Voice quality inspection method and device, equipment and storage medium

Abstract

The invention provides a voice quality inspection method, device and equipment and a storage medium, and belongs to the field of artificial intelligence, and the method comprises the steps: obtaining a single-task voice quality inspection model which is trained based on a preset neural network model and obtains at least two different voice quality inspection tasks; determining a target model parameter according to the model parameter of each single-task voice quality inspection model and the original model parameter of the preset neural network model; using the target model parameter as a model parameter of the preset neural network model to obtain a target voice quality inspection model, the target voice quality inspection model being used for processing a plurality of voice quality inspection tasks; and obtaining target voice to be subjected to quality inspection, and inputting the target voice into the target voice quality inspection model for voice quality inspection to obtain voice quality inspection results of the multiple voice quality inspection tasks. According to the invention, the voice quality inspection flexibility and accuracy are greatly improved. The invention further relates to a block chain technology, and the target voice quality inspection model can be stored in a block chain.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a speech quality inspection method, apparatus, device and storage medium. Background Art

[0002] Voice quality inspection is a key link in evaluating and improving the quality of customer service. It involves analyzing customer service conversation recordings to identify problems so that they can be improved. For example, voice quality inspection includes identifying user emotions, detecting conversation topics, and detecting conversation speed, etc. Traditional voice quality inspection relies on manual listening and evaluation, but this method is time-consuming and costly. Therefore, the technology of voice quality inspection through speech recognition and natural language processing technology has emerged, including but not limited to emotion detection models, conversation topic detection models, and conversation speed detection models, and voice quality inspection is performed based on each model. However, these systems are usually optimized for a single task and are difficult to adapt to the changing quality inspection needs and changing business scenarios. In addition, voice quality inspection through each single task is inefficient and consumes a large amount of computing power and running memory of the detection equipment.

[0003] Therefore, how to improve the efficiency and accuracy of speech quality detection is a problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of this application is to provide a speech quality inspection method, device, equipment and storage medium, aiming to improve the efficiency and accuracy of speech quality inspection.

[0005] In a first aspect, the present application provides a speech quality inspection method, the speech quality inspection method comprising the following steps:

[0006] Acquire a single-task speech quality inspection model for at least two different speech quality inspection tasks based on a preset neural network model training;

[0007] Determining target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model;

[0008] Using the target model parameters as model parameters of the preset neural network model to obtain a target speech quality inspection model, wherein the target speech quality inspection model is used to process a variety of speech quality inspection tasks;

[0009] The target speech to be quality-checked is obtained, and the target speech is input into the target speech quality-checking model for speech quality-checking, so as to obtain speech quality-checking results of various speech quality-checking tasks.

[0010] In a second aspect, the present application further provides a speech quality inspection device, the speech quality inspection device comprising an acquisition module and a generation module, wherein:

[0011] The acquisition module is used to acquire a single-task speech quality inspection model for at least two different speech quality inspection tasks obtained by training based on a preset neural network model;

[0012] The generating module is used to determine the target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model;

[0013] The generating module is further used to use the target model parameters as model parameters of the preset neural network model to obtain a target speech quality inspection model, wherein the target speech quality inspection model is used to process a variety of speech quality inspection tasks;

[0014] The acquisition module is also used to acquire the target speech to be quality inspected;

[0015] The generation module is also used to input the target speech into the target speech quality inspection model for speech quality inspection, so as to obtain speech quality inspection results of various speech quality inspection tasks.

[0016] In a third aspect, the present application also provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the speech quality inspection method as described above are implemented.

[0017] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored, wherein when the computer program is executed by a processor, the steps of the speech quality inspection method as described above are implemented.

[0018] The present application provides a speech quality inspection method, device, equipment and storage medium. The present application obtains a single-task speech quality inspection model for at least two different speech quality inspection tasks by training based on a preset neural network model; determines the target model parameters according to the model parameters of each single-task speech quality inspection model and the original model parameters of the preset neural network model; uses the target model parameters as the model parameters of the preset neural network model to obtain a target speech quality inspection model, which is used to process multiple speech quality inspection tasks; obtains the target speech to be inspected, and inputs the target speech into the target speech quality inspection model for speech quality inspection, and obtains speech quality inspection results of multiple speech quality inspection tasks. In the present application, the target model parameters of the model that can be used to process multiple speech quality inspection tasks are generated by the model parameters of the single-task speech quality inspection model and the original model parameters, and the target speech quality inspection model for processing multiple speech quality inspection tasks can be accurately obtained based on the target model parameters, and then speech quality inspection is performed based on the target speech quality inspection model to obtain speech quality inspection results of multiple speech quality inspection tasks, which greatly improves the flexibility and accuracy of speech quality inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 A flowchart of a voice quality inspection method provided in an embodiment of the present application;

[0021] Figure 2 A flow chart of another voice quality inspection method provided in an embodiment of the present application:

[0022] Figure 3 for Figure 1 A schematic diagram of the sub-step flow chart of the speech quality inspection method in FIG.

[0023] Figure 4 A schematic block diagram of a speech quality inspection device provided in an embodiment of the present application;

[0024] Figure 5 for Figure 4 A schematic block diagram of submodules of a speech quality inspection device in;

[0025] Figure 6 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application.

[0026] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0029] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0030] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0031] Voice quality inspection is a key link in evaluating and improving the quality of customer service. It involves analyzing customer service conversation recordings to identify problems so that they can be improved. For example, voice quality inspection includes identifying user emotions, detecting conversation topics, and detecting conversation speed, etc. Traditional voice quality inspection relies on manual listening and evaluation, but this method is time-consuming and costly. Therefore, the technology of voice quality inspection through speech recognition and natural language processing technology has emerged, including but not limited to emotion detection models, conversation topic detection models, and conversation speed detection models, and voice quality inspection is performed based on each model. However, these systems are usually optimized for a single task and are difficult to adapt to the changing quality inspection needs and changing business scenarios. In addition, voice quality inspection through each single task is inefficient and consumes a large amount of computing power and running memory of the detection equipment.

[0032] To solve the above problems, the embodiments of the present application provide a speech quality inspection method, device, equipment and storage medium. The speech quality inspection method includes: obtaining a single-task speech quality inspection model for at least two different speech quality inspection tasks based on a preset neural network model training; determining the target model parameters according to the model parameters of each single-task speech quality inspection model and the original model parameters of the preset neural network model; using the target model parameters as the model parameters of the preset neural network model to obtain a target speech quality inspection model, which is used to process a variety of speech quality inspection tasks; obtaining a target speech to be inspected, and inputting the target speech into the target speech quality inspection model for speech quality inspection, and obtaining speech quality inspection results for a variety of speech quality inspection tasks.

[0033] Among them, the voice quality inspection method can be applied to computer devices, and the computer devices can be electronic devices such as mobile phones, tablet computers, laptops, desktop computers, personal digital assistants and wearable devices.

[0034] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0035] Please refer to Figure 1 , Figure 1 A flowchart of a speech quality inspection method provided in an embodiment of the present application.

[0036] like Figure 1 As shown, the voice quality inspection method includes steps S101 to S104.

[0037] Step S101: obtaining a single-task speech quality inspection model for at least two different speech quality inspection tasks based on a preset neural network model training.

[0038] Among them, the preset neural network model can be selected according to actual conditions, and the embodiment of the present invention does not make specific limitations on this. For example, the preset neural network model includes but is not limited to a convolutional neural network model, a recurrent neural network model, and a recurrent convolutional neural network model. For example, the preset neural network model is ChatGLM3.

[0039] It should be noted that speech quality inspection tasks include but are not limited to emotion recognition, intention detection, topic detection, speech speed detection, etc. Among them, emotion recognition includes positive emotions, negative emotions and neutral emotions.

[0040] In some embodiments, Figure 2 As shown, before step S101, steps S201 to S205 are also included.

[0041] Step S201: Acquire a sample data set, where the sample data set includes a plurality of sample data, and the sample data includes sample speech and identified single-task speech quality inspection results.

[0042] The sample data set includes multiple sample data, each sample data includes a sample speech and a marked single-task speech quality inspection result. For example, the speech quality inspection task is to detect emotions, and the marked single-task speech quality inspection result is the emotion result corresponding to the sample speech.

[0043] In some embodiments, a sample speech is obtained, and a single-task speech quality inspection result of the sample speech is annotated to obtain an identified single-task speech quality inspection result, the sample speech and the identified single-task speech quality inspection result are integrated to obtain a sample data, and the steps of obtaining the sample speech and annotating the single-task speech quality inspection result of the sample speech to obtain the identified single-task speech quality inspection result, integrating the sample speech and the identified single-task speech quality inspection result to obtain a sample data are repeated in a loop to obtain a sample data set.

[0044] Step S202: Select a sample data from the sample data set as target sample data.

[0045] The method of selecting sample data may be selected according to actual conditions, and the embodiment of the present invention does not specifically limit this. For example, a sample data may be randomly selected as the target sample data; or sample data may be selected as the target sample data according to the time sequence of the sample data generation.

[0046] Step S203: input the sample speech in the target sample data into a preset neural network model for recognition to obtain a predicted single-task speech quality inspection result.

[0047] The sample speech in the target sample data is input into the preset neural network model for recognition to obtain the predicted single-task speech quality inspection result. By recognizing the sample speech through the preset neural network model, the predicted single-task speech quality inspection result can be accurately obtained.

[0048] Step S204: determining whether the preset neural network model has converged according to the predicted single-task speech quality inspection result and the identified single-task speech quality inspection result.

[0049] In some embodiments, the loss value of the preset neural network model is determined based on the predicted single-task speech quality inspection result and the identified single-task speech quality inspection result; if the loss value of the preset neural network model is less than or equal to the preset loss value, it is determined that the preset neural network model has converged; if the loss value of the preset neural network model is greater than the preset loss value, it is determined that the preset neural network model has not converged. Among them, the preset loss value can be set according to actual conditions, and the embodiment of the present invention does not make specific limitations on this. For example, the preset loss value can be 0.002. Through the loss value of the preset neural network model, the convergence of the preset neural network model can be accurately known, which greatly improves the efficiency and accuracy of training the preset neural network model.

[0050] In some embodiments, according to the predicted single-task speech quality inspection result and the identified single-task speech quality inspection result, the loss value of the preset neural network model can be determined by: calculating the cosine similarity of the predicted single-task speech quality inspection result and the identified single-task speech quality inspection result; and determining the value of the unit 1 minus the cosine similarity as the loss value of the preset neural network model. It should be noted that the loss value of the neural network model can also be calculated in other ways, and the implementation of the present invention does not specifically limit this.

[0051] Step S205: If the preset neural network model has not converged, adjust the model parameters of the preset neural network model, and loop through the step of selecting a sample data from the sample data set as the target sample data until the preset neural network model converges.

[0052] Among them, if the loss value of the preset neural network model is greater than the preset loss value, it is determined that the preset neural network model has not converged, and the model parameters of the preset neural network model are adjusted; a sample data is selected from the sample data set as the target sample data; the sample speech in the target sample data is input into the preset neural network model for recognition to obtain a predicted single-task speech quality inspection result; based on the predicted single-task speech quality inspection result and the identified single-task speech quality inspection result, it is determined whether the preset neural network model has converged; if the preset neural network model has not converged, the model parameters of the preset neural network model are adjusted and the above steps are returned to continue to execute the above steps in a loop until the preset neural network model converges to obtain a single-task speech quality inspection model.

[0053] Step S102: determining target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model.

[0054] Among them, the model parameters are the model parameter coefficients of the single-task speech quality inspection model, and the model parameter coefficients include multiple parameter coefficients of each layer of the model. For example, the single-task speech quality inspection model includes an encoder layer, a fully connected layer and a decoder layer, wherein the encoder layer includes 20 parameter coefficients, the fully connected layer includes 50 parameter coefficients and the decoder layer includes 20 parameter coefficients.

[0055] In some embodiments, the model parameter coefficients of each single-task speech quality inspection model are obtained to obtain a plurality of model parameter coefficients; the model parameters of the preset neural network model are obtained to obtain the original model parameters. According to the model parameters of each single-task speech quality inspection model and the original model parameters of the preset neural network model, the target model parameters are determined. Through the model parameters of the single-task speech quality inspection model and the original model parameters of the preset neural network model, the model parameters of the model used to process a variety of speech quality inspection tasks can be determined efficiently and accurately.

[0056] In some embodiments, Figure 3 As shown, step S102 includes sub-steps S1021 and S1022.

[0057] Sub-step S1021, determining a model parameter difference between each of the model parameters and the original model parameters to obtain a plurality of model parameter difference values.

[0058] Among them, both the model parameters and the original model parameters include multiple parameter coefficients.

[0059] In some embodiments, the difference operation is performed on the parameter coefficients in each model parameter at the same parameter coefficient position as the original model parameter in turn to obtain the model parameter difference between each model parameter and the original model parameter, and the model parameter difference includes the sub-parameter difference at each parameter coefficient position. By performing the difference operation on the parameter coefficients in each model parameter at the same parameter coefficient position as the original model parameter, multiple model parameter differences can be accurately obtained.

[0060] Exemplarily, a single-task speech quality inspection model includes an emotion recognition model and an intent detection model; the model parameters of the emotion recognition model are obtained to obtain first model parameters; the model parameters of the intent detection model are obtained to obtain second model parameters; the model parameters of the preset neural network model are obtained to obtain original model parameters; wherein, the first model parameters include parameter coefficient 1, parameter coefficient 2, parameter coefficient 3, parameter coefficient 4 and parameter coefficient 5; the second model parameters include parameter coefficient 6, parameter coefficient 7, parameter coefficient 8, parameter coefficient 9 and parameter coefficient 10; the original model parameters include parameter coefficient 11, parameter coefficient 12, parameter coefficient 13, parameter coefficient 14 and parameter coefficient 15.

[0061] Subtract parameter coefficient 11 from parameter coefficient 1 to obtain sub-parameter difference 1; subtract parameter coefficient 11 from parameter coefficient 2 to obtain sub-parameter difference 2; subtract parameter coefficient 11 from parameter coefficient 3 to obtain sub-parameter difference 3; subtract parameter coefficient 11 from parameter coefficient 4 to obtain sub-parameter difference 4; subtract parameter coefficient 11 from parameter coefficient 5 to obtain sub-parameter difference 5; integrate sub-parameter difference 1, sub-parameter difference 2, sub-parameter difference 3, sub-parameter difference 4 and sub-parameter difference 5 to obtain the first model parameter difference.

[0062] Subtract parameter coefficient 11 from parameter coefficient 6 to obtain sub-parameter difference 6; subtract parameter coefficient 11 from parameter coefficient 7 to obtain sub-parameter difference 7; subtract parameter coefficient 11 from parameter coefficient 8 to obtain sub-parameter difference 8; subtract parameter coefficient 11 from parameter coefficient 9 to obtain sub-parameter difference 9; subtract parameter coefficient 11 from parameter coefficient 10 to obtain sub-parameter difference 10; integrate sub-parameter difference 6, sub-parameter difference 7, sub-parameter difference 8, sub-parameter difference 9 and sub-parameter difference 10 to obtain the second model parameter difference.

[0063] Sub-step S1022: synthesizing the model parameters of the multi-task speech quality inspection model according to the model parameter differences and the original model parameters to obtain target model parameters.

[0064] In some embodiments, multiple sub-parameter differences in each model parameter difference are screened, and unscreened sub-parameter differences are set to zero to obtain multiple candidate model parameter differences; each sub-parameter difference in each candidate model parameter difference is amplified to obtain multiple target model parameter differences; and model parameters of a multi-task speech quality inspection model are synthesized based on multiple target model parameter differences and original model parameters to obtain target model parameters. By synthesizing model parameters of a multi-task speech quality inspection model based on multiple target model parameter differences and original model parameters, the target model parameters can be accurately obtained.

[0065] In some embodiments, the method for screening multiple sub-parameter differences in each model parameter difference can be: performing absolute value processing on the sub-parameter difference at each parameter coefficient position in the model parameter difference to obtain the absolute value of the sub-difference coefficient corresponding to each sub-parameter difference; sorting the absolute value of each sub-difference coefficient according to the absolute value to obtain a sub-difference coefficient sequence; selecting a preset proportion of sub-difference coefficients from the sub-difference coefficient sequence at a preset interval to obtain multiple target sub-difference coefficients, the preset proportion being the proportion of the number of target sub-parameter differences to the total number of sub-parameter differences; obtaining the sub-parameter difference corresponding to each target sub-difference coefficient, and determining the obtained sub-parameter difference as the screened target sub-parameter difference. Among them, the preset interval and the preset proportion can be set according to actual conditions, and the embodiment of the present invention does not specifically limit this. For example, the preset interval is 1 and the preset proportion is 0.2.

[0066] Exemplarily, the first model parameter difference includes sub-parameter difference 1, sub-parameter difference 2, sub-parameter difference 3, sub-parameter difference 4, sub-parameter difference 5, sub-parameter difference 6, sub-parameter difference 7, sub-parameter difference 8, sub-parameter difference 9 and sub-parameter difference 10; each sub-parameter difference is processed by absolute value, and the absolute value of each sub-difference coefficient is sorted according to the absolute value, so that the sub-difference coefficient sequence is sub-parameter difference 1, sub-parameter difference 5, sub-parameter difference 7, sub-parameter difference 3, sub-parameter difference 2, sub-parameter difference 10. The preset interval is 2, and the preset ratio is 0.3; then the target sub-parameter difference includes sub-parameter difference 1, sub-parameter difference 3, and sub-parameter difference 10; or the target sub-parameter difference includes sub-parameter difference 5, sub-parameter difference 2, and sub-parameter difference; or the target sub-parameter difference includes sub-parameter difference 7, sub-parameter difference 6, and sub-parameter difference 8; or the target sub-parameter difference includes sub-parameter difference 3, sub-parameter difference 10, and sub-parameter difference 9.

[0067] In some embodiments, each sub-parameter difference in each candidate model parameter difference is amplified to obtain multiple target model parameter differences by dividing each sub-parameter difference by the preset ratio to obtain the target model parameter difference corresponding to each sub-parameter difference.

[0068] In some embodiments, the model parameters of the multi-task speech quality inspection model are synthesized according to multiple target model parameter differences and original model parameters, and the target model parameters can be obtained by: performing mean processing on the sub-parameter differences at the same parameter coefficient position in each target model parameter difference to obtain the total model parameter difference; performing addition processing on the total model parameter difference and the sub-parameter differences at the same parameter coefficient position in the original model parameters to obtain the target model parameters. The target model parameters are accurately obtained by performing addition processing on the total model parameter difference and the sub-parameter differences at the same parameter coefficient position in the original model parameters.

[0069] Step S103: Using the target model parameters as model parameters of the preset neural network model to obtain a target speech quality inspection model, wherein the target speech quality inspection model is used to process a variety of speech quality inspection tasks.

[0070] In some embodiments, the target model parameters are used as model parameters of a preset neural network model to obtain a target speech quality inspection model, which is used to process a variety of speech quality inspection tasks. By using the target model parameters as model parameters of a preset neural network model, the target speech quality inspection model for a variety of speech quality inspection tasks can be accurately processed. By synthesizing a target speech quality inspection model for processing a variety of speech quality inspection tasks, a variety of speech quality inspection tasks can be completed, and the number of models can be reduced, thereby reducing the occupancy rate of the device and the CPU usage, thereby improving the fluency of the device.

[0071] It should be noted that the single-task speech quality inspection model includes an emotion recognition model, an intention detection model, a topic detection model and a speaking rate detection model; the synthesized target speech quality inspection model can perform emotion recognition, intention detection, topic detection and speaking rate detection.

[0072] Step S104: obtaining a target speech to be quality-checked, and inputting the target speech into the target speech quality-checking model for speech quality-checking, to obtain speech quality-checking results of various speech quality-checking tasks.

[0073] The target speech to be tested is obtained, and the target speech is input into the target speech quality inspection model for speech quality inspection, so as to obtain speech quality inspection results of various speech quality inspection tasks. This greatly improves the flexibility, efficiency and accuracy of speech quality inspection, thereby improving the user experience.

[0074] The speech quality inspection method provided in the above embodiment obtains a single-task speech quality inspection model for at least two different speech quality inspection tasks based on a preset neural network model training; determines the target model parameters according to the model parameters of each single-task speech quality inspection model and the original model parameters of the preset neural network model; uses the target model parameters as the model parameters of the preset neural network model to obtain a target speech quality inspection model, which is used to process a variety of speech quality inspection tasks; obtains the target speech to be inspected, and inputs the target speech into the target speech quality inspection model for speech quality inspection, and obtains speech quality inspection results for a variety of speech quality inspection tasks. In the present application, the target model parameters of the model that can be used to process a variety of speech quality inspection tasks are generated by the model parameters of the single-task speech quality inspection model and the original model parameters, and the target speech quality inspection model for processing a variety of speech quality inspection tasks can be accurately obtained based on the target model parameters, and then speech quality inspection is performed based on the target speech quality inspection model to obtain speech quality inspection results for a variety of speech quality inspection tasks, which greatly improves the flexibility and accuracy of speech quality inspection.

[0075] See also Figure 4 , Figure 4 A schematic block diagram of a speech quality inspection device provided in an embodiment of the present application.

[0076] like Figure 4 As shown, the speech quality inspection device 300 includes an acquisition module 310 and a generation module 320, wherein:

[0077] The acquisition module 310 is used to acquire a single-task speech quality inspection model for at least two different speech quality inspection tasks obtained by training based on a preset neural network model;

[0078] The generating module 320 is used to determine the target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model;

[0079] The generating module 320 is further used to use the target model parameters as model parameters of the preset neural network model to obtain a target speech quality inspection model, wherein the target speech quality inspection model is used to process a variety of speech quality inspection tasks;

[0080] The acquisition module 310 is also used to acquire the target speech to be quality inspected;

[0081] The generating module 320 is further used to input the target speech into the target speech quality inspection model for speech quality inspection, so as to obtain speech quality inspection results of various speech quality inspection tasks.

[0082] In some embodiments, Figure 5As shown, the generating module 320 includes a first generating submodule 321 and a second generating submodule 322, wherein:

[0083] The first generating submodule 321 is used to determine the model parameter difference between each of the model parameters and the original model parameter to obtain a plurality of model parameter difference values;

[0084] The second generating submodule 322 is used to synthesize the model parameters of the multi-task speech quality inspection model according to the model parameter differences and the original model parameters to obtain the target model parameters.

[0085] In some embodiments, the first generating submodule 321 is further used to:

[0086] The difference operation is performed on the parameter coefficients in each of the model parameters at the same parameter coefficient position as in the original model parameters in turn to obtain the model parameter difference between each of the model parameters and the original model parameters, wherein the model parameter difference includes the sub-parameter difference at each parameter coefficient position.

[0087] In some embodiments, the second generating submodule 322 is further used to:

[0088] Screening multiple sub-parameter differences in each of the model parameter differences, and setting unscreened sub-parameter differences to zero, to obtain multiple candidate model parameter differences;

[0089] Amplifying each sub-parameter difference in each of the candidate model parameter differences to obtain a plurality of target model parameter differences;

[0090] The model parameters of the multi-task speech quality inspection model are synthesized according to the multiple target model parameter differences and the original model parameters to obtain the target model parameters.

[0091] In some embodiments, the second generating submodule 322 is further used to:

[0092] Performing absolute value processing on the sub-parameter difference at each parameter coefficient position in the model parameter difference to obtain the absolute value of the sub-difference coefficient corresponding to each sub-parameter difference;

[0093] The absolute values ​​of the sub-difference coefficients are sorted according to their absolute values ​​to obtain a sub-difference coefficient sequence;

[0094] Selecting a preset proportion of sub-difference coefficients from the sub-difference coefficient sequence at preset intervals to obtain a plurality of target sub-difference coefficients, wherein the preset proportion is a ratio of the number of target sub-parameter differences to the total number of sub-parameter differences;

[0095] The sub-parameter difference corresponding to each target sub-difference coefficient is obtained, and the obtained sub-parameter difference is determined as the screened target sub-parameter difference.

[0096] In some embodiments, the second generating submodule 322 is further used to:

[0097] Performing mean processing on sub-parameter differences at the same parameter coefficient position in each of the target model parameter differences to obtain a total model parameter difference;

[0098] The total model parameter difference and the sub-parameter difference at the same parameter coefficient position in the original model parameter are added to obtain the target model parameter.

[0099] In some embodiments, the speech quality inspection device 300 is further used for:

[0100] Acquire a sample data set, the sample data set comprising a plurality of sample data, the sample data comprising sample speech and an identified single-task speech quality inspection result;

[0101] Selecting a sample data from the sample data set as target sample data;

[0102] Inputting the sample speech in the target sample data into a preset neural network model for recognition to obtain a predicted single-task speech quality inspection result;

[0103] Determining whether the preset neural network model converges according to the predicted single-task speech quality inspection result and the identified single-task speech quality inspection result;

[0104] If the preset neural network model has not converged, the model parameters of the preset neural network model are adjusted, and the step of selecting a sample data from the sample data set as the target sample data is executed repeatedly until the preset neural network model converges.

[0105] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above-mentioned speech quality inspection device can refer to the corresponding process in the aforementioned speech quality inspection method embodiment, and will not be repeated here.

[0106] See also Figure 6 , Figure 6 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application.

[0107] like Figure 6 As shown, the computer device 400 includes a processor 402 and a memory 403 connected via a system bus 401 , wherein the memory 403 may include a storage medium and an internal memory.

[0108] The storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any speech quality inspection method.

[0109] The processor 402 is used to provide computing and control capabilities and support the operation of the entire computer device.

[0110] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute any speech quality inspection method.

[0111] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0112] It should be understood that the processor 402 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0113] In one embodiment, the processor 402 is used to run a computer program stored in the memory to implement the following steps:

[0114] Acquire a single-task speech quality inspection model for at least two different speech quality inspection tasks based on a preset neural network model training;

[0115] Determining target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model;

[0116] Using the target model parameters as model parameters of the preset neural network model to obtain a target speech quality inspection model, wherein the target speech quality inspection model is used to process a variety of speech quality inspection tasks;

[0117] The target speech to be quality-checked is obtained, and the target speech is input into the target speech quality-checking model for speech quality-checking, so as to obtain speech quality-checking results of various speech quality-checking tasks.

[0118] In one embodiment, the processor 402, when determining the target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model, is used to implement:

[0119] Determine a model parameter difference between each of the model parameters and the original model parameter to obtain a plurality of model parameter difference values;

[0120] The model parameters of the multi-task speech quality inspection model are synthesized according to the model parameter differences and the original model parameters to obtain the target model parameters.

[0121] In one embodiment, each of the model parameters and the original model parameters include a plurality of parameter coefficients, and when the processor 402 determines the model parameter difference between each of the model parameters and the original model parameters to obtain a plurality of model parameter difference values, it is used to implement:

[0122] The difference operation is performed on the parameter coefficients in each of the model parameters at the same parameter coefficient position as in the original model parameters in turn to obtain the model parameter difference between each of the model parameters and the original model parameters, wherein the model parameter difference includes the sub-parameter difference at each parameter coefficient position.

[0123] In one embodiment, the processor 402, when implementing the synthesis of the model parameters of the multi-task speech quality inspection model according to each of the model parameter differences and the original model parameters to obtain the target model parameters, is used to implement:

[0124] Screening multiple sub-parameter differences in each of the model parameter differences, and setting unscreened sub-parameter differences to zero, to obtain multiple candidate model parameter differences;

[0125] Amplifying each sub-parameter difference in each of the candidate model parameter differences to obtain a plurality of target model parameter differences;

[0126] The model parameters of the multi-task speech quality inspection model are synthesized according to the multiple target model parameter differences and the original model parameters to obtain the target model parameters.

[0127] In one embodiment, when implementing the screening of the multiple sub-parameter differences in each of the model parameter differences, the processor 402 is used to implement:

[0128] Performing absolute value processing on the sub-parameter difference at each parameter coefficient position in the model parameter difference to obtain the absolute value of the sub-difference coefficient corresponding to each sub-parameter difference;

[0129] The absolute values ​​of the sub-difference coefficients are sorted according to their absolute values ​​to obtain a sub-difference coefficient sequence;

[0130] Selecting a preset proportion of sub-difference coefficients from the sub-difference coefficient sequence at preset intervals to obtain a plurality of target sub-difference coefficients, wherein the preset proportion is a ratio of the number of target sub-parameter differences to the total number of sub-parameter differences;

[0131] The sub-parameter difference corresponding to each target sub-difference coefficient is obtained, and the obtained sub-parameter difference is determined as the screened target sub-parameter difference.

[0132] In one embodiment, when the processor 402 implements the synthesis of model parameters of the multi-task speech quality inspection model according to the multiple target model parameter differences and the original model parameters to obtain the target model parameters, it is used to implement:

[0133] Performing mean processing on sub-parameter differences at the same parameter coefficient position in each of the target model parameter differences to obtain a total model parameter difference;

[0134] The total model parameter difference and the sub-parameter difference at the same parameter coefficient position in the original model parameter are added to obtain the target model parameter.

[0135] In one embodiment, the processor 402 is further configured to implement:

[0136] Acquire a sample data set, the sample data set comprising a plurality of sample data, the sample data comprising sample speech and an identified single-task speech quality inspection result;

[0137] Selecting a sample data from the sample data set as target sample data;

[0138] Inputting the sample speech in the target sample data into a preset neural network model for recognition to obtain a predicted single-task speech quality inspection result;

[0139] Determining whether the preset neural network model converges according to the predicted single-task speech quality inspection result and the identified single-task speech quality inspection result;

[0140] If the preset neural network model has not converged, the model parameters of the preset neural network model are adjusted, and the step of selecting a sample data from the sample data set as the target sample data is executed repeatedly until the preset neural network model converges.

[0141] It should be noted that technical personnel in the relevant field can clearly understand that for the convenience and brevity of description, the specific working process of the computer device described above can refer to the corresponding process in the aforementioned speech quality inspection method embodiment, and will not be repeated here.

[0142] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions. The method implemented when the program instructions are executed can refer to the various embodiments of the speech quality inspection method of the present application.

[0143] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may be non-volatile or volatile. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.

[0144] Furthermore, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the blockchain node, etc.

[0145] The blockchain referred to in this invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0146] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0147] It should also be understood that the term "and / or" used in the present specification refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0148] The serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application.

Claims

1. A speech quality inspection method, characterized in that: include: Acquire a single-task speech quality inspection model for at least two different speech quality inspection tasks based on a preset neural network model training; Determining target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model; Using the target model parameters as model parameters of the preset neural network model to obtain a target speech quality inspection model, wherein the target speech quality inspection model is used to process a variety of speech quality inspection tasks; The target speech to be quality-checked is obtained, and the target speech is input into the target speech quality-checking model for speech quality-checking, so as to obtain speech quality-checking results of various speech quality-checking tasks.

2. The speech quality inspection method according to claim 1, characterized in that: Determining target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model includes: Determine a model parameter difference between each of the model parameters and the original model parameter to obtain a plurality of model parameter difference values; The model parameters of the multi-task speech quality inspection model are synthesized according to the model parameter differences and the original model parameters to obtain the target model parameters.

3. The speech quality inspection method according to claim 2, characterized in that: Each of the model parameters and the original model parameters includes a plurality of parameter coefficients; the step of determining a model parameter difference between each of the model parameters and the original model parameters to obtain a plurality of model parameter differences includes: The difference operation is performed on the parameter coefficients in each of the model parameters at the same parameter coefficient position as in the original model parameters in turn to obtain the model parameter difference between each of the model parameters and the original model parameters, wherein the model parameter difference includes the sub-parameter difference at each parameter coefficient position.

4. The speech quality inspection method according to claim 3, characterized in that: The synthesizing of the model parameters of the multi-task speech quality inspection model according to the model parameter differences and the original model parameters to obtain the target model parameters includes: Screening multiple sub-parameter differences in each of the model parameter differences, and setting unscreened sub-parameter differences to zero, to obtain multiple candidate model parameter differences; Amplifying each sub-parameter difference in each of the candidate model parameter differences to obtain a plurality of target model parameter differences; The model parameters of the multi-task speech quality inspection model are synthesized according to the multiple target model parameter differences and the original model parameters to obtain the target model parameters.

5. The speech quality inspection method according to claim 4, characterized in that: The screening of the multiple sub-parameter differences in each of the model parameter differences includes: Performing absolute value processing on the sub-parameter difference at each parameter coefficient position in the model parameter difference to obtain the absolute value of the sub-difference coefficient corresponding to each sub-parameter difference; The absolute values ​​of the sub-difference coefficients are sorted according to their absolute values ​​to obtain a sub-difference coefficient sequence; Selecting a preset proportion of sub-difference coefficients from the sub-difference coefficient sequence at preset intervals to obtain a plurality of target sub-difference coefficients, wherein the preset proportion is a ratio of the number of target sub-parameter differences to the total number of sub-parameter differences; The sub-parameter difference corresponding to each target sub-difference coefficient is obtained, and the obtained sub-parameter difference is determined as the screened target sub-parameter difference.

6. The speech quality inspection method according to claim 4, characterized in that: The synthesizing the model parameters of the multi-task speech quality inspection model according to the multiple target model parameter differences and the original model parameters to obtain the target model parameters includes: Performing mean processing on sub-parameter differences at the same parameter coefficient position in each of the target model parameter differences to obtain a total model parameter difference; The total model parameter difference and the sub-parameter difference at the same parameter coefficient position in the original model parameter are added to obtain the target model parameter.

7. The speech quality inspection method according to any one of claims 1 to 6, characterized in that: The method further comprises: Acquire a sample data set, the sample data set comprising a plurality of sample data, the sample data comprising sample speech and an identified single-task speech quality inspection result; Selecting a sample data from the sample data set as target sample data; Inputting the sample speech in the target sample data into a preset neural network model for recognition to obtain a predicted single-task speech quality inspection result; Determining whether the preset neural network model converges according to the predicted single-task speech quality inspection result and the identified single-task speech quality inspection result; If the preset neural network model has not converged, the model parameters of the preset neural network model are adjusted, and the step of selecting a sample data from the sample data set as the target sample data is executed repeatedly until the preset neural network model converges.

8. A speech quality inspection device, characterized in that: The speech quality inspection device comprises an acquisition module and a generation module, wherein: The acquisition module is used to acquire a single-task speech quality inspection model for at least two different speech quality inspection tasks obtained by training based on a preset neural network model; The generating module is used to determine the target model parameters according to the model parameters of each of the single-task speech quality inspection models and the original model parameters of the preset neural network model; The generating module is further used to use the target model parameters as model parameters of the preset neural network model to obtain a target speech quality inspection model, wherein the target speech quality inspection model is used to process a variety of speech quality inspection tasks; The acquisition module is also used to acquire the target speech to be quality inspected; The generation module is also used to input the target speech into the target speech quality inspection model for speech quality inspection, so as to obtain speech quality inspection results of various speech quality inspection tasks.

9. A computer device, characterized in that: The computer device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the speech quality inspection method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the speech quality inspection method according to any one of claims 1 to 7 are implemented.

Images