Voice quality inspection method, device, equipment and storage medium

Abstract

The present application discloses a method, device, equipment and storage medium for speech quality inspection, the method comprising: receiving speech data to be inspected, converting the speech data to text to obtain text data to be inspected; identifying whether the text data to be inspected has various preset quality inspection items to obtain a target recognition result; and evaluating the speech data to be inspected according to the target recognition result. The present application solves the technical problem that speech quality inspection is performed manually in the prior art, the quality inspection efficiency is low and the quality inspection is not comprehensive.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology in financial technology (Fintech), and in particular to a voice quality inspection method, device, equipment and storage medium. Background Art

[0002] With the continuous development of financial technology, especially Internet technology finance, more and more technologies are applied in the financial field, but the financial industry also puts forward higher requirements for technology, such as the financial industry also has higher requirements for voice quality inspection.

[0003] At present, enterprises conduct voice collection through customer service telephones, such as financial collection by telephone. In order to evaluate the quality of telephone customer service and collection, it is often necessary to conduct quality inspection on audio files such as recordings of telephone collection. Among them, quality inspection is to extract the customer service voice content, determine whether the customer service voice content contains corresponding quality inspection items, and then conduct quality inspection scoring. The quality inspection extraction and scoring in the existing quality inspection process are all completed manually, and manual spot checks have certain limitations, consume a lot of human resources, and are difficult to conduct comprehensive spot checks. Summary of the invention

[0004] The main purpose of the present application is to provide a speech quality inspection method, device, equipment and storage medium, aiming to solve the technical problems in the prior art of performing speech quality inspection manually, with low quality inspection efficiency and incomplete quality inspection.

[0005] To achieve the above object, the present application provides a voice quality inspection method, the voice quality inspection method comprising:

[0006] Receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain text data to be inspected;

[0007] Identify whether the text data to be inspected has each preset quality inspection item to obtain a target recognition result;

[0008] The speech data to be quality-checked is evaluated according to the target recognition result.

[0009] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0010] The step of identifying whether the text data to be inspected has each preset quality inspection item to obtain a target recognition result includes:

[0011] Using the first recognition sub-model in the preset quality inspection model, the text data to be inspected is identified to determine whether it has each preset quality inspection element, thereby obtaining an initial recognition result;

[0012] Acquire initial text data corresponding to the initial recognition result, input the initial text data into a second recognition sub-model in the preset quality inspection model, and identify the strength of each preset quality inspection element of the initial text data through the second recognition sub-model to obtain the target recognition result;

[0013] Among them, the first recognition sub-model is obtained by iteratively training a first preset prediction model to be trained based on a first preset text set with various preset quality inspection element labels by executing a first preset training process, and the second recognition sub-model is obtained by iteratively training a second preset prediction model to be trained based on a second preset text set with various strength labels of various preset quality inspection elements by executing a second preset training process.

[0014] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0015] The step of identifying whether the text data to be inspected has each preset quality inspection item to obtain a target recognition result includes:

[0016] Performing matching processing of a preset first pattern on each first text at each preset determined position in the text data to be inspected, and obtaining a first identification sub-result of whether each first text has a preset quality inspection item;

[0017] Locating each second text associated with each preset quality inspection element from the text data to be quality inspected, inputting each second text into a preset classification model, obtaining a predicted quality inspection element label of each second text, and obtaining a second identification sub-result;

[0018] The first recognition sub-result and the second recognition sub-result are set as the target recognition result.

[0019] Optionally, the step of setting the first recognition sub-result and the second recognition sub-result as the target recognition result includes:

[0020] The first text and the second text are used as candidate texts, and a matching process of a preset second mode is performed to determine whether there is a misjudgment of a preset quality inspection element, and obtain a misjudgment recognition result;

[0021] Based on the misjudgment recognition result, misjudgment elimination processing is performed on the first recognition sub-result and the second recognition sub-result to obtain a target recognition result.

[0022] Optionally, before the step of locating each second text associated with each preset quality inspection element from the text data to be quality inspected and inputting each second text into a preset classification model, the method includes:

[0023] Acquire preset classified text data, extract a feature matrix from the preset classified text data based on a preset convolution layer and a preset pooling layer in a third preset prediction model to be trained, and perform prediction processing on the feature matrix according to a preset activation layer of the third preset prediction model to be trained to obtain a predicted quality inspection factor label set;

[0024] The predicted quality inspection element label set is compared with the actual label set in the preset classified text data to obtain a comparison result, so as to adjust the parameters of the preset convolution layer and the preset pooling layer until the similarity between the predicted quality inspection element label set and the actual label set is greater than a preset similarity threshold, so as to obtain the preset classification model.

[0025] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0026] The step of evaluating the speech data to be quality-checked according to the target recognition result comprises:

[0027] Obtaining the score weight ratio of each preset quality inspection item, and obtaining the score of the speech data to be quality inspected according to the score weight ratio and the target recognition result;

[0028] Based on the scoring of the speech data to be quality inspected, a scoring report is formed and reported.

[0029] Optionally, the step of receiving the voice data to be inspected, performing text conversion processing on the voice data to be inspected, and obtaining the text data to be inspected includes:

[0030] Receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain first processed data;

[0031] The first processed data is subjected to a preset preprocessing to obtain text data to be quality inspected.

[0032] The present application also provides a speech quality inspection device, the speech quality inspection device comprising:

[0033] A receiving module, used for receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain text data to be inspected;

[0034] An identification module is used to identify whether the text data to be inspected has various preset quality inspection items to obtain a target identification result;

[0035] An evaluation module is used to evaluate the speech data to be quality-checked according to the target recognition result.

[0036] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0037] The identification module comprises:

[0038] An identification unit, used to identify whether the text data to be quality inspected has each preset quality inspection element through a first identification sub-model in a preset quality inspection model, and obtain an initial identification result;

[0039] A first acquisition unit is used to acquire initial text data corresponding to the initial recognition result, input the initial text data into a second recognition sub-model in the preset quality inspection model, and identify the strength of each preset quality inspection element of the initial text data through the second recognition sub-model to obtain the target recognition result;

[0040] Among them, the first recognition sub-model is obtained by iteratively training a first preset prediction model to be trained based on a first preset text set with various preset quality inspection element labels by executing a first preset training process, and the second recognition sub-model is obtained by iteratively training a second preset prediction model to be trained based on a second preset text set with various strength labels of various preset quality inspection elements by executing a second preset training process.

[0041] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0042] The identification module also includes:

[0043] A matching unit, used for performing a matching process of a preset first mode on each first text at each preset determined position in the text data to be inspected, and obtaining a first identification sub-result of whether each first text has a preset quality inspection item;

[0044] A positioning unit, used to locate each second text associated with each preset quality inspection element from the text data to be quality inspected, input each second text into a preset classification model, obtain a predicted quality inspection element label of each second text, and obtain a second identification sub-result;

[0045] A setting unit is configured to set the first recognition sub-result and the second recognition sub-result as the target recognition result.

[0046] Optionally, the setting unit includes:

[0047] A matching subunit, configured to use the first text and the second text as candidate texts, and perform matching processing of a preset second mode to determine whether there is a misjudgment of a preset quality inspection element, and obtain a misjudgment recognition result;

[0048] The clearing subunit is used to perform misjudgment clearing processing on the first recognition sub-result and the second recognition sub-result based on the misjudgment recognition result to obtain a target recognition result.

[0049] Optionally, the speech quality inspection device further includes:

[0050] An acquisition module is used to acquire preset classified text data, extract a feature matrix from the preset classified text data based on a preset convolution layer and a preset pooling layer in a third preset prediction model to be trained, and perform prediction processing on the feature matrix according to a preset activation layer of the third preset prediction model to be trained to obtain a predicted quality inspection factor label set;

[0051] A comparison module is used to compare the predicted quality inspection element label set with the actual label set in the preset classified text data to obtain a comparison result, so as to adjust the parameters of the preset convolution layer and the preset pooling layer until the similarity between the predicted quality inspection element label set and the actual label set is greater than a preset similarity threshold, so as to obtain the preset classification model.

[0052] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0053] The evaluation module includes:

[0054] A second acquisition unit is used to acquire the score weight ratio of each preset quality inspection item, and obtain the score of the speech data to be quality inspected according to the score weight ratio and the target recognition result;

[0055] The reporting module is used to form a scoring report based on the scoring of the speech data to be quality inspected, and report the scoring report.

[0056] Optionally, the receiving module includes:

[0057] A receiving unit, configured to receive voice data to be inspected, and perform text conversion processing on the voice data to be inspected to obtain first processed data;

[0058] The preprocessing unit is used to perform preset preprocessing on the first processed data to obtain text data to be quality inspected.

[0059] The present application also provides a speech quality inspection device, which is a physical device, and includes: a memory, a processor, and a program of the speech quality inspection method stored in the memory and executable on the processor. When the program of the speech quality inspection method is executed by the processor, the steps of the speech quality inspection method as described above can be implemented.

[0060] The present application also provides a storage medium, on which is stored a program for implementing the above-mentioned voice quality inspection method. When the program of the voice quality inspection method is executed by a processor, the steps of the above-mentioned voice quality inspection method are implemented.

[0061] The present application receives voice data to be inspected, performs text conversion processing on the voice data to be inspected, and obtains text data to be inspected; identifies whether the text data to be inspected has various preset quality inspection items to obtain a target recognition result; and evaluates the voice data to be inspected based on the target recognition result. In the present application, after receiving the voice data to be quality inspected, the voice data to be quality inspected is converted into text to obtain the text data to be quality inspected, and then the text data to be quality inspected is automatically identified to see whether it has various preset quality inspection items to obtain a target recognition result. Since the text data to be quality inspected is automatically identified to see whether it has various preset quality inspection items to obtain a target recognition result, and based on the target recognition result, the voice data to be quality inspected is automatically evaluated instead of requiring manual quality inspection, thereby saving resources and improving quality inspection efficiency. Moreover, since quality inspection is performed immediately after receiving the voice data to be quality inspected, the comprehensiveness of quality inspection is improved (that is, all voice data to be quality inspected are quality inspected), thereby solving the technical problems of low quality inspection efficiency and incomplete quality inspection in the prior art that voice quality inspection is performed manually. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0063] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0064] Figure 1 This is a flow chart of the first embodiment of the speech quality inspection method of the present application;

[0065] Figure 2 A schematic flow chart of the refinement steps of identifying whether the text data to be inspected has each preset quality inspection item and obtaining a target recognition result in the first embodiment of the speech quality inspection method of the present application;

[0066] Figure 3 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0067] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0068] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0069] The present application embodiment provides a method for speech quality inspection. In the first embodiment of the speech quality inspection method of the present application, refer to Figure 1 , the voice quality inspection method comprises:

[0070] Step S10, receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain text data to be inspected;

[0071] Step S20, identifying whether the text data to be inspected has each preset quality inspection item, and obtaining a target recognition result;

[0072] Step S30: evaluating the speech data to be quality-checked according to the target recognition result.

[0073] The specific steps are as follows:

[0074] Step S10, receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain text data to be inspected;

[0075] In the present embodiment, it should be noted that the voice quality inspection method is applied to a voice quality inspection system, which is subordinate to a voice quality inspection device. The voice quality inspection system communicates with a telephone collection customer service platform, wherein the telephone collection customer service platform records the telephone collection content of each telephone collector and obtains the recording content. The voice data to be inspected can be obtained from the recording content. Specifically, since the recording content includes the voice content of the person being collected, in order to reduce the data to be processed, the recording content is voice-cut to remove the recording of the person being collected (specifically, the recording of the person being collected is determined through language recognition technology) to obtain the recording data of the collector. The data of the collection personnel's recording is used as the voice data to be inspected and sent to the voice quality inspection system. After receiving the voice data to be inspected, the voice quality inspection system performs text conversion processing on the voice data to be inspected to obtain text data to be inspected.

[0076] The step of receiving the voice data to be inspected, converting the voice data to be inspected into text, and obtaining the text data to be inspected comprises:

[0077] Step S11, receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain first processed data;

[0078] In this embodiment, after obtaining the speech data to be inspected, the speech data to be inspected is converted into text using a preset language conversion model, such as using a preset ASR model to convert the speech data to be inspected into text, to obtain first processed data.

[0079] Step S12, performing preset preprocessing on the first processed data to obtain text data to be quality inspected.

[0080] After obtaining the first processed data, due to errors in the text conversion process or because the debt collector has an accent, etc., it is necessary to perform preset preprocessing on the first processed data. Specifically, the text data is preprocessed by performing word segmentation, error correction, rewriting and other preprocessing through preset NLP (Natural Language Processing) and other technologies to obtain the first processed data. In this embodiment, it should be noted that due to the preprocessing, the content that needs quality inspection can be accurately obtained, laying the foundation for accurate quality inspection.

[0081] Step S20, identifying whether the text data to be inspected has each preset quality inspection item, and obtaining a target recognition result;

[0082] After obtaining the text data to be inspected, the text data to be inspected is identified to determine whether it has various preset quality inspection items, and a target identification result is obtained, wherein the quality inspection items refer to the text of business knowledge, pressure points, communication skills, opening remarks, closing remarks, etc. For example, the text of the opening remarks is "Hello! This is a staff of xx Bank calling you. Is this Mr. / Ms. XXX?" etc. Identifying whether the text data to be inspected has various preset quality inspection items includes identifying whether the text data to be inspected has all preset quality inspection items, wherein each preset quality inspection item includes multiple preset quality inspection elements, for example, the quality inspection item "pressure points" includes multiple quality inspection elements such as informing the overdue days, informing that interest and corresponding penalty interest will be charged for overdue, informing that the credit report of the People's Bank will be reported, informing the severity of the credit report, informing that the bank does not exclude sending letters to the customer's unit or home for continued overdue, informing that attention should be paid to personal credit records, and business knowledge includes multiple quality inspection elements such as which applications to repay, how to repay, and how to check whether the repayment is successful. Therefore, identifying whether the text data to be quality inspected has each preset quality inspection item refers to identifying whether the text data to be quality inspected has each preset quality inspection element in each preset quality inspection item. By identifying whether the text data to be quality inspected has each preset quality inspection item (including each preset quality inspection element), the target recognition result is obtained.

[0083] Wherein, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0084] In this embodiment, a method for obtaining a target recognition result is provided, namely, method 1: the step of identifying whether the text data to be inspected has each preset quality inspection item to obtain the target recognition result includes:

[0085] Step S21, identifying whether the text data to be inspected has each preset quality inspection element through the first recognition sub-model in the preset quality inspection model, and obtaining an initial recognition result;

[0086] Step S22, obtaining initial text data corresponding to the initial recognition result, inputting the initial text data into a second recognition sub-model in the preset quality inspection model, and performing recognition of the strength of each preset quality inspection element on the initial text data through the second recognition sub-model to obtain the target recognition result;

[0087] Among them, the first recognition sub-model is obtained by iteratively training a first preset prediction model to be trained based on a first preset text set with various preset quality inspection element labels by executing a first preset training process, and the second recognition sub-model is obtained by iteratively training a second preset prediction model to be trained based on a second preset text set with various strength labels of various preset quality inspection elements by executing a second preset training process.

[0088] In the present embodiment, there is a preset quality inspection model, which includes a first recognition sub-model and a second recognition sub-model. Through the setting of the first recognition sub-model, the text data to be quality inspected can be accurately identified as to whether it has various preset quality inspection elements to obtain an initial recognition result. Through the setting of the second recognition sub-model, the initial recognition result corresponds to each intensity label of each preset quality inspection element of the initial text data to obtain the target recognition result. Since the recognition in the present embodiment adds the recognition of each intensity label of each preset quality inspection element, and only recognizes each intensity label of each preset quality inspection element in the initial text data, the recognition depth is increased and the data content to be identified is reduced. Therefore, the recognition accuracy is improved, and the recognition efficiency is improved.

[0089] In this embodiment, the first recognition sub-model is capable of identifying whether the text data to be inspected has various preset quality inspection elements to obtain an initial recognition result, wherein the first recognition sub-model is obtained by iteratively training the first preset prediction model to be trained based on the first preset text set with various preset quality inspection element labels by executing the first preset training process, so that the text data to be inspected can be accurately identified. Specifically, in the training process of executing the first preset training process, the first feature extraction is performed on the first preset text set with various preset quality inspection element labels through the first upsampling layer in the first preset prediction model to be trained to obtain the first extracted feature. After obtaining the first extracted feature, the first extracted feature is predicted through the activation layer in the first preset prediction model to be trained to obtain a first prediction result, and the first prediction result is compared with the first actual result pre-stored in the first preset text set to obtain the similarity between the first prediction result and the first actual result. The parameters of the first upsampling layer are continuously adjusted according to the similarity between the first prediction result and the first actual result until the similarity between the first prediction result and the first actual result is greater than the first preset recognition similarity, thereby obtaining the first recognition sub-model.

[0090] In this embodiment, initial text data corresponding to the initial recognition result is obtained, and the initial text data is input into the second recognition sub-model in the preset quality inspection model. The initial text data is identified by the second recognition sub-model with respect to the strength of each preset quality inspection element (determined by features such as speaking speed and intonation) to obtain the target recognition result; wherein, the second recognition sub-model is obtained by iteratively training a second preset prediction model to be trained based on a second preset text set having each strength label of each preset quality inspection element by executing a second preset training process. Specifically, during the training process of executing the second preset training flow, the second feature extraction is performed on the second preset text set having each preset quality inspection element strength label through the upsampling layer in the second preset prediction model to be trained to obtain the second extracted feature. After obtaining the second extracted feature, the second extracted feature is predicted through the activation layer in the second preset prediction model to be trained to obtain the second prediction result. The second prediction result is compared with the pre-stored second actual result in the second preset text set to obtain the similarity between the second prediction result and the second actual result. The parameters of the second upsampling layer are continuously adjusted according to the similarity between the second prediction result and the second actual result until the similarity between the second prediction result and the second actual result is greater than the second preset recognition similarity, thereby obtaining the second recognition sub-model.

[0091] In this embodiment, since the preset quality inspection model has been trained in advance, the target recognition result can be automatically obtained by directly inputting the text data to be inspected, thus laying the foundation for efficiently obtaining speech quality inspection evaluation.

[0092] Each of the preset quality inspection items is composed of various preset quality inspection elements;

[0093] Among them, refer to Figure 2 In this embodiment, another method for obtaining a target recognition result is provided, namely, method 2: the step of identifying whether the text data to be inspected has each preset quality inspection item to obtain a target recognition result includes:

[0094] Step S23, performing a matching process of a preset first pattern on each first text at each preset determined position in the text data to be inspected, and obtaining a first identification sub-result of whether each first text has a preset quality inspection item;

[0095] In the present embodiment, another method for obtaining the target recognition result is provided. In this method, first, each first text (the first text at the starting position or the first text at the ending position) at each preset determined position (such as the starting position or the ending position) in the text data to be quality inspected is matched with a preset first pattern to find the quality inspection item or the quality inspection element in the quality inspection item. For example, to find the quality inspection element "xx bank staff" in the opening remarks quality inspection item, the preset first pattern r"(*)(here)(*)(xx bank staff|xx bank)(*)(call you)(*)" can be used. It should be noted that the preset first pattern has a certain degree of generalization and can cover a variety of expressions. For example, it can cover sentences such as "WeBank calls you here" (which can be preset). In the present embodiment, it should be noted that the preset first pattern is first matched with each first text at the preset determined position, thereby realizing a quick initial quality inspection of the voice data to determine the quality inspection scoring level or improve the efficiency of the quality inspection (this is because if there is no corresponding quality inspection item in the initial position or the ending position quality inspection, it indicates that the current service level is poor).

[0096] Step S24, locating each second text associated with each preset quality inspection element from the text data to be quality inspected, inputting each second text into a preset classification model, obtaining a predicted quality inspection element label of each second text, and obtaining a second identification sub-result;

[0097] The second texts associated with the preset quality inspection elements are located from the text data to be inspected. Specifically, the second texts associated with the preset quality inspection elements are located from the text data to be inspected by word matching. For example, the quality inspection elements to be inspected (for example, "Please repay the full amount before noon tomorrow") are generally found from the text data to be inspected, that is, from a long text, and their start and end positions are determined. In the text data to be inspected, there are generally multiple sentences connected together without punctuation marks in the middle, for example, "Well, please repay the full amount before noon tomorrow, otherwise the credit report will be reported to the central bank." Therefore, it is necessary to find all the possible second texts where the quality inspection element is located by the preset fuzzy matching method. After obtaining each second text, the second text is input into the preset classification model to obtain the predicted quality inspection element label of each second text to obtain the second identification sub-result, that is, accurately classified by the preset classification model to obtain the second identification sub-result.

[0098] After obtaining the first recognition sub-result and the second recognition sub-result, the first recognition sub-result and the second recognition sub-result are comprehensively set as the target recognition result.

[0099] Step S30: evaluating the speech data to be quality-checked according to the target recognition result.

[0100] In this embodiment, after obtaining the target recognition result, the speech data to be inspected is evaluated according to the target recognition result. For example, if the target recognition result is that the text data to be inspected has all preset quality inspection items, the speech data to be inspected is full marks.

[0101] Each of the preset quality inspection items is composed of various preset quality inspection elements;

[0102] The step of evaluating the speech data to be quality-checked according to the target recognition result comprises:

[0103] Step S31, obtaining the score weight ratio of each preset quality inspection item, and obtaining the score of the speech data to be quality inspected according to the score weight ratio and the target recognition result;

[0104] In this embodiment, after obtaining the target recognition result, the preset quality inspection items in the target recognition result and the weight ratio of the preset quality inspection items are obtained based on the preset quality inspection items not included in the target recognition result and the score weight ratio of each preset quality inspection item, and based on the total score, the score of the speech data to be quality inspected is obtained.

[0105] Step S32: Based on the score of the speech data to be quality inspected, a score report is generated and the score report is submitted.

[0106] After obtaining the score of the voice data to be inspected, a score report is formed based on the score of the voice data to be inspected, and the score report is reported for subsequent query or random inspection. During the random inspection, a comparison is made based on the score in the score report and the manual score to determine whether the score in the score report is accurate.

[0107] The present application receives voice data to be inspected, performs text conversion processing on the voice data to be inspected, and obtains text data to be inspected; identifies whether the text data to be inspected has various preset quality inspection items to obtain a target recognition result; and evaluates the voice data to be inspected based on the target recognition result. In the present application, after receiving the voice data to be quality inspected, the voice data to be quality inspected is converted into text to obtain the text data to be quality inspected, and then the text data to be quality inspected is automatically identified to see whether it has various preset quality inspection items to obtain a target recognition result. Since the text data to be quality inspected is automatically identified to see whether it has various preset quality inspection items to obtain a target recognition result, and based on the target recognition result, the voice data to be quality inspected is automatically evaluated instead of requiring manual quality inspection, thereby saving resources and improving quality inspection efficiency. Moreover, since quality inspection is performed immediately after receiving the voice data to be quality inspected, the comprehensiveness of quality inspection is improved (that is, all voice data to be quality inspected are quality inspected), thereby solving the technical problems of low quality inspection efficiency and incomplete quality inspection in the prior art that voice quality inspection is performed manually.

[0108] Further, based on the first embodiment of the present application, in another embodiment of the present application, the step of setting the first identification sub-result and the second identification sub-result as the target identification result includes:

[0109] Step A1, taking the first text and the second text as candidate texts, performing matching processing of a preset second mode to determine whether there is a misjudgment of a preset quality inspection element, and obtaining a misjudgment recognition result;

[0110] In this embodiment, after obtaining the first identification sub-result and the second identification sub-result, in order to avoid possible misjudgment, the first text and the second text are also used as candidate texts, and the preset second mode matching processing is performed to determine whether there is a misjudgment of the preset quality inspection element, and obtain a misjudgment recognition result. Specifically, in the telephone collection recording, the customer service or the collection staff often express some negative statements such as "If the full amount is repaid before the evening, the credit report will not be reported to the central bank", and the negative statement does not belong to the preset "communication skills" quality inspection item, nor does it belong to the preset repayment plan (quality inspection element). Only through the matching of the preset first mode and text positioning, there may be recognition errors, so the candidate text needs to be re-predicted. Specifically, it is determined whether there is a preset negative statement in the candidate text. If there is a preset negative statement, it is determined that the candidate text does not belong to the preset "communication skills" quality inspection item, nor does it belong to the preset repayment plan quality inspection element.

[0111] Step A2: performing misjudgment elimination processing on the first recognition sub-result and the second recognition sub-result based on the misjudgment recognition result to obtain a target recognition result.

[0112] After obtaining the misjudgment recognition result, the first recognition sub-result and the second recognition sub-result are cleared of misjudgments based on the misjudgment recognition result, such as deleting the candidate texts that do not belong to the preset "communication skills" quality inspection item, to obtain the target recognition result.

[0113] In this embodiment, the first text and the second text are used as candidate texts, and a matching process of a preset second mode is performed to determine whether there is a misjudgment of a preset quality inspection element, and a misjudgment recognition result is obtained; based on the misjudgment recognition result, the first recognition sub-result and the second recognition sub-result are misjudged and cleared to obtain a target recognition result. In this embodiment, misjudgments are accurately identified, and the accuracy of recognition is improved.

[0114] Further, based on the first embodiment and the second embodiment of the present application, before the step of locating each second text associated with each preset quality inspection element from the text data to be inspected and inputting each second text into a preset classification model, the method includes:

[0115] Step B1, obtaining preset classified text data, extracting a feature matrix from the preset classified text data based on a preset convolution layer and a preset pooling layer in a third preset prediction model to be trained, and performing prediction processing on the feature matrix according to a preset activation layer of the third preset prediction model to be trained to obtain a predicted quality inspection factor label set;

[0116] In this embodiment, a preset classification model is accurately obtained to accurately obtain the predicted quality inspection element labels of each second text. Specifically, preset classification text data (including the actual label set of each quality inspection element) is obtained, and a feature matrix is ​​extracted from the training sub-data of the preset classification text data based on the preset convolution layer and the preset pooling layer in the third preset prediction model to be trained. The feature matrix is ​​predicted according to the preset activation layer of the third preset prediction model to be trained to obtain a predicted quality inspection element label set, wherein the predicted quality inspection element label set involves each quality inspection element label.

[0117] Step B2, comparing the predicted quality inspection element label set with the actual label set in the preset classified text data to obtain a comparison result, so as to adjust the parameters of the preset convolutional layer and the preset pooling layer until the similarity between the predicted quality inspection element label set and the actual label set is greater than the preset similarity threshold, so as to obtain the preset classification model.

[0118] The predicted quality inspection factor label set is compared with the actual label set in the preset classified text data to obtain a comparison result. After obtaining the comparison result, the parameters (including matrix weights) of the preset convolutional layer and the preset pooling layer are adjusted in order according to the comparison result until the similarity between the predicted quality inspection factor label set and the actual label set is greater than a preset similarity threshold, and a target classification model is obtained. When the similarity is greater than the preset similarity threshold, the target classification model is verified based on the verification sub-data of the preset classified text data. When the verification passes, the preset classification model is obtained.

[0119] This embodiment obtains preset classified text data, extracts a feature matrix from the preset classified text data based on the preset convolution layer and the preset pooling layer in the third preset prediction model to be trained, and performs prediction processing on the feature matrix according to the preset activation layer of the third preset prediction model to be trained to obtain a predicted quality inspection element label set; the predicted quality inspection element label set is compared with the actual label set in the preset classified text data to obtain a comparison result, so as to adjust the parameters of the preset convolution layer and the preset pooling layer until the similarity between the predicted quality inspection element label set and the actual label set is greater than the preset similarity threshold, so as to obtain the preset classification model. In this embodiment, the preset classification model is accurately obtained to lay the foundation for accurately conducting quality inspection of speech.

[0120] Reference Figure 3 , Figure 3 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0121] like Figure 3As shown, the speech quality inspection device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0122] Optionally, the voice quality inspection device may also include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. The rectangular user interface may include a display screen (Display), an input submodule such as a keyboard (Keyboard), and the optional rectangular user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0123] Those skilled in the art will understand that Figure 3 The structure of the speech quality inspection device shown in the figure does not constitute a limitation on the speech quality inspection device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0124] like Figure 3 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module and a voice quality inspection program. The operating system is a program that manages and controls the hardware and software resources of the voice quality inspection device, and supports the operation of the voice quality inspection program and other software and / or programs. The network communication module is used to realize the communication between the components inside the memory 1005, and the communication with other hardware and software in the voice quality inspection system.

[0125] exist Figure 3 In the voice quality inspection device shown, the processor 1001 is used to execute the voice quality inspection program stored in the memory 1005 to implement the steps of any of the above-mentioned voice quality inspection methods.

[0126] The specific implementation of the speech quality inspection device of the present application is basically the same as the embodiments of the above-mentioned speech quality inspection method, and will not be repeated here.

[0127] The present application also provides a speech quality inspection device, the speech quality inspection device comprising:

[0128] A receiving module, used for receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain text data to be inspected;

[0129] An identification module is used to identify whether the text data to be inspected has various preset quality inspection items to obtain a target identification result;

[0130] An evaluation module is used to evaluate the speech data to be quality-checked according to the target recognition result.

[0131] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0132] The identification module comprises:

[0133] An identification unit, used to identify whether the text data to be quality inspected has each preset quality inspection element through a first identification sub-model in a preset quality inspection model, and obtain an initial identification result;

[0134] A first acquisition unit is used to acquire initial text data corresponding to the initial recognition result, input the initial text data into a second recognition sub-model in the preset quality inspection model, and identify the strength of each preset quality inspection element of the initial text data through the second recognition sub-model to obtain the target recognition result;

[0135] Among them, the first recognition sub-model is obtained by iteratively training a first preset prediction model to be trained based on a first preset text set with various preset quality inspection element labels by executing a first preset training process, and the second recognition sub-model is obtained by iteratively training a second preset prediction model to be trained based on a second preset text set with various strength labels of various preset quality inspection elements by executing a second preset training process.

[0136] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0137] The identification module also includes:

[0138] A matching unit, used for performing a matching process of a preset first mode on each first text at each preset determined position in the text data to be inspected, and obtaining a first identification sub-result of whether each first text has a preset quality inspection item;

[0139] A positioning unit, used to locate each second text associated with each preset quality inspection element from the text data to be quality inspected, input each second text into a preset classification model, obtain a predicted quality inspection element label of each second text, and obtain a second identification sub-result;

[0140] A setting unit is configured to set the first recognition sub-result and the second recognition sub-result as the target recognition result.

[0141] Optionally, the setting unit includes:

[0142] A matching subunit, configured to use the first text and the second text as candidate texts, and perform matching processing of a preset second mode to determine whether there is a misjudgment of a preset quality inspection element, and obtain a misjudgment recognition result;

[0143] The clearing subunit is used to perform misjudgment clearing processing on the first recognition sub-result and the second recognition sub-result based on the misjudgment recognition result to obtain a target recognition result.

[0144] Optionally, the speech quality inspection device further includes:

[0145] An acquisition module is used to acquire preset classified text data, extract a feature matrix from the preset classified text data based on a preset convolution layer and a preset pooling layer in a third preset prediction model to be trained, and perform prediction processing on the feature matrix according to a preset activation layer of the third preset prediction model to be trained to obtain a predicted quality inspection factor label set;

[0146] A comparison module is used to compare the predicted quality inspection element label set with the actual label set in the preset classified text data to obtain a comparison result, so as to adjust the parameters of the preset convolution layer and the preset pooling layer until the similarity between the predicted quality inspection element label set and the actual label set is greater than a preset similarity threshold, so as to obtain the preset classification model.

[0147] Optionally, each of the preset quality inspection items is composed of various preset quality inspection elements;

[0148] The evaluation module includes:

[0149] A second acquisition unit is used to acquire the score weight ratio of each preset quality inspection item, and obtain the score of the speech data to be quality inspected according to the score weight ratio and the target recognition result;

[0150] The reporting module is used to form a scoring report based on the scoring of the speech data to be quality inspected, and report the scoring report.

[0151] Optionally, the receiving module includes:

[0152] A receiving unit, configured to receive voice data to be inspected, and perform text conversion processing on the voice data to be inspected to obtain first processed data;

[0153] The preprocessing unit is used to perform preset preprocessing on the first processed data to obtain text data to be quality inspected.

[0154] The specific implementation of the speech quality inspection device of the present application is basically the same as the embodiments of the above-mentioned speech quality inspection method, and will not be repeated here.

[0155] An embodiment of the present application provides a storage medium, and the storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to implement the steps of any of the above-mentioned speech quality inspection methods.

[0156] The specific implementation of the storage medium of the present application is basically the same as the embodiments of the above-mentioned voice quality inspection method, and will not be repeated here.

[0157] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.

Claims

1. A speech quality inspection method, characterized in that: The speech quality inspection method comprises: Receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain text data to be inspected; Identify whether the text data to be inspected has each preset quality inspection item to obtain a target recognition result; According to the target recognition result, the speech data to be inspected is evaluated; Each of the preset quality inspection items is composed of various preset quality inspection elements; The step of identifying whether the text data to be inspected has each preset quality inspection item to obtain a target recognition result includes: Performing matching processing of a preset first pattern on each first text at each preset determined position in the text data to be inspected, and obtaining a first identification sub-result of whether each first text has a preset quality inspection item; Locating each second text associated with each preset quality inspection element from the text data to be inspected, inputting each second text into a preset classification model, obtaining a predicted quality inspection element label of each second text, so as to obtain a second identification sub-result; wherein the positioning method includes a fuzzy matching method; The first recognition sub-result and the second recognition sub-result are set as the target recognition result.

2. The speech quality inspection method according to claim 1, characterized in that: The step of setting the first recognition sub-result and the second recognition sub-result as the target recognition result comprises: The first text and the second text are used as candidate texts, and a matching process of a preset second mode is performed to determine whether there is a misjudgment of a preset quality inspection element, and obtain a misjudgment recognition result; Based on the misjudgment recognition result, misjudgment elimination processing is performed on the first recognition sub-result and the second recognition sub-result to obtain a target recognition result.

3. The speech quality inspection method according to claim 2, characterized in that: Before the step of locating each second text associated with each preset quality inspection element from the text data to be inspected and inputting each second text into a preset classification model, the method includes: Acquire preset classified text data, extract a feature matrix from the preset classified text data based on a preset convolution layer and a preset pooling layer in a third preset prediction model to be trained, and perform prediction processing on the feature matrix according to a preset activation layer of the third preset prediction model to be trained to obtain a predicted quality inspection factor label set; The predicted quality inspection element label set is compared with the actual label set in the preset classified text data to obtain a comparison result, so as to adjust the parameters of the preset convolution layer and the preset pooling layer until the similarity between the predicted quality inspection element label set and the actual label set is greater than a preset similarity threshold, so as to obtain the preset classification model.

4. The voice quality inspection method according to any one of claims 1 to 3, characterized in that: Each of the preset quality inspection items is composed of various preset quality inspection elements; The step of evaluating the speech data to be quality-checked according to the target recognition result comprises: Obtaining the score weight ratio of each preset quality inspection item, and obtaining the score of the speech data to be quality inspected according to the score weight ratio and the target recognition result; Based on the scoring of the speech data to be quality inspected, a scoring report is formed and reported.

5. The speech quality inspection method according to claim 1, characterized in that: The step of receiving the voice data to be inspected, performing text conversion processing on the voice data to be inspected, and obtaining the text data to be inspected comprises: Receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain first processed data; The first processed data is subjected to a preset preprocessing to obtain text data to be quality inspected.

6. A speech quality inspection device, characterized in that: The speech quality inspection device comprises: A receiving module, used for receiving voice data to be inspected, and performing text conversion processing on the voice data to be inspected to obtain text data to be inspected; An identification module is used to identify whether the text data to be inspected has various preset quality inspection items to obtain a target identification result; An evaluation module, used for evaluating the speech data to be quality-checked according to the target recognition result; Each of the preset quality inspection items is composed of various preset quality inspection elements; The speech quality inspection device is used to achieve: Performing matching processing of a preset first pattern on each first text at each preset determined position in the text data to be inspected, and obtaining a first identification sub-result of whether each first text has a preset quality inspection item; Locating each second text associated with each preset quality inspection element from the text data to be inspected, inputting each second text into a preset classification model, obtaining a predicted quality inspection element label of each second text, so as to obtain a second identification sub-result; wherein the positioning method includes a fuzzy matching method; The first recognition sub-result and the second recognition sub-result are set as the target recognition result.

7. A voice quality inspection device, characterized in that: The speech quality inspection device comprises: a memory, a processor, and a program stored in the memory for implementing the speech quality inspection method. The memory is used to store a program for implementing the voice quality inspection method; The processor is used to execute a program for implementing the speech quality inspection method to implement the steps of the speech quality inspection method as claimed in any one of claims 1 to 5.

8. A storage medium, characterized in that: The storage medium stores a program for implementing the speech quality inspection method, and the program for implementing the speech quality inspection method is executed by a processor to implement the steps of the speech quality inspection method as claimed in any one of claims 1 to 5.

Images