernel and convolution kernel offset of the convolution layer to obtain the first output feature map. The convolution layer here can be multi-layered, and the specific number of layers of the convolution layer and the size of the convolution kernel are not limited. The convolution kernel of the convolution layer can be used to perform the convolution sum operation on the input feature vector and add offset, and then output the result through the excitation function to obtain the first output feature map.

Then, the statistical pooling layer is used to perform statistical pooling processing on the first output feature map to obtain the first feature map. The statistical pooling layer can capture the statistical information between all frames of the audio to be recognized, which is conducive to modeling global information. The statistical pooling layer here can be multi-layered, and the specific number of layers and the size of the receptive field of the statistical pooling layer are not limited. In the next layer, the first feature map is obtained.

In the second step, the first feature map extracted from the first network is input into the second network and classified to obtain the probability value of the feature vector in each preset age interval.

In one embodiment, after the first feature map is extracted by the first network, the first feature map is input into the second network for classification, and the probability value of the feature vector in each preset age interval is output. The second network may include Fully connected layer + multi-class logistic regression layer (Softmax layer), 1*1 convolution layer + multi-class logistic regression layer, global pooling layer (GlobalAverage Pooling, GAP) + multi-class logistic regression layer, etc., the second network The specific settings can be flexibly adjusted by the user according to the actual situation, which will not be repeated here.

As an embodiment, the second network may include a fully connected layer and a multi-class logistic regression layer. In this case, the first feature map extracted from the first network is input into the second network and classified, and the feature vector is obtained in each prediction. The process of setting the probability value of the age interval can include:

First, the first feature map extracted from the first network is input to the fully connected layer, and the fully connected layer is used to reduce the dimension of the first feature map to obtain the first vector. The first vector is a one-dimensional vector, and the fully connected layer here may be multiple layers, and the specific number of layers of the fully connected layer is not limited. The fully connected layer may be used to flatten the first feature map into a one-dimensional vector.

Then, the first vector is input into the multi-class logistic regression layer to obtain the probability value of the feature vector in each preset age interval. The multi-class logistic regression layer is the final output layer of the information recognition model, which can output values between 0 and 1, representing the probability value of the feature vector belonging to each preset age interval. For example, the output probability of 10-20 years old is 0, 20 The probability of -30 years old is 0.6, and the probability of 30-40 years old is 0.4. Assuming that the feature vector is R, the probability that the feature vector belongs to the jth preset age category can be obtained by the following formula:

其中，P(y⁽ⁱ⁾)为特征矢量属于第j个预设年龄类别的概率，ω为权重，b为偏差。

Among them, P(y ⁽ⁱ⁾ ) is the probability that the feature vector belongs to the jth preset age category, ω is the weight, and b is the bias.

In one embodiment, before using the information identification model to identify the age of the user, the information identification model needs to be trained, and the training process of the information identification model may include the following steps:

First, obtain a plurality of first audio samples and age information of each first audio sample, where the age information is a specific age, for example, 27.3 years old. The plurality of first audio samples include audio samples of all age ranges that the information recognition model can recognize. For example, if the information recognition model is used to recognize multiple age ranges such as 10-20 years old, 20-30 years old, 30-40 years old , the multiple first audio samples include audio samples corresponding to multiple age ranges, such as 10-20 years old, 20-30 years old, and 30-40 years old.

Then, according to the age labeling information of each first audio sample and each preset age interval, the label of each first audio sample is obtained. The preset age interval may include all age intervals that the information recognition model can recognize. For example, if the information recognition model can recognize 3 age intervals of 10-20 years old, 20-30 years old, and 30-40 years old, the preset age interval includes 10-20 years old, 20-30 years old, 30-40 years old.

The label of the first audio sample is related information of the age range to which the first audio sample belongs, that is, the label of the first audio sample is the actual age range to which it belongs, for example, the age label information is 21 years old, 22 years old, 23 years old. 3 first audio samples, all labeled 20-30 years old. In an optional embodiment, the label of each first audio sample may be obtained by manual labeling. In another optional embodiment, each first audio sample may be obtained by collecting audio of users in a specific age range, and a label of each first audio sample may be obtained according to the actual age of the user.

Next, for each acquired first audio sample, determine the first sample feature vector representing the acoustic feature of the first audio sample, the acquisition method of the first sample feature vector, and the feature of the audio to be identified in step S102 The acquisition method of the vector is similar, and will not be repeated here;

Finally, the information recognition model is trained based on the first sample feature vector and the label corresponding to the first sample feature vector. Specifically, the first sample feature vector is input into the information identification model, the information identification model is used to identify the first sample feature vector, and the probability value of the first sample feature vector in each preset age interval is output, and The preset age interval corresponding to the maximum probability value is used as the predicted age interval of the first sample eigenvector, since the predicted age interval of the first sample eigenvector identified by the information recognition model should be consistent with the label corresponding to the first sample eigenvector , therefore, if the predicted age interval is inconsistent with the corresponding label, adjust the parameters of the information recognition model, input the first sample feature vector into the information recognition model after parameter adjustment, and repeat the above process until the preset model training is satisfied The cut-off condition, after the training of the information recognition model.

The preset model training cut-off conditions may include the following two situations: first, the number of training times reaches a preset number of times (for example, 200 times); second, the trained information recognition model is tested by using the test sample set. The recognition accuracy of the model reaches a preset threshold (eg, 90%). In both cases, the information recognition model after the last parameter adjustment is used as the trained information recognition model.

Sub-step S1032, according to the probability value of the feature vector in each preset age interval, determine the age recognition result corresponding to the audio to be recognized.

In one embodiment, after obtaining the probability value of the feature vector in each preset age interval according to the method introduced in sub-step S1031, the probability value of each preset age interval can be interpolated to obtain the age corresponding to the audio to be recognized. Identify the results. In addition, the probability values of the feature vectors in each preset age interval can also be weighted and averaged to obtain the age recognition result corresponding to the audio to be recognized. In order to facilitate the calculation, the middle value of each preset age interval can be taken and the weighted average can be calculated. , the information recognition model outputs that the probability of 10-20 years old is 0, the probability of 20-30 years old is 0.6, and the probability of 30-40 years old is 0.4, then the age recognition result is 25+(35-25)*0.4=29 years old.

In other embodiments, when the service tag includes a gender identification result, please refer to FIG. 4 , the process of obtaining the gender identification result based on the feature vector and the information identification model may include sub-steps S1033 to S1034, which are described in detail below:

Sub-step S1033, the feature vector is input into the information identification model, and the probability value of the feature vector in each preset gender category is obtained.

In one embodiment, when the service tag includes a gender recognition result, an information recognition model can be used to perform gender recognition on the audio to be recognized. In this case, the information recognition model can include GMM, a neural network model, etc., or a combination thereof; the neural network model can Including BP neural network, CNN, RNN, LSTM, etc.

As an implementation manner, the information recognition model includes a third network and a fourth network, and the third network is used to extract the feature vector and output the probability value of the feature vector in each preset gender category. In this embodiment, male , female, and uncertain are the three gender categories. The uncertainty here refers to the recognition error caused by noise interference. At this time, the feature vector is input into the information recognition model to obtain the probability value of the feature vector in each preset gender category. process, which can include:

The first step is to input the feature vector into the information recognition model, and use the third network to perform feature extraction on the feature vector to obtain a second feature map.

In one embodiment, after the feature vector is input into the information recognition model, the first network of the information recognition model can perform feature extraction on the feature vector, and the third network can include a convolution layer, a convolution layer+pooling layer, a convolution layer + pooling layer + random loss layer, fully connected layer, etc. The specific settings of the third network can be flexibly adjusted by the user according to the actual situation, and will not be repeated here.

As an embodiment, the third network includes a convolution layer and a long-term and short-term memory layer. At this time, the feature vector is input into the information recognition model, and the third network is used to extract the feature vector to obtain the second feature map. Process, you can include:

First, the feature vector is input into the information recognition model, and the feature vector is convolved with the convolution kernel and the convolution kernel offset of the convolution layer to obtain the second output feature map. The convolution layer here can be multi-layered, and the specific number of layers and the size of the convolution kernel of the convolution layer are not limited.

Then, the long and short-term memory layer is used to capture the sequence information of the second output feature map, and the second feature map is obtained. Compared with the local features of the traditional model, the sequence information captured by the long and short-term memory layer can reflect the audio and gender to be recognized for a period of time. association.

In the second step, the fourth network is used to classify the second feature map output by the third network, and the probability value of the feature vector in each preset gender category is obtained.

In one embodiment, after the second feature map is extracted by the third network, the second feature map is output to the fourth network for classification, and the probability value of the feature vector in each preset gender category is output. The fourth network may include all Connection layer + multi-class logistic regression layer, 1*1 convolution layer + multi-class logistic regression layer, global pooling layer + multi-class logistic regression layer, etc. The specific settings of the second network can be flexibly adjusted by the user according to the actual situation. It is not repeated here.

As an embodiment, the fourth network includes a fully connected layer and a multi-class logistic regression layer. At this time, the fourth network is used to classify the second feature map output by the third network, and the feature vectors in each preset gender category are obtained. Probability value steps, including:

First, input the second feature map output by the third network into the fully connected layer, and use the fully connected layer to reduce the dimension of the second feature map to obtain a second vector. The second vector is a one-dimensional vector, and the fully connected layer here can be It is a multi-layer, and the specific number of the fully connected layer is not limited. The fully connected layer can be used to flatten the second feature map into a one-dimensional vector.

Then, the multi-class logistic regression layer is used to process the second vector to obtain the probability value of the feature vector in each preset gender category, for example, the output probability of male is 0.1, the probability of female is 0.1, and the probability of uncertainty is 0.8.

In one embodiment, before using the information identification model to identify the probability value of the feature vector in each preset gender category, the information identification model needs to be trained, and the specific training process may include the following steps:

First, acquire a plurality of second audio samples and gender labeling information of each second audio sample, where the plurality of second audio samples include male and female audio samples, and the gender labeling information includes male and female. In an optional embodiment, the gender labeling information of each second audio sample may be obtained by manual labeling. In another optional embodiment, each second audio sample may be obtained by collecting audio of a user of a specific gender.

Then, for each acquired second audio sample, determine the second sample feature vector representing the acoustic feature of the second audio sample, the acquisition method of the second sample feature vector, and the acquisition of the feature vector of the audio to be identified in step S102 The method is similar and will not be repeated here;

Finally, the information recognition model is trained based on the second sample feature vector and the gender labeling information of the second sample feature vector. Specifically, the second sample feature vector is input into the information identification model, the second sample feature vector is identified by the information identification model, and the probability value of the second sample feature vector in each preset gender category is output, and the maximum probability The gender category corresponding to the value is used as the predicted gender of the second sample feature vector. Since the predicted gender of the second sample feature vector should be consistent with the gender annotation information of the second sample feature vector, if the predicted gender is inconsistent with the gender annotation information, the The information recognition model adjusts parameters, and inputs the second sample feature vector into the parameter-adjusted information recognition model, and repeats the above process until the preset model training cutoff condition is met, and then the information recognition model is trained.

In addition, in practical applications, the information recognition model can also be optimized by obtaining the actual gender of the user, for example, by conducting a survey of the user or performing real-name authentication on the user to know the user's gender, so as to correct the recognition error of the information recognition model. .

Sub-step S1034, using the pre-trained hidden Markov model to constrain the probability value of the feature vector in each preset gender category, to obtain a gender recognition result corresponding to the audio to be recognized.

In one embodiment, after obtaining the probability value of each preset gender type of the feature vector according to the method introduced in sub-step S1033, the preset gender category corresponding to the maximum probability value is used as the predicted gender of the feature vector, and the pre-trained gender is used. Hidden Markov Model (HMM) adds constraints in the time domain to limit the jump of preset gender categories, that is, the constraints predict the jump between the three categories of male, female, and uncertain. , specifically, limiting the jumping of the category can be achieved by setting the HMM hidden state jump probability, that is, increasing the probability of state self-jumping and reducing the probability of jumping to other states, for example, the information recognition model outputs a probability of 0.1, If the probability of female is 0.1, and the probability of uncertainty is 0.8, then set the probability of self-jumping for the three categories of male, female, and uncertainty as 0.9, 0.8, and 0.6, and jump to other states for the three categories of male, female, and uncertainty. The probability of is 0.1, 0.2, 0.4.

In other embodiments, when the service tag is the identity verification result, please refer to FIG. 5 , the process of obtaining the identity verification result based on the feature vector and the information recognition model may include sub-steps S1035 to S1038, which are described in detail below:

Sub-step S1035, obtaining the standard audio corresponding to the audio to be recognized.

In one embodiment, the database 140 of the information identification system 100 pre-stores the standard audio of each user to be authenticated. The identity is verified, and the standard audio corresponding to the identity to be verified is obtained from the database 140 .

According to the actual identity of the user, the standard audio stored in the database 140 can be obtained in different ways. For example, for the driver and passenger in the online car-hailing scenario, the audio can be obtained when the driver or passenger is registered, and the registered audio can be used as the driver. Or the standard audio of the passenger. In addition, the standard audio may include the audio of the driver or passenger in different channels (eg, telephone channel, mobile wireless channel, etc.) Standard audio for driver and passenger.

Sub-step S1036, determining a feature vector representing the acoustic feature of the standard audio.

The acquisition method of the feature vector of the standard audio is similar to the acquisition method of the feature vector of the audio to be recognized in step S102, and details are not repeated here.

Sub-step S1037, input the feature vector corresponding to the audio to be recognized and the feature vector corresponding to the standard audio into the information recognition model to obtain the similarity score between the audio to be recognized and the standard audio.

In one embodiment, when the service tag includes an authentication result, an information identification model may be used to authenticate the audio to be identified. In this case, the information identification model may include GMM, and probabilistic linear discriminant analysis may include probabilistic linear discriminant analysis (Probabilistic Linear Discriminant Analysis). Analysis, PLDA) model, or neural network model, when using GMM, PLDA model, or neural network model to calculate the similarity score between the audio to be recognized and the standard audio, what is compared is the feature vector corresponding to the audio to be recognized and the corresponding standard audio The weighted average of the eigenvectors, rather than the individual eigenvectors, can make the comparison results more accurate.

The similarity score includes any one of Euclidean distance, Mahalanobis distance, included angle cosine, Hamming distance, correlation coefficient, correlation distance, and information entropy.

In sub-step S1038, an identity verification result corresponding to the audio to be recognized is determined according to the similarity score, and the identity verification result includes a verification pass result or a verification failure result.

In one embodiment, the identity verification result corresponding to the audio to be recognized may be determined according to the magnitude relationship between the similarity score and the preset similarity threshold. Specifically, the similarity score and the preset similarity are determined. The threshold (for example, 0.8) is compared, and when the similarity score is greater than or equal to the preset similarity threshold (for example, 0.8), it is determined that the identity verification result corresponding to the audio to be recognized is verified as passed; when the similarity score is When it is less than a preset similarity threshold (for example, 0.8), it is determined that the authentication result corresponding to the audio to be recognized is an authentication failure.

Combined with the online car-hailing scenario, when the driver's identity needs to be verified, the driver's identity is verified by obtaining the driver's to-be-recognized audio and standard audio, and comparing the feature vector of the to-be-recognized audio and the standard audio feature vector, that is, Verify whether the driver who actually uses the car-hailing is the same person as the registered driver; when the driver's identity needs to be verified, the characteristics of the audio to be recognized are obtained by obtaining the audio to be recognized of the passenger and the standard audio in the passenger voiceprint database. The vector is compared with the feature vector of standard audio to confirm the identity of the passenger, that is, to confirm the identity of the passenger who actually uses the car-hailing.

It should be pointed out that this application uses the information recognition model to obtain the service tag corresponding to the audio to be recognized, and the service tag can be one of the age recognition result, gender recognition result, and identity verification result, or any combination of these three items. , that is, based on the feature vector of the user's audio to be recognized, the information recognition model can be used to perform at least one of age recognition, gender recognition, and identity verification on the user. In practical applications, the structure of the information recognition model can be flexibly adjusted according to requirements. to get the final service tag. For example, when applied to the online car-hailing scenario, age recognition, gender recognition and identity verification can be performed on the driver at the same time, and the obtained age recognition results, gender recognition results, and identity verification results can be compared with the real-name registration information of the driver. Determine whether the driver who actually uses the car-hailing and the registered driver are actually the same person, so as to get more accurate results.

In step S104, service information matching the service tag is determined according to the service tag.

In one embodiment, after obtaining the service tag corresponding to the audio to be recognized according to the method introduced in step S104, that is, after using the information recognition model to identify the age and gender of the user and verifying the actual identity of the user, the Service tags match users to corresponding services.

In some embodiments, when the service tag includes an age identification result, the server 110 may pre-store a plurality of preset age intervals and the service content corresponding to each preset age interval. For example, when applied to an online car-hailing scenario, the server 110 The age ranges of 10-20 years old, 20-30 years old, 30-40 years old, 40-50 years old, and over 50 years old can be stored in advance, as well as the drivers corresponding to each age range. For example, 20-30 years old corresponds to good reputation. Drivers over the age of 50 correspond to drivers with stable driving, etc.

At this time, the process of determining the service information matching the service tag according to the service tag may include: first, according to the age recognition result, determining the target age range to which the age recognition result belongs from a plurality of preset age ranges, for example, If the age recognition result is 29 years old, the target age range to which the age recognition result belongs is 20-30 years old; then, obtain the service content corresponding to the target age range, and obtain the service information that matches the service tag, for example, determine the age recognition result If the target age range is 20-30 years old, one will be randomly selected from the corresponding drivers of 20-30 years old to dispatch the order.

Age recognition plays an important role in some scenarios. For example, when applied to online car-hailing scenarios, the age of the passenger cannot usually be known because the passenger terminal does not have real-name authentication. In this case, the information recognition method introduced in the embodiment of this application can be used to The age of the passengers and the allocation of orders according to their age. For example, allocating the orders of young passengers to drivers with good reputation can effectively avoid accidents in the process of taxiing; another example, when applied to online shopping scenarios, due to different age groups The users of , may have different consumption habits and consumption needs, so the age portrait can be obtained by identifying the user's age. At this time, the user can be recommended for products based on the age portrait.

In other embodiments, when the service tag includes a gender identification result, the server 110 may pre-store multiple preset gender categories and service content corresponding to each preset gender category. For example, when applied to an online car-hailing scenario, the server 110 may pre-store two gender categories, male and female, and drivers corresponding to each gender category, such as female drivers corresponding to female drivers, male drivers corresponding to male drivers, and the like.

At this time, the process of determining the service information matching the service tag according to the service tag may include: obtaining target service content corresponding to the gender identification result from a plurality of service contents according to the gender identification result, and obtaining a target service content that matches the service tag. For service information, for example, if the gender identification result is female, one of the female drivers will be randomly selected to dispatch the order.

Gender recognition plays an important role in some scenarios. For example, when applied to online car-hailing scenarios, since the passenger terminal does not have real-name authentication, it is usually impossible to know the gender of the passenger. In this case, the information recognition method introduced in the embodiment of this application can be used to identify The gender of the passenger, and assign the order according to the gender. For example, allocating the order of the female passenger to the female driver can effectively avoid the accident during the taxi process; another example, applied to the online shopping scene, can identify the gender of the user. , and make personalized recommendations for users based on gender.

In other embodiments, when the service tag includes the identity verification result, the server 110 may pre-store the identity verification result and the service content corresponding to the identity verification result. For normal order dispatch, the result of verification failure corresponds to stop dispatch; for passengers, the result of verification pass corresponds to the generation of an order, and the result of verification failure corresponds to the stop of order generation.

At this time, the process of determining the service information matching the service tag according to the service tag may include: when the identity verification result is a verification pass result, acquiring the service content corresponding to the verification pass result as the service information matching the service tag; When the authentication result is an authentication failure result, the service content corresponding to the authentication failure result is acquired as the service information matching the service tag.

Identity verification plays an important role in some scenarios. For example, it is applied to online car-hailing scenarios. With the rapid development of the Internet, online car-hailing plays an important role in people's travel. Convenience also has many potential safety hazards. For example, the user (driver or passenger) who actually uses the car-hailing may not be the same person as the registered user, especially the user who actually uses the car-hailing may have a driving accident record or even A person with a criminal record, therefore, can use the information identification method introduced in the embodiment of this application to perform identity verification on the driver or passenger, and allocate orders according to the result of the identity verification, for example, stop allocating orders when the identity verification fails. Access to the service platform of the public security organ to assist the public security organ in law enforcement.

Based on the same inventive concept, the embodiment of the present application also provides an information identification device 300 corresponding to the information identification method. For the implementation, refer to the implementation of the method, and the repetition will not be repeated.

Please refer to FIG. 6 . FIG. 6 shows a schematic diagram of an information identification device 300 provided by an embodiment of the present application. The information identification device 300 includes an audio acquisition module 301 , a first execution module 302 , a second execution module 303 and a third execution module 304 .

The audio acquisition module 301 is used to acquire the audio to be recognized.

The first execution module 302 is configured to determine a feature vector representing the acoustic feature of the audio to be identified.

In an optional implementation manner, the acoustic feature includes at least one of a Mel-frequency cepstral coefficient MFCC feature and a bottleneck feature BNF feature.

The second execution module 303 is configured to obtain a service tag corresponding to the audio to be recognized based on the feature vector and the pre-trained information recognition model.

In an optional embodiment, the service tag includes an age identification result; the second execution module 303 is specifically configured to: input the feature vector into the information identification model to obtain the probability value of the feature vector in each preset age interval; The probability value of the preset age interval is used to determine the age recognition result corresponding to the audio to be recognized.

In an optional embodiment, the information identification model includes a first network and a second network; the second execution module 303 executes the method of inputting the feature vector into the information identification model to obtain the probability value of the feature vector in each preset age interval, including: : Input the feature vector into the information recognition model, and use the first network to perform feature extraction on the feature vector to obtain the first feature map; input the first feature map extracted from the first network into the second network and classify, and obtain the feature vector in The probability value of each preset age interval.

In an optional embodiment, the first network includes a convolution layer and a statistical pooling layer; the second execution module 303 executes the feature vector input information recognition model, and uses the first network to perform feature extraction on the feature vector to obtain the first feature. The method of graphing includes: inputting the feature vector into the information recognition model, using the convolution kernel and convolution kernel offset of the convolution layer to convolve the feature vector, and obtaining the first output feature map; using the statistical pooling layer to An output feature map is subjected to statistical pooling to obtain a first feature map.

In an optional implementation manner, the second network includes a fully connected layer and a multi-class logistic regression layer; the second execution module 303 executes the input of the first feature map extracted from the first network into the second network and performs classification to obtain features. The method of the probability value of the vector in each preset age interval includes: inputting the first feature map extracted from the first network into the fully connected layer, and using the fully connected layer to reduce the dimension of the first feature map to obtain the first vector; The first vector is input into the multi-class logistic regression layer, and the probability value of the feature vector in each preset age interval is obtained.

In an optional embodiment, the second execution module 303 executes the method of determining the age recognition result corresponding to the audio to be recognized according to the probability value of the feature vector in each preset age interval, including: placing the feature vector at each preset age. The probability value of the interval is weighted and averaged to obtain the age recognition result corresponding to the audio to be recognized.

In an optional implementation manner, the service label includes the gender identification result; the second execution module 303 executes the information identification model based on the feature vector and pre-trained to obtain the service label corresponding to the audio to be identified, including: inputting the feature vector into the information. Identify the model to obtain the probability value of the feature vector in each preset gender category; use the pre-trained hidden Markov model to constrain the probability value of the feature vector in each preset gender category to obtain the gender recognition result corresponding to the audio to be recognized.

In an optional embodiment, the information identification model includes a third network and a fourth network; the second execution module 303 executes the method of inputting the feature vector into the information identification model to obtain the probability value of the feature vector in each preset gender category, including: : Input the feature vector into the information recognition model, and use the third network to extract the feature vector to obtain the second feature map; use the fourth network to classify the second feature map output by the third network to obtain the feature vector in each preset Probability values for gender categories.

In an optional embodiment, the third network includes a convolution layer and a long-term and short-term memory layer; the second execution module 303 executes the feature vector input information recognition model, and uses the third network to perform feature extraction on the feature vector to obtain the second feature. The method includes: inputting the feature vector into the information recognition model, using the convolution kernel and convolution kernel offset of the convolution layer to convolve the feature vector, and obtaining the second output feature map; using the long and short-term memory layer to capture the first Second, the sequence information of the output feature map is obtained to obtain a second feature map.

In an optional implementation manner, the fourth network includes a fully connected layer and a multi-class logistic regression layer; the second execution module 303 uses the fourth network to classify the second feature map output by the third network, and obtains the feature vector in each The method of presetting the probability value of the gender category includes: inputting the second feature map output by the third network into the fully connected layer, and using the fully connected layer to reduce the dimension of the second feature map to obtain a second vector; using multi-class logistic regression The layer processes the second vector to obtain the probability value of the feature vector in each preset gender category.

In an optional implementation manner, the service tag includes an identity verification result; the second execution module 303 executes the information recognition model based on the feature vector and pre-trained to obtain the service tag corresponding to the audio to be recognized, including: obtaining the corresponding audio to be recognized. the standard audio; determine the feature vector representing the acoustic features of the standard audio; input the feature vector corresponding to the audio to be recognized and the feature vector corresponding to the standard audio into the information recognition model to obtain the similarity score between the audio to be recognized and the standard audio; The similarity score determines the identity verification result corresponding to the audio to be recognized, and the identity verification result includes a verification pass result or a verification failure result.

The third execution module 304 is configured to determine service information matching the service tag according to the service tag.

In an optional implementation manner, the server 110 pre-stores a plurality of preset age intervals and the service content corresponding to each preset age interval; the third execution module 304 executes to determine the service information matching the service tag according to the service tag. The method includes: determining the target age range to which the age recognition result belongs from a plurality of preset age ranges according to the age recognition result; obtaining service content corresponding to the target age range, and obtaining service information matching the service tag.

In an optional embodiment, the server 110 pre-stores a plurality of preset gender categories and the service content corresponding to each preset gender category; the third execution module 304 executes to determine the service information matching the service tag according to the service tag. The method includes: acquiring the target service content corresponding to the gender identification result from a plurality of service contents according to the gender identification result, and obtaining the service information matching the service tag.

In an optional embodiment, the server 110 pre-stores the identity verification result and the service content corresponding to the identity verification result; the third execution module 304 executes the method of determining the service information matching the service tag according to the service tag, including: when When the authentication result is the verification pass result, the service content corresponding to the verification pass result is obtained as the service information matching the service label; when the authentication result is the verification failure result, the service content corresponding to the verification failure result is obtained as the service content corresponding to the service label. matching service information.

The embodiment of the present application further provides an electronic device 60 , as shown in FIG. 7 , which is a schematic structural diagram of the electronic device 60 provided by the embodiment of the present application, including: a processor 61 , a memory 62 , and a bus 63 . The memory 62 stores machine-readable instructions executable by the processor 61 (for example, the audio acquisition module 301, the first execution module 302, the second execution module 303, and the third execution module 304 in the device in FIG. Execution instructions, etc.), when the electronic device 60 is running, the processor 61 communicates with the memory 62 through the bus 63, and the machine-readable instructions are executed by the processor 61 when the above-mentioned information identification method is executed. step.

Embodiments of the present application further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the above information identification method when the computer program is run by a processor. Specifically, the storage medium can be a general storage medium, such as a removable disk, a hard disk, etc., when the computer program on the storage medium is run, the above-mentioned information identification method can be executed, so as to provide more suitable information without the user's perception. services and improve the user experience.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, for the specific working process of the system and device described above, reference may be made to the corresponding process in the method embodiment, which is not repeated in this application. In the several embodiments provided in this application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. The device embodiments described above are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces, indirect coupling or communication connection of devices or modules, which may be in electrical, mechanical or other forms.

The modules described as separate components may or may not be physically separated, and components shown as modules may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a processor-executable non-volatile computer-readable storage medium. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk and other mediums that can store program codes.

The above are only the specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the technical scope disclosed in the present application can easily think of changes or replacements, which should be covered within the scope of the present application. within the scope of protection of this application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims (17)

1. an information identification method, is characterized in that, described method comprises: Get the audio to be recognized; determining a feature vector characterizing the acoustic feature of the audio to be identified; Based on the feature vector and the pre-trained information recognition model, obtain the service tag corresponding to the audio to be recognized; Service information matching the service tag is determined according to the service tag. 2. The method according to claim 1, wherein the service tag includes an age identification result; The step of obtaining the service tag corresponding to the audio to be recognized based on the feature vector and the pre-trained information recognition model includes: Inputting the feature vector into the information identification model to obtain the probability value of the feature vector in each preset age interval; According to the probability value of the feature vector in each preset age interval, the age recognition result corresponding to the audio to be recognized is determined. 3. The method according to claim 2, wherein the information identification model comprises a first network and a second network; The step of inputting the feature vector into the information identification model to obtain the probability value of the feature vector in each preset age interval includes: Inputting the feature vector into the information recognition model, and using the first network to perform feature extraction on the feature vector to obtain a first feature map; The first feature map extracted from the first network is input into the second network and classified, and the probability value of the feature vector in each preset age interval is obtained. 4. The method according to claim 3, wherein the first network comprises a convolution layer and a statistical pooling layer; The step of inputting the feature vector into the information identification model, and using the first network to perform feature extraction on the feature vector to obtain a first feature map, includes: Inputting the feature vector into the information recognition model, and using the convolution kernel and the convolution kernel offset of the convolution layer to perform convolution processing on the feature vector to obtain a first output feature map; Using the statistical pooling layer to perform statistical pooling processing on the first output feature map to obtain a first feature map. 5. The method according to claim 3, wherein the second network comprises a fully connected layer and a multi-class logistic regression layer; The step of inputting the first feature map extracted from the first network into the second network and classifying to obtain the probability value of the feature vector in each preset age interval includes: Inputting the first feature map extracted from the first network to the fully connected layer, and using the fully connected layer to reduce the dimension of the first feature map to obtain a first vector; The first vector is input into the multi-class logistic regression layer to obtain the probability value of the feature vector in each preset age interval. 6. The method according to claim 2, wherein the step of determining the corresponding age recognition result of the audio to be recognized according to the probability value of the feature vector in each preset age interval comprises: A weighted average of the probability values of the feature vector in each preset age interval is performed to obtain an age recognition result corresponding to the audio to be recognized. 7. The method according to claim 2, wherein the server pre-stores a plurality of preset age intervals and service content corresponding to each preset age interval; The step of determining the service information matching the service tag according to the service tag includes: According to the age recognition result, determine the target age range to which the age recognition result belongs from the plurality of preset age ranges; Acquire the service content corresponding to the target age range, and obtain service information matching the service tag. 8. The method of claim 1, wherein the service tag includes a gender identification result; The step of obtaining the service tag corresponding to the audio to be recognized based on the feature vector and the pre-trained information recognition model includes: Inputting the feature vector into the information recognition model to obtain the probability value of the feature vector in each preset gender category; Using a pre-trained hidden Markov model to constrain the probability value of the feature vector in each preset gender category, a gender recognition result corresponding to the to-be-recognized audio is obtained. 9. The method according to claim 8, wherein the information identification model comprises a third network and a fourth network; The step of inputting the feature vector into the information recognition model to obtain the probability value of the feature vector in each preset gender category includes: Inputting the feature vector into the information recognition model, and using the third network to perform feature extraction on the feature vector to obtain a second feature map; The fourth network is used to classify the second feature map output by the third network, and the probability value of the feature vector in each preset gender category is obtained. 10. The method according to claim 9, wherein the third network comprises a convolutional layer and a long short-term memory layer; The step of inputting the feature vector into the information recognition model, and using the third network to perform feature extraction on the feature vector to obtain a second feature map, includes: Inputting the feature vector into the information recognition model, and using the convolution kernel of the convolution layer and the convolution kernel offset to perform convolution processing on the feature vector to obtain a second output feature map; The sequence information of the second output feature map is captured by the long and short-term memory layer to obtain a second feature map. 11. The method according to claim 9, wherein the fourth network comprises a fully connected layer and a multi-class logistic regression layer; The step of using the fourth network to classify the second feature map output by the third network to obtain the probability value of the feature vector in each preset gender category includes: Inputting the second feature map output by the third network into the fully connected layer, and using the fully connected layer to reduce the dimension of the second feature map to obtain a second vector; The second vector is processed by using the multi-class logistic regression layer to obtain the probability value of the feature vector in each preset gender category. 12. The method according to claim 8, wherein the server pre-stores a plurality of preset gender categories and service content corresponding to each preset gender category; The step of determining the service information matching the service tag according to the service tag includes: According to the gender identification result, the target service content corresponding to the gender identification result is acquired from a plurality of service contents, and the service information matching the service tag is obtained. 13. The method of claim 1, wherein the service tag comprises an authentication result; The step of obtaining the service tag corresponding to the audio to be recognized based on the feature vector and the pre-trained information recognition model includes: Obtain the standard audio corresponding to the to-be-recognized audio; determining a feature vector characterizing the acoustic features of the standard audio; Inputting the feature vector corresponding to the audio to be recognized and the feature vector corresponding to the standard audio into the information recognition model to obtain the similarity score of the audio to be recognized and the standard audio; An identity verification result corresponding to the to-be-recognized audio is determined according to the similarity score, and the identity verification result includes a verification pass result or a verification failure result. 14. The method according to claim 13, wherein the server pre-stores an identity verification result and service content corresponding to the identity verification result; The step of determining the service information matching the service tag according to the service tag includes: When the identity verification result is a verification pass result, obtain the service content corresponding to the verification pass result as the service information matching the service tag; When the identity verification result is a verification failure result, the service content corresponding to the verification failure result is acquired as service information matching the service tag. 15. An information identification device, characterized in that the device comprises: The audio acquisition module is used to acquire the audio to be recognized; a first execution module for determining a feature vector representing the acoustic feature of the audio to be identified; a second execution module, configured to obtain a service tag corresponding to the audio to be recognized based on the feature vector and the pre-trained information recognition model; A third execution module, configured to determine service information matching the service tag according to the service tag. 16. An electronic device, comprising: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device runs, the processor is connected to a bus. The storage media communicate through a bus, and the processor executes the machine-readable instructions to execute the steps of the information identification method according to any one of claims 1 to 14 at runtime. 17. A computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, the information identification method according to any one of claims 1 to 14 is executed. step.

Images