Voice tag acquisition method and device, equipment, medium and product

Abstract

The invention discloses a voice tag acquisition method and device, equipment, a medium and a product. The voice tag acquisition method comprises the following steps: in response to a voice device pull-up operation of a user, generating an authorization request and sending the authorization request to the user; in response to an authorization request confirmation operation of the user, collecting the to-be-detected audio frame data through a voice real-time collection device; performing signal enhancement on the to-be-detected audio frame data in a pre-trained context front-end processing model to obtain signal-enhanced to-be-detected audio frame data; and inputting the to-be-detected audio frame data into a pre-constructed voice tag acquisition model to obtain a tag identification result matched with the to-be-detected audio frame data. According to the technical scheme of the invention, the voice tag can be obtained, and the accuracy of the tag recognition result of the voice is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, device, equipment, medium and product for acquiring a voice tag. Background Art

[0002] In today’s digital financial services sector, the quality and efficiency of bank customer service is crucial to improving customer satisfaction and loyalty. With the development of technology, the customer service section of bank apps and telephone intelligent customer service are playing an increasingly important role in providing services to customers.

[0003] However, the existing deep learning-based speech tag acquisition technology still has many shortcomings in the application of bank customer service. The main technical solutions currently include the use of models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) and their variants (such as long short-term memory networks (LSTMs)). Although CNN can extract effective features from raw speech signals, capture local features and reduce feature dimensions through local perception domains and pooling layers, and improve model robustness, it is necessary to save a large number of intermediate results when processing high-dimensional data such as long speech signals commonly seen in bank customer service, resulting in high memory consumption and weak ability to capture global temporal information. RNNs and their variants can capture temporal dependencies in speech signals through recurrent connection structures, and have advantages in understanding speech context. However, when processing complex long-sequence speech in bank customer service, it is necessary to gradually calculate the output of each time step, which has high computational complexity and is prone to problems such as overfitting, gradient vanishing or gradient explosion.

[0004] At the same time, in the actual application scenarios of bank customer service, complex environments such as noise and background noise interference frequently occur, and problems such as non-standard user pronunciation and dialect differences are also common. Due to the lack of existing technology, these factors often lead to speech recognition errors, which in turn affects the subsequent dialogue understanding and question answering.

[0005] In summary, the existing voice tag acquisition method has the problem of inaccurate voice tag recognition results. Summary of the invention

[0006] The present invention provides a method, device, equipment, medium and product for acquiring voice tags, which can solve the problem of inaccurate voice tag recognition results caused by existing voice tag acquisition methods.

[0007] In a first aspect, an embodiment of the present invention provides a method for acquiring a voice tag, the method comprising:

[0008] In response to the user's voice device pull-up operation, generate an authorization request and send it to the user;

[0009] In response to the user's authorization request confirmation operation, the audio frame data to be detected is collected by a real-time voice collection device;

[0010] Performing signal enhancement on the audio frame data to be detected in a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement;

[0011] The audio frame data to be detected is input into a pre-built voice tag acquisition model to obtain a tag recognition result matching the audio frame data to be detected.

[0012] In a fourth aspect, an embodiment of the present invention provides a voice tag acquisition device, the device comprising:

[0013] A data acquisition module, used to collect audio frame data to be detected through a real-time voice acquisition device;

[0014] A signal enhancement module, used to perform signal enhancement on the audio frame data to be detected through a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement;

[0015] The model processing module is used to input the audio frame data to be detected into the voice tag acquisition model trained by the construction method of the voice tag acquisition model to obtain a tag recognition result matching the audio frame data to be detected.

[0016] In a fifth aspect, an embodiment of the present invention provides an electronic device, the electronic device comprising:

[0017] at least one processor; and

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a voice tag acquisition method described in any embodiment of the present invention.

[0020] In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement a voice tag acquisition method described in any embodiment of the present invention when executed.

[0021] The technical solution of the embodiment of the present invention first obtains the original sample set, processes each target audio frame based on a preset label strength extractor, obtains the label strength score matching each target audio frame, and performs mask processing on each target audio frame and the label strength score matching each target audio frame based on a preset mask strategy according to each label strength score to obtain each sample audio frame, then forms a training sample set according to each sample audio frame and each label marking result, trains the mockingbird model using the training sample set to obtain a speech label acquisition model, then collects the audio frame data to be detected through a real-time speech acquisition device, and performs signal enhancement on the audio frame data to be detected through a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement, and finally inputs the audio frame data to be detected into the speech label acquisition model to obtain a label recognition result matching the audio frame data to be detected, thereby solving the problem of inaccurate speech label recognition result caused by the existing speech label acquisition method, realizing the construction of the speech label acquisition model, and using the model to acquire speech labels to improve the accuracy of the speech label recognition result.

[0022] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 is a flowchart of a method for acquiring voice tags according to Embodiment 1 of the present invention;

[0025] Figure 2 is a flowchart of a method for constructing a voice tag acquisition model provided according to Embodiment 2 of the present invention;

[0026] Figure 3 Schematic diagram of a voice tag acquisition device according to Embodiment 3 of the present invention

[0027] Figure 4 The present invention is a schematic diagram of the structure of an electronic device for implementing a method for acquiring a voice tag according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, any variation of the terms "including" and "having" is intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] Embodiment 1

[0031] Figure 1 A flowchart of a voice tag acquisition method provided in Embodiment 1 of the present invention. This embodiment is applicable to the situation of acquiring voice tags. The method can be executed by a voice tag acquisition device. The voice tag acquisition device can be implemented in the form of hardware and / or software. The voice tag acquisition device can be configured in the customer service section of a banking application with a voice tag acquisition function.

[0032] like Figure 1 As shown, the method includes:

[0033] S110. In response to the user's voice device pull-up operation, generate an authorization request and send it to the user.

[0034] Among them, the voice device pull-up operation is that the user takes specific actions or instructions to start or activate the voice device, so that it changes from a standby or non-working state to a state ready to receive and process voice commands. For example, the user presses a specific button on the voice device, says a specific wake-up word, clicks the start icon in the relevant application, and so on; further, the authorization request refers to a request issued to the user to obtain specific permissions or allow specific operations to be performed. For example, requesting permission to access the user's microphone, requesting permission to collect and process voice data, etc.

[0035] S120 . In response to the user's authorization request confirmation operation, the audio frame data to be detected is collected by a real-time voice collection device.

[0036] The authorization request confirmation operation refers to an action in which the user gives explicit consent or approval to the received authorization request, such as clicking an "Agree" button, saying "Confirm", or other affirmative instructions.

[0037] Specifically, in the customer service section of a bank application, real-time voice acquisition devices are often used to collect audio frame data to be detected; wherein, the real-time voice acquisition device is a type of tool that plays a key role in the bank customer service scenario, used to capture the customer's voice signal and convert it into digital form. For example, a common microphone is a real-time voice acquisition device. In the actual application of bank customer service, various types of microphones may be used, such as dynamic microphones, which have strong durability and can work stably in a more complex environment, and are suitable for customers in noisy environments; and condenser microphones, because of their high sensitivity and wide frequency response, can more clearly collect subtle changes in the customer's voice, suitable for scenarios with high requirements for sound quality, such as customers need to elaborate on complex financial issues. These real-time voice acquisition devices can accurately and timely convert the customer's voice into electrical signals in different scenarios and conditions of communication between bank customer service and customers, and further process it into digital audio frame data to be detected through advanced analog-to-digital conversion and other technologies. These data provide the original and key basic materials for the subsequent bank customer service system to analyze, process and accurately identify customer needs, which helps to improve the service quality and efficiency of bank customer service and provide customers with a better and more accurate service experience.

[0038] S130, performing signal enhancement on the audio frame data to be detected through a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement.

[0039] Specifically, in the customer service section related technology of the bank application, the audio frame data to be detected is subjected to signal enhancement in a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement; wherein, the context front-end processing model is a model built based on deep learning or machine learning algorithms, which can process and optimize the input audio frame data. For example, the common model architecture based on convolutional neural network (CNN) or recurrent neural network (RNN). Convolutional neural network is good at extracting local features of audio data, while recurrent neural network is more suitable for processing data with time series characteristics. In this process, the model will analyze the noise, attenuation and other problems in the audio frame data to be detected, and perform targeted enhancement processing. For example, for signal attenuation caused by network transmission, the model can enhance the signal strength by adjusting parameters such as audio amplitude and frequency; for interference from environmental noise, the model can use filtering algorithms and other technologies to reduce the impact of noise, so that the audio is clearer and recognizable, providing a better data foundation for subsequent speech analysis and processing.

[0040] S140: Input the audio frame data to be detected into a voice tag acquisition model to obtain a tag recognition result matching the audio frame data to be detected.

[0041] Optionally, after obtaining the tag recognition result that matches the audio frame data to be detected, the method further includes: based on the tag recognition result and a preset policy rule, obtaining a voice policy that matches the tag recognition result and pushing it to the user.

[0042] The strategy rules are pre-set criteria and specifications for judging and handling different situations. For example, when the tag recognition result is anger, the corresponding strategy rule is to take soothing measures.

[0043] For example, when the real-time analysis finds that the user's label recognition result is a strong negative emotion such as anger, excitement, speechlessness, etc., the system will quickly obtain the corresponding soothing voice strategy according to the preset strategy rules, such as gentle apology words, caring explanations, etc., and push them to the user in a timely manner. This can effectively help users relieve negative emotions and improve their experience and satisfaction.

[0044] The technical solution of the embodiment of the present invention is to collect audio frame data to be detected through a real-time speech acquisition device, and perform signal enhancement on the audio frame data to be detected through a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement, and finally input the audio frame data to be detected into a speech label acquisition model to obtain a label recognition result matching the audio frame data to be detected, thereby realizing speech label acquisition and improving the accuracy of speech label recognition results.

[0045] Embodiment 2

[0046] Figure 2 This is a flow chart of a method for constructing a voice tag acquisition model provided in the second embodiment of the present invention. This embodiment is supplemented on the basis of the above embodiment, and specifically supplements the method for constructing a voice tag acquisition model in this embodiment.

[0047] like Figure 2 As shown, the method includes:

[0048] S210: Obtain an original sample set.

[0049] The original sample includes at least one target audio frame and a label marking result matching the target audio frame.

[0050] Specifically, in this embodiment, it is necessary to obtain an original sample set. In this original sample set, at least one target audio frame is included; wherein the target audio frame is a small audio segment with specific features in the audio data or selected for analysis, and each target audio frame is matched with a corresponding label marking result; further, the label marking result is a clear marking of the label expressed by the audio frame, such as joy, sadness, anger, etc. These label marking results may be obtained through manual marking, automatic generation by machine learning algorithms or other methods, and are intended to provide a clear reference and basis for subsequent analysis and processing.

[0051] S220. Process each target audio frame based on a preset label strength extractor to obtain a label strength score matching each target audio frame, and perform mask processing on each target audio frame and the label strength score matching each target audio frame based on a preset masking strategy according to each label strength score to obtain each sample audio frame.

[0052] Among them, the label strength extractor is a tool or algorithm specially designed to analyze and evaluate the strength of the label contained in the audio frame. For example, it can judge the strength of the label by analyzing the frequency, amplitude, duration and other features of the audio, and obtain the label strength score matching each target audio frame after processing; further, in this embodiment, the label strength extractor is composed of an acoustic encoder and a strength predictor. The acoustic encoder is composed of 12 convolutional layers, which can extract high-level features H from a given input Mel-spectrogram sequence X, and then input the high-level features H into the label strength predictor for predicting the label strength score. The strength predictor includes a BiLSTM layer, two fully connected layers (FC) and an average pooling layer, which can be used to obtain feature-related label strength scores based on features. .

[0053] Among them, according to each label strength score, each target audio frame and the label strength score matching each target audio frame are masked based on a preset masking strategy to obtain each sample audio frame, including: sorting the target audio frames by label strength according to the label strength score of each target audio frame to obtain an audio frame sorting result; summing the target audio frame and the label strength score matching the target audio frame to obtain data to be masked that matches the target audio frame; based on the audio frame sorting result, masking each data to be masked is performed according to a preset masking strategy to obtain sample audio frames that match each target audio frame respectively.

[0054] Further, the masking process is an operation of partially hiding or modifying the original data through specific rules and algorithms; specifically, in the related technology of audio processing, the label strength score generated in the above steps of this embodiment is used to assign a higher probability to the audio frames with higher scores, that is, after calculating the label strength score for each audio frame, they will be arranged in order, and then a probability value k will be selected to mask the position corresponding to the first k% frame; further, in order to effectively improve the mismatch problem between training and reasoning in this embodiment, a sub-random process is deliberately introduced. This sub-random process is a special random processing method that aims to increase the diversity and uncertainty of processing. It is specifically divided into three steps: 80% of the time, the selected target audio frame is masked to zero; 10% of the time, the selected target audio frame is replaced with a random audio frame; and the remaining 10% of the time, the target audio frame is kept unchanged. In addition, in order to avoid the model relying solely on the local smoothness of the audio frame, an additional continuous mask is also used, which is a means of processing continuous data segments. In this way, the model can be forced to reason about the global structure instead of relying solely on local information, thereby improving the generalization ability and accuracy of the model.

[0055] S230: Form a training sample set according to each sample audio frame and each label marking result, wherein the label marking result is used as annotated data in the training sample.

[0056] S240: Use the training sample set to train the Mockingjay model to obtain a speech label acquisition model.

[0057] The mockingbird model includes an input layer, a feature extraction layer, an adaptive attention transformation network layer, a fully connected layer and a loss function layer which are connected in sequence.

[0058] The Mockingbird model is a complex architecture model specially designed to process voice tag related tasks, and in this embodiment can be used to obtain a voice tag acquisition model after training.

[0059] Further, the Mockingjay model is trained using the training sample set to obtain a voice tag acquisition model, including: obtaining a target training sample from the training sample set, and inputting the target training sample into the Mockingjay model through an input layer; performing a feature extraction operation on a sample audio frame in the target training sample through a feature extraction layer in the Mockingjay model to obtain a target audio frame feature; processing the target audio frame feature through an adaptive attention transformation network layer in the Mockingjay model to obtain a target information point matching the target audio frame feature; processing the target information point through a fully connected layer in the Mockingjay model to obtain a label prediction result; adjusting parameters of the Mockingjay model according to the label prediction result and the label marking result in the target training sample through a loss function layer in the Mockingjay model; returning to execute the operation of obtaining the target training sample in the training sample set until a training termination condition is met, and determining the trained Mockingjay model as a voice tag acquisition model.

[0060] Specifically, the Mockingjay model is trained using a training sample set to obtain a voice tag acquisition model, including: obtaining a target training sample from a rich and diverse training sample set, and inputting the target training sample into the Mockingjay model through an input layer as an entry; performing an in-depth feature extraction operation on a sample audio frame in the target training sample through a feature extraction layer in the Mockingjay model, so as to accurately obtain the target audio frame feature; performing intelligent processing on the extracted target audio frame feature through an adaptive attention transformation network layer in the Mockingjay model to obtain a target information point that is highly matched with the target audio frame feature; performing comprehensive processing on these target information points through a fully connected layer in the Mockingjay model to finally obtain a tag prediction result; performing fine parameter adjustment on the Mockingjay model according to the obtained tag prediction result and the accurate tag marking result in the target training sample through a loss function layer in the Mockingjay model, and it should be noted that, in this embodiment, the loss function layer can specifically use an L1 loss function; then returning to perform the operation of obtaining the target training sample in the training sample set, and repeating the cycle until a preset end training condition is met, and the mature Mockingjay model obtained through training is determined as a model that can accurately perform voice tag acquisition.

[0061] The technical solution of the embodiment of the present invention first obtains the original sample set, processes each target audio frame based on a preset label strength extractor, obtains a label strength score matching each target audio frame, and according to each label strength score, performs mask processing on each target audio frame and the label strength score matching each target audio frame based on a preset masking strategy to obtain each sample audio frame, and finally forms a training sample set according to each sample audio frame and each label marking result, uses the training sample set to train a mockingjay model to obtain a speech label acquisition model, realizes the construction of the speech label acquisition model, and thus improves the accuracy of the speech label recognition result.

[0062] Embodiment 3

[0063] Figure 3 This is a schematic diagram of the structure of a voice tag acquisition device provided by Embodiment 3 of the present invention. Figure 3 As shown, the device comprises:

[0064] The authorization request module 310 is used to generate an authorization request in response to the user's voice device pull-up operation and send it to the user;

[0065] The data collection module 320 is used to collect the audio frame data to be detected through the real-time voice collection device in response to the user's authorization request confirmation operation;

[0066] The signal enhancement module 330 is used to perform signal enhancement on the audio frame data to be detected through a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement;

[0067] The model processing module 340 is used to input the audio frame data to be detected into a pre-built voice tag acquisition model to obtain a tag recognition result matching the audio frame data to be detected.

[0068] The technical solution of the embodiment of the present invention is to collect audio frame data to be detected through a real-time speech acquisition device, and perform signal enhancement on the audio frame data to be detected through a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement, and finally input the audio frame data to be detected into a speech label acquisition model to obtain a label recognition result matching the audio frame data to be detected, thereby realizing speech label acquisition and improving the accuracy of speech label recognition results.

[0069] Based on the above embodiment, the model processing module 340 is also used to: after obtaining the label recognition result that matches the audio frame data to be detected, based on the label recognition result and the preset policy rules, obtain the voice strategy that matches the label recognition result and push it to the user.

[0070] On the basis of the above embodiment, the voice tag acquisition device further includes: a voice tag acquisition model construction module, which is used to: obtain an original sample set, wherein the original sample includes at least one target audio frame and a tag marking result matching the target audio frame; process each target audio frame based on a preset tag strength extractor to obtain a tag strength score matching each target audio frame, and mask each target audio frame and the tag strength score matching each target audio frame based on a preset masking strategy according to each tag strength score to obtain each sample audio frame; form a training sample set according to each sample audio frame and each tag marking result, and the tag marking result is used as annotation data in the training sample; use the training sample set to train the mockingjay model to obtain a voice tag acquisition model.

[0071] On the basis of the above embodiment, the voice tag acquisition device is also used to: sort the label strength of each target audio frame according to the label strength score of each target audio frame to obtain an audio frame sorting result; sum the target audio frame and the label strength score matching the target audio frame to obtain the data to be masked that matches the target audio frame; based on the audio frame sorting result, mask each data to be masked according to a preset masking strategy to obtain sample audio frames that match each target audio frame.

[0072] On the basis of the above embodiment, the speech tag acquisition device is further used to: acquire a target training sample from a training sample set, and input the target training sample into a mockingjay model through an input layer; perform a feature extraction operation on a sample audio frame in the target training sample through a feature extraction layer in the mockingjay model to obtain a target audio frame feature; process the target audio frame feature through an adaptive attention transformation network layer in the mockingjay model to obtain a target information point matching the target audio frame feature; process the target information point through a fully connected layer in the mockingjay model to obtain a label prediction result; adjust parameters of the mockingjay model according to the label prediction result and the label marking result in the target training sample through a loss function layer in the mockingjay model; return to execute the operation of acquiring the target training sample in the training sample set until the training end condition is met, and determine the trained mockingjay model as the model for speech tag acquisition.

[0073] Embodiment 4

[0074] Figure 4A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0075] like Figure 4 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, ROM 12 and RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0076] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0077] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a method for obtaining a voice tag.

[0078] Accordingly, a method for obtaining a voice tag includes:

[0079] In response to the user's voice device pull-up operation, generate an authorization request and send it to the user;

[0080] In response to the user's authorization request confirmation operation, the audio frame data to be detected is collected by a real-time voice collection device;

[0081] Performing signal enhancement on the audio frame data to be detected in a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement;

[0082] The audio frame data to be detected is input into a pre-trained voice tag acquisition model to obtain a tag recognition result matching the audio frame data to be detected.

[0083] In some embodiments, a voice tag acquisition method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the voice tag acquisition method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform a voice tag acquisition method in any other appropriate manner (e.g., by means of firmware).

[0084] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0085] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0086] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, 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 foregoing.

[0087] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0088] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0089] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0090] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

Claims

1. A method for acquiring a voice tag, characterized in that: include: In response to the user's voice device pull-up operation, generate an authorization request and send it to the user; In response to the user's authorization request confirmation operation, the audio frame data to be detected is collected by a real-time voice collection device; Performing signal enhancement on the audio frame data to be detected in a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement; The audio frame data to be detected is input into a pre-built voice tag acquisition model to obtain a tag recognition result matching the audio frame data to be detected.

2. The method according to claim 1, characterized in that After obtaining the tag recognition result matching the audio frame data to be detected, the method further includes: Based on the tag recognition result and preset policy rules, a voice policy matching the tag recognition result is obtained and pushed to the user.

3. The method according to claim 1, characterized in that The process of constructing the voice tag acquisition model includes: Acquire an original sample set, wherein the original sample includes at least one target audio frame and a label marking result matching the target audio frame; Processing each target audio frame based on a preset label strength extractor to obtain a label strength score matching each target audio frame, and performing masking processing on each target audio frame and the label strength score matching each target audio frame based on a preset masking strategy according to each label strength score to obtain each sample audio frame; Forming a training sample set according to each sample audio frame and each label marking result, wherein the label marking result is used as the annotation data in the training sample; The mockingjay model is trained using the training sample set to obtain a speech label acquisition model.

4. The method according to claim 3, characterized in that: According to each label strength score, each target audio frame and the label strength score matching each target audio frame are masked based on a preset mask strategy to obtain each sample audio frame, including: Sort the target audio frames by their label strength according to their label strength scores, and obtain an audio frame sorting result; The target audio frame and the label strength score matching the target audio frame are summed to obtain the to-be-masked data matching the target audio frame; Based on the audio frame sorting result, mask processing is performed on each to-be-masked data according to a preset masking strategy to obtain sample audio frames that match each target audio frame.

5. The method according to claim 3, characterized in that: The mockingjay model includes an input layer, a feature extraction layer, an adaptive attention transformation network layer, a fully connected layer, and a loss function layer which are connected in sequence.

6. The method according to any one of claims 3 to 5, characterized in that: Use the training sample set to train the mockingbird model to obtain the speech label acquisition model, including: Obtain a target training sample from the training sample set, and input the target training sample into the Mockingjay model through an input layer; Performing feature extraction operation on the sample audio frame in the target training sample through the feature extraction layer in the mockingbird model to obtain the target audio frame feature; Processing the target audio frame features through an adaptive attention transformation network layer in a mockingjay model to obtain target information points matching the target audio frame features; The target information point is processed by a fully connected layer in the Mockingjay model to obtain a label prediction result; Adjusting parameters of the Mockingjay model according to the label prediction results and the label marking results in the target training sample through a loss function layer in the Mockingjay model; Return to execute the operation of obtaining the target training sample in the training sample set until the training end condition is met, and determine the trained Mockingbird model as the model for voice tag acquisition.

7. A voice tag acquisition device, characterized in that: include: An authorization request module, for generating an authorization request in response to a user's voice device pull-up operation and sending the request to the user; A data collection module, used to collect the audio frame data to be detected through a real-time voice collection device in response to the user's authorization request confirmation operation; A signal enhancement module, used to perform signal enhancement on the audio frame data to be detected through a pre-trained context front-end processing model to obtain the audio frame data to be detected after signal enhancement; A model processing module is used to input the audio frame data to be detected into a voice tag acquisition model trained by the method according to any one of claims 1 to 4 to obtain a tag recognition result matching the audio frame data to be detected.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a voice tag acquisition method according to any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement a voice tag acquisition method according to any one of claims 1 to 5 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program implements a method for acquiring a voice tag according to any one of claims 1 to 5.

Images