Speech recognition method, device, computer equipment and storage medium

Abstract

The present application relates to a speech recognition method, apparatus, computer device, computer-readable storage medium, and computer program product. The method comprises: inputting multiple candidate segmentations of an initial recognition text corresponding to a speech to be recognized into a first model for recognition, obtaining multiple target segmentations corresponding to the multiple candidate segmentations; arranging the multiple target segmentations and the initial segmentations to obtain multiple intermediate texts; inputting the multiple intermediate texts into a second model to obtain logical scores corresponding to the multiple intermediate texts; and determining the target recognition text of the speech to be recognized based on the logical scores corresponding to the multiple intermediate texts. The speech recognition method provided in the present application can improve the recognition accuracy of speech in financial institutions, thereby improving the quality inspection effect of financial institutions on the service quality of customer service personnel.

Description

Technical Field

[0001] The present application relates to the field of speech processing technology, and in particular to a speech recognition method, apparatus, computer equipment, storage medium, and computer program product. Background Art

[0002] Financial institutions typically set up customer service centers, where dedicated customer service personnel provide voice services such as business consultation and transaction processing. To improve the quality of service provided by financial institutions and verify whether customer service personnel provide high-quality services in accordance with regulations and standards, they record all voice calls between customer service personnel and users. Financial institutions then use these recordings to conduct quality inspections of customer service personnel.

[0003] Then, in the process of quality inspection of customer service personnel through recording, the recording is usually converted into text form to provide the quality inspection department with a more intuitive quality inspection basis and facilitate the quality inspection operation of the quality inspection department.

[0004] However, the speech recognition models currently used by financial institutions are mostly general speech recognition models, which have low recognition accuracy for proprietary vocabulary in financial institutions, so that they cannot provide accurate text information for quality inspection departments, resulting in poor quality inspection results for customer service personnel. Summary of the Invention

[0005] The present application provides a speech recognition method, apparatus, computer device, computer-readable storage medium, and computer program product, which perform general and specialized recognition of speech to be recognized by financial institutions based on multiple models to improve the recognition accuracy of speech to be recognized by financial institutions.

[0006] In a first aspect, the present application provides a speech recognition method, the method comprising:

[0007] Inputting multiple candidate segmented words of the initial recognition text corresponding to the speech to be recognized into the first model for recognition, and obtaining multiple target segmented words corresponding to the multiple candidate segmented words, the first model being used to recognize proprietary vocabulary of the financial institution;

[0008] Arrange multiple target segmented words and initial segmented words to obtain multiple intermediate texts, where the initial segmented words are the segmented words other than the multiple candidate segmented words in the initial recognition text;

[0009] Inputting the plurality of intermediate texts into the second model to obtain logical scores corresponding to the plurality of intermediate texts;

[0010] The target recognition text of the speech to be recognized is determined based on the logical scores corresponding to the multiple intermediate texts.

[0011] In a second aspect, the present application further provides a speech recognition device, comprising:

[0012] a first input acquisition module, configured to input multiple candidate segmented words of the initial recognition text corresponding to the speech to be recognized into the first model for recognition, and obtain multiple target segmented words corresponding to the multiple candidate segmented words, wherein the first model is used to recognize proprietary vocabulary of financial institutions;

[0013] an arrangement module, configured to arrange the target segmented words and the initial segmented words to obtain a plurality of intermediate texts, wherein the initial segmented words are the segmented words other than the candidate segmented words in the initial recognition text;

[0014] A second input obtaining module is used to input a plurality of intermediate texts into the second model to obtain logical scores corresponding to the plurality of intermediate texts;

[0015] The determination module is used to determine the target recognition text of the speech to be recognized based on the logical scores corresponding to the multiple intermediate texts.

[0016] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any of the above methods when executing the computer program:

[0017] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above methods when the computer program is executed by a processor.

[0018] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, which implements the steps of any of the above methods when executed by a processor.

[0019] The present application provides a speech recognition method, apparatus, computer device, computer-readable storage medium, and computer program product. The method comprises: inputting multiple candidate segmentations of an initial recognition text corresponding to a speech to be recognized into a first model for recognition, obtaining multiple target segmentations corresponding to the multiple candidate segmentations; arranging the multiple target segmentations and the initial segmentations to obtain multiple intermediate texts; inputting the multiple intermediate texts into a second model to obtain logical scores corresponding to the multiple intermediate texts; and determining the target recognition text of the speech to be recognized based on the logical scores corresponding to the multiple intermediate texts. The speech recognition method provided in the present application, after recognizing the speech to be recognized of a financial institution using a general model, further recognizes proprietary vocabulary that cannot be recognized by the general model based on a first model that recognizes proprietary vocabulary of the financial institution, then arranges the text recognized by the general model and the text recognized by the first model, and then logically scores the obtained multiple intermediate texts using a second model for logical scoring, so as to select texts that are more consistent with logical word order from the multiple intermediate files through the logical scoring, thereby improving the recognition accuracy of the speech of the financial institution and thereby improving the quality inspection effect of the financial institution on the service quality of customer service personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A diagram of an application environment of a speech recognition method in one embodiment;

[0021] Figure 2 1 is a flow chart of a speech recognition method according to an embodiment;

[0022] Figure 3 is a flow chart of a speech recognition method according to another embodiment;

[0023] Figure 4 is a flow chart of a speech recognition method according to another embodiment;

[0024] Figure 5 is a flow chart of a speech recognition method according to another embodiment;

[0025] Figure 6 is a structural block diagram of a speech recognition method in one embodiment;

[0026] Figure 7 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0028] The speech recognition method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store data that server 104 needs to process. The data storage system can be integrated with server 104, or placed on a cloud or other network server. Terminal 102 collects the speech of all customer service personnel during customer service and sends it to server 104 via the network. Server 104 inputs multiple candidate segmentations of the initial recognition text corresponding to the speech to be recognized into a first model for recognition, obtaining multiple target segmentations corresponding to the multiple candidate segmentations. The multiple target segmentations and the initial segmentations are then arranged to obtain multiple intermediate texts. The multiple intermediate texts are then input into a second model to obtain logical scores corresponding to the multiple intermediate texts. Based on the logical scores corresponding to the multiple intermediate texts, target recognition texts for the speech to be recognized are determined, thereby improving the recognition accuracy of speech in financial institutions. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, scanners, voice recorders, and portable wearable devices. Portable wearable devices can be smart watches, smart glasses, smart bracelets, head-mounted devices, etc. Server 104 can be implemented as a standalone server or a server cluster consisting of multiple servers.

[0029] In one embodiment, Figure 2 As shown, a speech recognition method is provided, which is applied to Figure 1 The following steps are used as an example to illustrate the server in the example:

[0030] In step S202, multiple candidate word segments of the initial recognition text corresponding to the speech to be recognized are input into the first model for recognition to obtain multiple target word segments corresponding to the multiple candidate word segments. The first model is used to recognize proprietary vocabulary of financial institutions.

[0031] The speech to be recognized can be recorded from a customer service representative's phone call at a financial institution's customer service center, using a landline phone used by customer service personnel to provide customer service to users. The speech to be recognized includes questions and after-sales inquiries raised by users to customer service personnel, as well as responses from customer service personnel to these questions. To better serve users, financial institutions require pre-employment training for customer service personnel so that they can provide services that comply with the financial institution's regulations. Therefore, financial institutions need to use call records between customer service personnel and users to determine whether the personnel are complying with the financial institution's regulations. After receiving the speech stored in the customer service representative's phone, the server needs to convert the speech into text for more intuitive quality control of the customer service personnel. This requires translating the speech into text, and the accuracy of the text translation affects the accuracy of the customer service personnel quality control. Therefore, when the server inputs the speech to be recognized into a universal recognition model for recognition, the universal recognition model is more accurate in recognizing common words, such as "I need," "apply," "bank card," "credit card," "ID card," "name," "age," and "gender," which are commonly used by financial institutions, as well as words commonly found in real life. However, the recognition of some proprietary terms of various financial institutions is inaccurate, which cannot provide an accurate basis for customer service personnel's quality inspection.

[0032] Therefore, the server inputs the speech to be recognized into the universal recognition model for recognition. After obtaining multiple initial recognition texts, it filters out some proprietary words of financial institutions that are incorrectly recognized by the universal recognition model, and then recognizes the proprietary words through the first model to obtain the correct text expression corresponding to the proprietary words. Because the first model is obtained through training based on all the proprietary words already in financial institutions, the first model has a high recognition accuracy rate for the proprietary words of financial institutions. Through the recognition of the first model, the recognition accuracy of the speech to be recognized can be improved. For example, the text recognized by the universal recognition model is: I want to inquire about the benefits of Zhaozhaobao, and the candidate segmentation word is determined to be "Zhaozhaobao". The candidate segmentation word "Zhaozhaobao" is recognized by the first model to obtain the corresponding target segmentation word "Chaochaobao".

[0033] It should be noted that the first model can be another model independent of the general recognition model, or the first model and the general recognition model can be a single recognition model. In this case, as part of the general recognition model, after the server inputs the speech to be recognized into the general recognition model, the first model can recognize words with pronunciations similar to or identical to proprietary words, replace the proprietary words recognized in the general recognition model, and output multiple intermediate texts. This application is not limited to this.

[0034] Step S204 , arranging the multiple target segmented words and the initial segmented words to obtain multiple intermediate texts, where the initial segmented words are the other segmented words in the initial recognition text except the multiple candidate segmented words.

[0035] Among them, the server obtains multiple segmented words through the universal recognition model and the above-mentioned first model. Next, it is necessary to combine the multiple segmented words to obtain a complete text that is easy to read. The multiple segmented words can be combined in a variety of different ways, so that multiple intermediate texts can be obtained. For example, the multiple intermediate texts obtained by arranging are: I want the benefits of Chao Chao Bao consultation, Chao Chao Bao I want to consult the benefits, I want to consult Chao Chao Bao's benefits, I want the benefits of Chao Chao Bao consultation, etc. By performing multiple possible arrangements of the multiple segmented words to obtain multiple intermediate texts, it is easy to filter out the target text from the multiple intermediate texts, further improving the accuracy of the speech recognition to be recognized.

[0036] Step S206: Input the plurality of intermediate texts into the second model to obtain logical scores corresponding to the plurality of intermediate texts.

[0037] Among them, the second model is used to score the expression logic of the intermediate text to output the intermediate text that is most consistent with the logical expression. The second model can be obtained by training based on a large number of common general expression sentences, so that the judgment of the logic of the text is more accurate. For example, the above-mentioned multiple intermediate texts are input into the second model, and after recognition by the second model, the scores corresponding to the multiple intermediate texts are obtained as follows: I want to consult Chao Chao Bao for 5 points, I want to consult Chao Chao Bao for 3 points, I want to consult Chao Chao Bao for 8 points, and I want to consult Chao Chao Bao for 4 points. Through the logical scoring output by the second model, the server can better filter out the final recognition results for the speech to be recognized from multiple intermediate texts.

[0038] Step S208 : determining a target recognition text of the speech to be recognized based on the logical scores corresponding to the multiple intermediate texts.

[0039] The server may select the intermediate text with the highest logical score as the target text for speech recognition based on the logical scores of the multiple intermediate texts output by the second model. Since a higher logical score indicates that the expression of the text more closely conforms to standard logical expressions, the server may use the text with the highest logical score as the basis for performing service quality inspections on corresponding customer service personnel.

[0040] The present application provides a speech recognition method, the method comprising: inputting multiple candidate segmentations of an initial recognition text corresponding to a speech to be recognized into a first model for recognition, obtaining multiple target segmentations corresponding to the multiple candidate segmentations; arranging the multiple target segmentations and the initial segmentations to obtain multiple intermediate texts; inputting the multiple intermediate texts into a second model to obtain logical scores corresponding to the multiple intermediate texts; and determining a target recognition text for the speech to be recognized based on the logical scores corresponding to the multiple intermediate texts. The speech recognition method provided in the present application, after recognizing the speech to be recognized of a financial institution using a general model, further recognizes proprietary vocabulary that cannot be recognized in the general model based on a first model that recognizes proprietary vocabulary of the financial institution, then arranges the text recognized by the general model and the text recognized by the first model, and then logically scores the obtained multiple intermediate texts using a second model for logical scoring, so as to select texts that are more consistent with a logical word order from the multiple intermediate files through the logical scoring, thereby improving the recognition accuracy of the speech of the financial institution and thereby improving the quality inspection effect of the financial institution on the service quality of customer service personnel.

[0041] In one embodiment, this embodiment is an optional method embodiment for obtaining initial recognition text, and the method includes:

[0042] The speech to be recognized is input into the universal recognition model to obtain the initial recognition text. The universal recognition model is used to recognize universal word segmentation.

[0043] Among them, according to the above description, under normal circumstances, most of the conversations between customer service personnel and users are still common word segmentations and some common sentences. Therefore, after the server obtains the voice to be recognized, it is necessary to first input the voice to be recognized into the universal recognition model for universal word segmentation recognition to obtain the initial recognition text. If there is no proprietary vocabulary of the financial institution in the voice, then there is no need for the first model to participate in the recognition of the voice to be recognized.

[0044] The speech recognition method provided in the present application obtains multiple initial recognition texts by recognizing the speech to be recognized through a universal recognition model. The universal recognition model has a high recognition accuracy rate for universal word segmentations, and the proportion of general universal word segmentations in the speech to be recognized is relatively large, so the recognition efficiency of the speech to be recognized is high.

[0045] In one embodiment, Figure 3 As shown, this embodiment is an optional method embodiment for segmenting the initial text, and the method embodiment includes the following steps:

[0046] Step S302 : matching the initial recognition text with the dictionary library, and dividing the initial recognition text based on the matching result to obtain a plurality of initial segmented words.

[0047] The server may be equipped with a dictionary library, which may be generated based on common segmentation words and proprietary vocabulary of financial institutions. When the server recognizes the speech to be recognized based on the common recognition model, it obtains the initial recognition text, such as "I want to inquire about the benefits of the recruitment package." Then, the server matches the text with the dictionary library to obtain multiple initial segmentation words of the initial recognition text, thereby achieving the division of the overall text and facilitating the subsequent determination of candidate segmentation words.

[0048] Step S304 : annotating the speech to be recognized based on the speech features of the speech to be recognized to obtain a plurality of annotated texts.

[0049] Among them, the multiple initial segmentations obtained based on the above-mentioned division of the initial recognition text can be achieved by manually annotating the correct text of each initial segmentation based on the speech to be recognized; or the server can make possible text expressions for the multiple initial segmentations based on the speech features of the speech to be recognized to obtain multiple annotated texts. For example, the labeling of I want, consult, Zhaozhaobao, of, and income can be obtained as: I want, consult, Chaochaobao, of, and income.

[0050] Step S306 : comparing the multiple annotated texts with the multiple initial segmented words, and determining multiple candidate segmented words from the multiple initial segmented words according to the comparison results, wherein the similarity between the candidate segmented words and the annotated texts is less than a similarity threshold.

[0051] Among them, based on the annotated text corresponding to the multiple initial participles obtained by annotating the multiple initial participles mentioned above, the server can compare the annotated text with the multiple initial participles identified by the general recognition model, calculate the similarity between the two, and determine the candidate participles from the multiple initial participles based on the calculation result of the similarity, for example, the candidate participles obtained by screening are the "zhaozhaobao".

[0052] The speech recognition method provided in this application segments the initial recognition text through a dictionary library and divides the initial recognition text into word strengths, which facilitates the subsequent screening of candidate segmentations from multiple initial segmentations, improves the efficiency of screening candidate segmentations, and further improves the recognition efficiency of the speech to be recognized.

[0053] In one embodiment, Figure 4 As shown, this embodiment is an optional method embodiment for training the first model, and the method embodiment includes the following steps:

[0054] Step S402: obtaining a plurality of proprietary terms of the financial institution, and setting a plurality of model weights for each of the plurality of proprietary terms;

[0055] Step S404: performing model training based on the plurality of proprietary words and the plurality of model weights corresponding to each of the plurality of proprietary words to obtain target model weights corresponding to the plurality of proprietary words;

[0056] Step S406: Obtain a first model based on the plurality of proprietary words and target model weights corresponding to the plurality of proprietary words.

[0057] Among them, multiple proprietary words can be obtained from the server of the financial institution. The principle of obtaining the first model in this application is: each proprietary word will correspond to multiple different text expressions, and the corresponding pronunciations of the multiple text expressions are similar or identical, that is, multiple words with the same pronunciation but different characters. By continuously improving the weight of the correct proprietary word among the multiple words with the same pronunciation but different characters, the weight of all proprietary words in the financial institution in the recognition model is increased. Then, if a text with the same pronunciation but different text as the proprietary word in the financial institution is input for recognition, since the weight of the proprietary word in the financial institution is greater than the other multiple words with the same pronunciation but different characters, the correct text expression of the proprietary word of the financial institution can be output to achieve text correction of the proprietary word. Of course, when training through the neural network model, the weight of the correct text expression of the proprietary word in the financial institution cannot be adjusted appropriately at one time, so it is necessary to continuously adjust the weight of the proprietary word based on the results of each training output to obtain the target model weight corresponding to the proprietary word. It should be noted that since the similar texts, number of similar texts, degree of similarity, etc. of each proprietary word are different, the weight of each proprietary word can be different. In this way, the weight of each proprietary word in the financial institution can be accurately set to improve the recognition accuracy of the first model.

[0058] In one embodiment, Figure 5 As shown, this embodiment is an optional method embodiment for training the second model, and the method embodiment includes the following steps:

[0059] Step 502: Acquire multiple historical industry texts and multiple basic texts of a financial institution;

[0060] Step 504: pre-processing the multiple historical industry texts and the multiple basic texts;

[0061] Step 506: Perform model training based on the pre-processed multiple historical industry texts and multiple basic texts to obtain a second model.

[0062] The second model is used to perform logical scoring on text. Therefore, it needs to be trained on a large amount of text with correct logical expressions. Because the text in this application includes a large amount of proprietary vocabulary from financial institutions, it requires not only basic text, such as: "I want to apply for a card, I want to consult, I want to ask how to activate it, I want to ask how to operate it, I want to ask how to handle the service, I need guidance, can you send me guidance, can you send me a text tutorial, can you help me cancel it?" It also requires historical industry text from financial institutions, such as: "Chao Chao Bao's returns, Chao Chao Bao's holding period, Shanghai Pudong Development Bank credit card, Shanghai Pudong Development Bank debit card, Shanghai Pudong Development Fund, Shanghai Pudong Development Gold, Shanghai Pudong Development Stock, Shanghai Pudong Development Wealth Management, Shanghai Pudong Development Ruyi Fund, Shanghai Pudong Development Pengfa Wealth Management Products, etc." The neural network model is trained using this historical industry text and basic text to obtain the second model.

[0063] After obtaining multiple historical industry texts and multiple basic texts, the server can combine the multiple historical industry texts and multiple basic texts to obtain an integrated text containing quasi-vocabulary of financial institutions, and obtain a second model based on the integrated text training.

[0064] The speech recognition method provided in this application obtains a second model through historical industry texts and basic text training of financial institutions, so as to adapt the second model training to the needs of text logical expression scoring in financial institutions, and further obtain text that is more compatible with the speech to be recognized through the second model.

[0065] It should be noted that the training process of the first model and the second model can be performed in the terminal or in the server. If the model training process is performed in the terminal, the server can download the trained first model and second model from the terminal to perform proprietary vocabulary recognition and logical scoring of intermediate text. This application is not limited to this.

[0066] Optionally, the server may determine the intermediate text with the highest logical score output from the second model as the target recognition text for the speech to be recognized. Because a higher logical score indicates that the text more closely conforms to the logical expression of the correct text, the server can perform quality inspection of the customer service personnel's customer service based on the intermediate text with the highest logical score, thereby improving the accuracy of the quality inspection of the customer service personnel's customer service.

[0067] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0068] Based on the same inventive concept, the present application also provides a speech recognition device for implementing the aforementioned speech recognition method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations in the one or more speech recognition device embodiments provided below can be found in the above-mentioned limitations on the speech recognition method and will not be further elaborated here.

[0069] In one embodiment, Figure 6 As shown, a speech recognition device is provided, comprising: a first input obtaining module 602, an arrangement module 604, a second input obtaining module 606 and a determination module 608, wherein:

[0070] A first input acquisition module 602 is configured to input multiple candidate segmented words of the initial recognition text corresponding to the speech to be recognized into a first model for recognition, and obtain multiple target segmented words corresponding to the multiple candidate segmented words. The first model is configured to recognize proprietary vocabulary of financial institutions.

[0071] an arrangement module 604 for arranging the plurality of target segmented words and the initial segmented words to obtain a plurality of intermediate texts, wherein the initial segmented words are the segmented words other than the plurality of candidate segmented words in the initial recognition text;

[0072] A second input obtaining module 606 is configured to input a plurality of intermediate texts into a second model to obtain logical scores corresponding to the plurality of intermediate texts;

[0073] The determination module 608 is configured to determine a target recognition text of the speech to be recognized based on the logical scores corresponding to the plurality of intermediate texts.

[0074] In one embodiment, the first input obtaining module 602 is further configured to input the speech to be recognized into a universal recognition model to obtain the initial recognition text, and the universal recognition model is configured to recognize universal word segmentation.

[0075] In one embodiment, the first input obtaining module 602 further includes a screening unit.

[0076] The screening unit is used to match the initial recognition text with the dictionary library, divide the initial recognition text based on the matching results to obtain multiple initial segmentations; annotate the speech to be recognized based on the speech features of the speech to be recognized to obtain multiple annotated texts; compare the multiple annotated texts with the multiple initial segmentations, and determine multiple candidate segmentations from the multiple initial segmentations based on the comparison results, where the similarity between the candidate segmentations and the annotated texts is less than a similarity threshold.

[0077] In one embodiment, the above device further comprises a training module,

[0078] A training module is used to obtain multiple proprietary words of a financial institution and set multiple model weights for each of the multiple proprietary words; perform model training based on the multiple proprietary words and the multiple model weights corresponding to each of the multiple proprietary words to obtain target model weights corresponding to the multiple proprietary words; and obtain a first model based on the multiple proprietary words and the target model weights corresponding to the multiple proprietary words.

[0079] In one embodiment, the above-mentioned training module is also used to obtain multiple historical industry texts and multiple basic texts of financial institutions; preprocess the multiple historical industry texts and multiple basic texts; perform model training based on the preprocessed multiple historical industry texts and multiple basic texts to obtain a second model.

[0080] In one embodiment, the determination module 608 is specifically configured to determine the intermediate text with the highest logical score among the multiple intermediate texts as the target recognition text for the speech to be recognized.

[0081] In one embodiment, the above device further includes a quality inspection module.

[0082] The quality inspection module is used to perform quality inspection on customer service based on the target recognition text of the speech to be recognized.

[0083] Each module in the above-mentioned speech recognition device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0084] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store coordinate information and text information data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a speech recognition method is implemented.

[0085] Those skilled in the art will understand that Figure 7 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 shown in the figure, or combine certain components, or have a different component arrangement.

[0086] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0087] Inputting multiple candidate segmented words of the initial recognition text corresponding to the speech to be recognized into the first model for recognition, and obtaining multiple target segmented words corresponding to the multiple candidate segmented words, the first model being used to recognize proprietary vocabulary of the financial institution;

[0088] Arrange multiple target segmented words and initial segmented words to obtain multiple intermediate texts, where the initial segmented words are the segmented words other than the multiple candidate segmented words in the initial recognition text;

[0089] Inputting the plurality of intermediate texts into the second model to obtain logical scores corresponding to the plurality of intermediate texts;

[0090] The target recognition text of the speech to be recognized is determined based on the logical scores corresponding to the multiple intermediate texts.

[0091] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0092] The speech to be recognized is input into a universal recognition model to obtain the initial recognition text, and the universal recognition model is used to recognize universal word segmentation.

[0093] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0094] The initial recognition text is matched with the dictionary library, and the initial recognition text is divided based on the matching results to obtain multiple initial segmentations; the speech to be recognized is annotated based on the speech features of the speech to be recognized to obtain multiple annotated texts; the multiple annotated texts are compared with the multiple initial segmentations, and multiple candidate segmentations are determined from the multiple initial segmentations based on the comparison results, and the similarity between the candidate segmentations and the annotated texts is less than a similarity threshold.

[0095] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0096] Acquire multiple proprietary words of a financial institution and set multiple model weights for each of the multiple proprietary words; perform model training based on the multiple proprietary words and the multiple model weights corresponding to each of the multiple proprietary words to obtain target model weights corresponding to the multiple proprietary words; obtain a first model based on the multiple proprietary words and the target model weights corresponding to the multiple proprietary words.

[0097] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0098] Acquire multiple historical industry texts and multiple basic texts of financial institutions; preprocess the multiple historical industry texts and multiple basic texts; perform model training based on the preprocessed multiple historical industry texts and multiple basic texts to obtain a second model.

[0099] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0100] The intermediate text with the highest logical score among the multiple intermediate texts is determined as the target recognition text of the speech to be recognized.

[0101] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0102] Perform quality control on customer service based on the target recognition text of the speech to be recognized.

[0103] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0104] Inputting multiple candidate segmented words of the initial recognition text corresponding to the speech to be recognized into the first model for recognition, and obtaining multiple target segmented words corresponding to the multiple candidate segmented words, the first model being used to recognize proprietary vocabulary of the financial institution;

[0105] Arrange multiple target segmented words and initial segmented words to obtain multiple intermediate texts, where the initial segmented words are the segmented words other than the multiple candidate segmented words in the initial recognition text;

[0106] Inputting the plurality of intermediate texts into the second model to obtain logical scores corresponding to the plurality of intermediate texts;

[0107] The target recognition text of the speech to be recognized is determined based on the logical scores corresponding to the multiple intermediate texts.

[0108] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0109] The speech to be recognized is input into a universal recognition model to obtain the initial recognition text, and the universal recognition model is used to recognize universal word segmentation.

[0110] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0111] The initial recognition text is matched with the dictionary library, and the initial recognition text is divided based on the matching results to obtain multiple initial segmentations; the speech to be recognized is annotated based on the speech features of the speech to be recognized to obtain multiple annotated texts; the multiple annotated texts are compared with the multiple initial segmentations, and multiple candidate segmentations are determined from the multiple initial segmentations based on the comparison results, and the similarity between the candidate segmentations and the annotated texts is less than a similarity threshold.

[0112] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0113] Acquire multiple proprietary words of a financial institution and set multiple model weights for each of the multiple proprietary words; perform model training based on the multiple proprietary words and the multiple model weights corresponding to each of the multiple proprietary words to obtain target model weights corresponding to the multiple proprietary words; obtain a first model based on the multiple proprietary words and the target model weights corresponding to the multiple proprietary words.

[0114] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0115] Acquire multiple historical industry texts and multiple basic texts of financial institutions; preprocess the multiple historical industry texts and multiple basic texts; perform model training based on the preprocessed multiple historical industry texts and multiple basic texts to obtain a second model.

[0116] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0117] The intermediate text with the highest logical score among the multiple intermediate texts is determined as the target recognition text of the speech to be recognized.

[0118] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0119] Perform quality control on customer service based on the target recognition text of the speech to be recognized.

[0120] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0121] Inputting multiple candidate segmented words of the initial recognition text corresponding to the speech to be recognized into the first model for recognition, and obtaining multiple target segmented words corresponding to the multiple candidate segmented words, the first model being used to recognize proprietary vocabulary of the financial institution;

[0122] Arrange multiple target segmented words and initial segmented words to obtain multiple intermediate texts, where the initial segmented words are the segmented words other than the multiple candidate segmented words in the initial recognition text;

[0123] Inputting the plurality of intermediate texts into the second model to obtain logical scores corresponding to the plurality of intermediate texts;

[0124] The target recognition text of the speech to be recognized is determined based on the logical scores corresponding to the multiple intermediate texts.

[0125] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0126] The speech to be recognized is input into a universal recognition model to obtain the initial recognition text, and the universal recognition model is used to recognize universal word segmentation.

[0127] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0128] The initial recognition text is matched with the dictionary library, and the initial recognition text is divided based on the matching results to obtain multiple initial segmentations; the speech to be recognized is annotated based on the speech features of the speech to be recognized to obtain multiple annotated texts; the multiple annotated texts are compared with the multiple initial segmentations, and multiple candidate segmentations are determined from the multiple initial segmentations based on the comparison results, and the similarity between the candidate segmentations and the annotated texts is less than a similarity threshold.

[0129] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0130] Acquire multiple proprietary words of a financial institution and set multiple model weights for each of the multiple proprietary words; perform model training based on the multiple proprietary words and the multiple model weights corresponding to each of the multiple proprietary words to obtain target model weights corresponding to the multiple proprietary words; obtain a first model based on the multiple proprietary words and the target model weights corresponding to the multiple proprietary words.

[0131] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0132] Acquire multiple historical industry texts and multiple basic texts of financial institutions; preprocess the multiple historical industry texts and multiple basic texts; perform model training based on the preprocessed multiple historical industry texts and multiple basic texts to obtain a second model.

[0133] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0134] The intermediate text with the highest logical score among the multiple intermediate texts is determined as the target recognition text of the speech to be recognized.

[0135] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0136] Perform quality control on customer service based on the target recognition text of the speech to be recognized.

[0137] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0138] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, and the like.

[0139] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0140] The above embodiments merely illustrate several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A speech recognition method, characterized in that: The method comprises: Inputting multiple candidate word segments of an initial recognition text corresponding to the speech to be recognized into a first model for recognition, and obtaining multiple target word segments corresponding to the multiple candidate word segments, wherein the candidate word segments are words in the initial text that have the same or similar pronunciation as the proprietary words, and the first model is used to recognize the proprietary words of the financial institution; the training process of the first model includes: obtaining multiple proprietary words of the financial institution, and obtaining multiple other words corresponding to each proprietary word that have the same pronunciation but different characters, and setting multiple model weights for the multiple proprietary words and the multiple other words corresponding to each proprietary word; performing model training based on the multiple proprietary words, the multiple other words corresponding to each proprietary word, and the multiple model weights to obtain target model weights; and obtaining the first model based on the multiple proprietary words and the target model weights; Arranging the plurality of target segmented words and initial segmented words to obtain a plurality of intermediate texts, wherein the initial segmented words are segmented words other than the plurality of candidate segmented words in the initial recognition text; Inputting the plurality of intermediate texts into a second model to obtain logical scores corresponding to the plurality of intermediate texts; The target recognition text of the speech to be recognized is determined based on the logical scores corresponding to the multiple intermediate texts.

2. The method according to claim 1, characterized in that The method further comprises: The speech to be recognized is input into a universal recognition model to obtain the initial recognition text, and the universal recognition model is used to recognize universal word segmentation.

3. The method according to claim 2, characterized in that The method further comprises: Matching the initial recognition text with a dictionary library, and dividing the initial recognition text based on the matching result to obtain a plurality of initial segmented words; Annotating the speech to be recognized based on the speech features of the speech to be recognized to obtain multiple annotated texts; The multiple annotated texts are compared with the multiple initial participles, and the multiple candidate participles are determined from the multiple initial participles according to the comparison results, wherein the similarity between the candidate participles and the annotated texts is less than a similarity threshold.

4. The method according to claim 1, wherein The training process of the second model includes: Obtaining multiple historical industry texts and multiple basic texts of the financial institution; Preprocessing the multiple historical industry texts and the multiple basic texts; Model training is performed based on the preprocessed multiple historical industry texts and the multiple basic texts to obtain the second model.

5. The method according to claim 1, characterized in that The determining the target recognition text of the speech to be recognized based on the logical scores corresponding to the multiple intermediate texts includes: The intermediate text with the highest logical score among the multiple intermediate texts is determined as the target recognition text of the speech to be recognized.

6. The method according to claim 5, characterized in that The method further comprises: Perform quality control on customer service based on the target recognition text of the speech to be recognized.

7. A speech recognition device, characterized in that: The device comprises: A first input acquisition module is configured to input multiple candidate word segments of an initial recognition text corresponding to a speech to be recognized into a first model for recognition, and obtain multiple target word segments corresponding to the multiple candidate word segments, wherein the candidate word segments are words in the initial text that have the same or similar pronunciation as proprietary words, and the first model is configured to recognize proprietary words of financial institutions; the training process of the first model includes: obtaining multiple proprietary words of the financial institution, and obtaining multiple other words corresponding to each proprietary word that have the same pronunciation but different characters, and setting multiple model weights for the multiple proprietary words and the multiple other words corresponding to each proprietary word; performing model training based on the multiple proprietary words, the multiple other words corresponding to each proprietary word, and the multiple model weights to obtain target model weights; and obtaining the first model based on the multiple proprietary words and the target model weights. an arranging module, configured to arrange the plurality of target segmented words and initial segmented words to obtain a plurality of intermediate texts, wherein the initial segmented words are segmented words other than the plurality of candidate segmented words in the initial recognition text; A second input obtaining module, configured to input the plurality of intermediate texts into a second model to obtain logical scores corresponding to the plurality of intermediate texts; A determination module is used to determine the target recognition text of the speech to be recognized based on the logical scores corresponding to the multiple intermediate texts.

8. The device according to claim 7, characterized in that The first input obtaining module is further configured to: The speech to be recognized is input into a universal recognition model to obtain the initial recognition text, and the universal recognition model is used to recognize universal word segmentation.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images