A pronunciation recognition method, device, apparatus and storage medium thereof

Abstract

The embodiment of the application belongs to the technical field of artificial intelligence and financial technology, is applied to the process of voice customer service pronunciation accuracy identification in financial business, and relates to a pronunciation identification method, device and equipment and a storage medium thereof, which comprises optimizing GOP algorithm according to similar phonemes, so as to improve the pronunciation identification accuracy. In the overall scheme, the traditional GOP scoring algorithm is used for preliminary screening, the inaccurate phonemes are quickly excluded by setting the pronunciation score threshold, the prior knowledge of the similar phoneme table is introduced into the GOP algorithm for the second time, the interference of the similar phonemes on the final GOP score is reduced, the application range of the GOP algorithm is expanded, the discrimination ability of the algorithm for the similar phonemes is improved, the overall algorithm process is clear and simple, easy to apply, and the pronunciation identification accuracy is further improved.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and financial technology, and is applied to the process of recognizing the accuracy of voice customer service pronunciation in financial business. In particular, it relates to a pronunciation recognition method, device, equipment and its storage medium. Background Technology

[0002] With the development of the computer industry, traditional financial business is gradually transforming into financial technology business. In particular, in the field of telemarketing customer service in financial insurance and banking, customer service agents still undertake more and more financial business-related inquiries, sales and follow-up business.

[0003] Especially in the training of telemarketing personnel, oral assessment is basically done manually. For example, taking some professional oral assessment exams is very costly in terms of both time and money. Moreover, due to the limitations of manual assessment, professionals can only give a broad evaluation of the oral assessment results. Learners cannot know their specific pronunciation problems through such evaluation, which is not helpful for improving their pronunciation level in the future. The pronunciation correction model designed by the traditional GOP algorithm judges the audio to be tested. Due to the existence of similar pronunciations and interference, the pronunciation recognition results are prone to being inaccurate. Summary of the Invention

[0004] The purpose of this application is to provide a pronunciation recognition method, apparatus, device and storage medium to solve the problem that the pronunciation recognition results are not accurate enough when the pronunciation correction model designed using the traditional GOP algorithm is used to judge the audio to be tested.

[0005] To address the aforementioned technical problems, this application provides a pronunciation recognition method, employing the following technical solution:

[0006] A pronunciation recognition method includes the following steps:

[0007] Obtain a pre-organized similar phoneme table, collect standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio from a preset standard pronunciation library, wherein the similar phoneme table contains all target phonemes with similar phonemes, and the similar phonemes corresponding to each target phoneme;

[0008] Based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio, obtain the pronunciation score value of each phoneme in the standard pronunciation audio;

[0009] Calculate the pronunciation score threshold based on the pronunciation score value of each phoneme in the standard pronunciation audio;

[0010] Obtain the audio of the pronunciation to be tested;

[0011] The pronunciation of each phoneme in the audio to be tested is detected and scored to obtain the detection score value of each phoneme in the audio to be tested.

[0012] The accuracy of pronunciation of each phoneme in the audio to be tested is identified by using preset detection rules, detection scores of each phoneme in the audio to be tested, and the pronunciation score threshold.

[0013] Furthermore, before performing the step of obtaining the pronunciation score value of each phoneme in the standard pronunciation audio based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio, the method further includes:

[0014] Perform pre-emphasis, framing, and windowing operations on the standard pronunciation audio in the standard pronunciation library to obtain the standard audio preprocessing results;

[0015] Based on the preset acoustic feature extraction component and the standard audio preprocessing results, the MFCC acoustic features of the standard pronunciation audio are obtained;

[0016] Based on the MFCC acoustic features of the standard pronunciation audio and the preset voiceprint recognition model, the logarithmic phoneme posterior probability of each phoneme in the standard pronunciation audio is obtained, wherein the preset voiceprint recognition model includes a Kaldi-based voiceprint recognition model.

[0017] The step of obtaining the pronunciation score value of each phoneme in the standard pronunciation audio based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio specifically includes:

[0018] The maximum value is selected from the logarithmic posterior probabilities of each phoneme in the standard pronunciation audio.

[0019] The maximum logarithmic phoneme posterior probability is pre-set to the sum of the logarithmic phoneme posterior probabilities of all phonemes in the phoneme annotation set;

[0020] Iterate through all phonemes in the phoneme annotation set, and take different phonemes in the phoneme annotation set as target phonemes in turn;

[0021] According to the preset first algorithm formula: GOP(p1)=PP(p1)-∑ q1∈QLPP(q1) is used to obtain the pronunciation score of each phoneme in the standard pronunciation audio, where p1 represents the target phoneme, GOP(p1) represents the pronunciation score of the target phoneme p1, LPP(p1) represents the logarithmic phoneme posterior probability of the target phoneme p1, Q represents all phonemes in the phoneme annotation set, q1 represents the phoneme with the highest pronunciation score in the phoneme annotation set, and LPP(q1) represents the logarithmic phoneme posterior probability of phoneme q1, which is the sum of the logarithmic phoneme posterior probabilities of all phonemes in the phoneme annotation set.

[0022] Furthermore, the step of calculating the pronunciation score threshold based on the pronunciation score values ​​of each phoneme in the standard pronunciation audio specifically includes:

[0023] Based on the pronunciation score and average value algorithm formula of each phoneme in the standard pronunciation audio, the average pronunciation score of all phonemes in the phoneme annotation set is calculated;

[0024] Based on the pronunciation score of each phoneme in the standard pronunciation audio, the mean and standard deviation algorithm formula of the pronunciation scores, the standard deviation of the pronunciation scores of all phonemes in the phoneme annotation set is calculated;

[0025] The pronunciation score threshold is obtained by using the preset threshold calculation formula: T=μ+α*σ+β, where T represents the pronunciation score threshold, μ represents the mean pronunciation score of all phonemes in the phoneme annotation set, σ represents the standard deviation of the pronunciation scores of all phonemes in the phoneme annotation set, and α and β are preset empirical constants.

[0026] Furthermore, after performing the step of acquiring the audio of the pronunciation to be tested, the method further includes:

[0027] The audio to be tested is subjected to pre-emphasis, framing, and windowing operations to obtain the preprocessing result of the audio to be tested;

[0028] Based on the acoustic feature extraction component and the preprocessing result of the audio to be tested, the MFCC acoustic features of the audio to be tested are obtained, wherein the parameters set when performing pre-emphasis, framing and windowing operations on the audio to be tested are consistent with the parameters set when performing pre-emphasis, framing and windowing operations on the standard audio in the standard pronunciation library;

[0029] Based on the MFCC acoustic features of the audio to be tested and the Kaldi-based voiceprint recognition model, the logarithmic posterior probability of each phoneme in the audio to be tested is obtained.

[0030] Furthermore, the step of performing pronunciation detection and scoring on each phoneme in the audio to be tested, and obtaining the detection score value of each phoneme in the audio to be tested, specifically includes:

[0031] Traverse all phonemes in the audio of the pronunciation to be tested, and take the different phonemes in the audio of the pronunciation to be tested as the phonemes to be tested in turn;

[0032] According to the preset second algorithm formula: GOP(p2)=PP(p2)-∑ q1∈Q LPP(q1) performs pronunciation detection and scoring on each phoneme in the audio to be tested, and obtains the detection score value of each phoneme in the audio to be tested. Here, p2 represents the phoneme to be tested, GOP(p2) represents the detection score value of the phoneme p2, LPP(p2) represents the logarithmic phoneme posterior probability of the phoneme p2, Q represents all phonemes in the phoneme annotation set, q1 represents the phoneme with the highest pronunciation score in the phoneme annotation set, and LPP(q1) represents the logarithmic phoneme posterior probability of phoneme q1, which is the sum of the logarithmic phoneme posterior probabilities of all phonemes in the phoneme annotation set.

[0033] The detection scores of each phoneme in the audio to be tested are organized into a set of detection scores.

[0034] Furthermore, the step of identifying the pronunciation accuracy of each phoneme in the audio to be tested through preset detection rules, the detection score value of each phoneme in the audio to be tested, and the pronunciation score threshold specifically includes:

[0035] Sequentially obtain the elements in the set of detection scores;

[0036] Each element in the set of detection scores obtained sequentially is compared with the pronunciation score threshold to obtain the comparison result;

[0037] Based on the comparison results, determine the relationship between each element in the set of detection score values ​​and the pronunciation score threshold;

[0038] If the current element is not greater than the pronunciation score threshold, then the phoneme corresponding to the current element is identified as being inaccurately pronounced;

[0039] If the current element is greater than the pronunciation score threshold, then according to the similar phoneme table, the detection score value of the current phoneme is re-acquired, and the re-acquired detection score value of the current phoneme is compared with the pronunciation score threshold to obtain the comparison result. If the current element is not greater than the pronunciation score threshold, then the pronunciation of the phoneme corresponding to the current element is identified as inaccurate; if the current element is greater than the pronunciation score threshold, then the pronunciation of the phoneme corresponding to the current element is identified as accurate.

[0040] Furthermore, the step of re-obtaining the detection score value of the current phoneme based on the similar phoneme table specifically includes:

[0041] Based on the similar phoneme table, all similar phonemes corresponding to the current phoneme are identified, wherein each similar phoneme in the similar phoneme table has a pre-set corresponding logarithmic phoneme posterior probability;

[0042] Obtain the logarithmic phoneme posterior probability of all similar phonemes corresponding to the current phoneme, and filter out the maximum value, setting it as the sum of the logarithmic phoneme posterior probabilities of all similar phonemes;

[0043] According to the preset third algorithm formula: Calculate the detection score of the current phoneme, where p3 represents the current phoneme, GOP(p3) represents the detection score of the current phoneme p3, LPP(p3) represents the logarithmic posterior probability of the current phoneme p3, and Q... x q1 represents the set of all similar phonemes corresponding to the current phoneme, and q2 represents the set of similar phonemes Q. x The phoneme with the highest pronunciation score is represented by LPP(q2), which is the logarithmic phoneme posterior probability of phoneme q2, i.e., the sum of the logarithmic phoneme posterior probabilities of all similar phonemes.

[0044] To address the aforementioned technical problems, this application also provides a pronunciation recognition device, which employs the following technical solution:

[0045] A pronunciation recognition device, comprising:

[0046] The first data acquisition module is used to acquire a pre-organized similar phoneme table, collect standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio from a preset standard pronunciation library, wherein the similar phoneme table contains all target phonemes with similar phonemes, and the similar phonemes corresponding to each target phoneme;

[0047] The standard score calculation module is used to obtain the pronunciation score value of each phoneme in the standard pronunciation audio based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio.

[0048] The scoring threshold calculation module is used to calculate the pronunciation score threshold based on the pronunciation score value of each phoneme in the standard pronunciation audio.

[0049] The second data acquisition module is used to acquire the audio of the pronunciation to be tested;

[0050] The detection score calculation module is used to perform pronunciation detection and scoring on each phoneme in the audio to be tested, and to obtain the detection score value of each phoneme in the audio to be tested.

[0051] The accuracy recognition module is used to identify the pronunciation accuracy of each phoneme in the audio to be tested by using preset detection rules, the detection score value of each phoneme in the audio to be tested, and the pronunciation score threshold.

[0052] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution:

[0053] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the pronunciation recognition method described above.

[0054] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below:

[0055] A computer-readable storage medium storing computer-readable instructions that, when executed by a processor, implement the steps of the pronunciation recognition method described above.

[0056] Compared with the prior art, the embodiments of this application have the following main advantages:

[0057] The pronunciation recognition method described in this application involves: acquiring a pre-organized table of similar phonemes; collecting standard pronunciation audio and its corresponding phoneme annotation set from a preset standard pronunciation library; obtaining the pronunciation score value of each phoneme in the standard pronunciation audio according to a preset first algorithm formula, the standard pronunciation audio, and the corresponding phoneme annotation set; obtaining a pronunciation score threshold based on the pronunciation score value of each phoneme in the standard pronunciation audio and a preset threshold calculation formula; acquiring a pronunciation audio to be tested; performing pronunciation detection and scoring on each phoneme in the pronunciation audio to be tested according to a preset second algorithm formula to obtain the detection score value of each phoneme in the pronunciation audio to be tested; and identifying the pronunciation accuracy of each phoneme in the pronunciation audio to be tested by using preset detection rules, the detection score value of each phoneme in the pronunciation audio to be tested, and the pronunciation score threshold. By optimizing the algorithm for similar phonemes, the accuracy of pronunciation recognition is improved. In terms of the overall solution, the traditional GOP scoring algorithm is used for initial screening. By setting a pronunciation score threshold, phonemes with inaccurate pronunciation are quickly eliminated. By introducing prior knowledge of the similar phoneme table into the GOP algorithm, a second scoring is performed to reduce the interference of similar phonemes on the final GOP score. This expands the applicability of the GOP algorithm and improves the algorithm's ability to distinguish similar phonemes. The overall algorithm process is clear, concise, and easy to apply, further improving the accuracy of pronunciation recognition. Attached Figure Description

[0058] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0060] Figure 2 A flowchart of an embodiment of the pronunciation recognition method according to this application;

[0061] Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 202 shown;

[0062] Figure 4 yes Figure 2 A flowchart of a specific embodiment of step 203 shown;

[0063] Figure 5 yes Figure 2 A flowchart of a specific embodiment of step 205 shown;

[0064] Figure 6 yes Figure 2 A flowchart of a specific embodiment of step 206 shown;

[0065] Figure 7 A schematic diagram of the structure of an embodiment of the speech recognition device according to this application;

[0066] Figure 8 yes Figure 7 A schematic diagram of a specific embodiment of module 706 shown;

[0067] Figure 9 A schematic diagram of the structure of an embodiment of the computer device according to this application. Detailed Implementation

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0069] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0070] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0071] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0072] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0073] Terminal devices 101, 102, and 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers, etc.

[0074] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal devices 101, 102, and 103.

[0075] It should be noted that the pronunciation recognition method provided in this application embodiment is generally executed by a server, and correspondingly, the pronunciation recognition device is generally set in the server.

[0076] It should be understood that Figure 1The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0077] Continue to refer to Figure 2 A flowchart of an embodiment of the pronunciation recognition method according to this application is shown. The pronunciation recognition method includes the following steps:

[0078] Step 201: Obtain a pre-organized similar phoneme table, collect standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio from a preset standard pronunciation library, wherein the similar phoneme table contains all target phonemes with similar phonemes, and the similar phonemes corresponding to each target phoneme.

[0079] In this embodiment, the relationship between each target phoneme and its corresponding similar phoneme is one-to-one or one-to-many. That is, a target phoneme may have only one similar phoneme or it may have multiple similar phonemes.

[0080] In this embodiment, the standard pronunciation audio is a pre-recorded standard pronunciation audio based on the audio to be tested. The audio to be tested is the pronunciation audio collected from the test personnel. Specifically, the pronunciation audio collected from the test personnel can be business or sales scripts learned by insurance, banking, or e-commerce telemarketing personnel. By performing pronunciation recognition on the business or sales scripts learned by insurance, banking, or e-commerce telemarketing personnel, the pronunciation accuracy of insurance, banking, or e-commerce telemarketing personnel can be standardized, avoiding incorrect pronunciation during calls with customers.

[0081] In this embodiment, the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio are, for example, English pronunciation audio, which is a sequence of English pronunciation audio, and the phoneme annotation set is the phonetic symbol sequence corresponding to the English pronunciation audio sequence; similarly, for Chinese pronunciation audio, the standard pronunciation audio is a sequence of Chinese pronunciation audio, and the phoneme annotation set is the Chinese phonetic symbol sequence corresponding to the Chinese pronunciation audio sequence.

[0082] Step 202: Based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio, obtain the pronunciation score value of each phoneme in the standard pronunciation audio.

[0083] In this embodiment, before performing the step of obtaining the pronunciation score of each phoneme in the standard pronunciation audio based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio, the method further includes: performing pre-emphasis, frame segmentation, and windowing operations on the standard pronunciation audio in the standard pronunciation library to obtain the standard audio preprocessing result; obtaining the MFCC acoustic features of the standard pronunciation audio based on a preset acoustic feature extraction component and the standard audio preprocessing result; and obtaining the logarithmic phoneme posterior probability of each phoneme in the standard pronunciation audio based on the MFCC acoustic features of the standard pronunciation audio and a preset voiceprint recognition model, wherein the preset voiceprint recognition model includes a Kaldi-based voiceprint recognition model.

[0084] The acoustic feature extraction component can be an MFCC acoustic feature extraction program implemented using MATLAB and Python code. MATLAB is a commercial mathematical software produced by MathWorks, used in fields such as data analysis, wireless communication, deep learning, image processing and computer vision, signal processing, quantitative finance and risk management, robotics, and control systems.

[0085] Kaldi is a speech recognition tool developed in C++ and licensed under the Apache License. For Kaldi-based voiceprint recognition, traditional acoustic features, including Mel-frequency cepstral coefficients (MFCC), perceptual linear prediction coefficients (PLP), deep features, and energy-normalized spectral coefficients (PNCC), can all be selected as acoustic features for voiceprint recognition at the feature extraction level and perform well. By performing voiceprint recognition on these acoustic features, the logarithmic phoneme posterior probability (LPP) value of each phoneme in the target audio can be further obtained through GOP-NN (Neural Network-Goodness of Pronunciation).

[0086] Continue to refer to Figure 3 , Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 202 shown includes:

[0087] Step 301: Select the maximum value from the logarithmic posterior probability of each phoneme in the standard pronunciation audio.

[0088] Step 302: The maximum logarithmic phoneme posterior probability is pre-set to the sum of the logarithmic phoneme posterior probabilities of all phonemes in the phoneme annotation set;

[0089] Step 303: Traverse all phonemes in the phoneme annotation set and take different phonemes in the phoneme annotation set as target phonemes in turn;

[0090] Step 304, according to the preset first algorithm formula: GOP(p1)=PP(p1)-∑ q1∈Q LPP(q1) is used to obtain the pronunciation score of each phoneme in the standard pronunciation audio, where p1 represents the target phoneme, GOP(p1) represents the pronunciation score of the target phoneme p1, LPP(p1) represents the logarithmic phoneme posterior probability of the target phoneme p1, Q represents all phonemes in the phoneme annotation set, q1 represents the phoneme with the highest pronunciation score in the phoneme annotation set, and LPP(q1) represents the logarithmic phoneme posterior probability of phoneme q1, which is the sum of the logarithmic phoneme posterior probabilities of all phonemes in the phoneme annotation set.

[0091] In this embodiment, since the logarithmic phoneme posterior probability of phoneme q1 is the maximum value selected from the logarithmic phoneme posterior probabilities of each phoneme in the standard pronunciation audio, the range of the pronunciation score values ​​of each phoneme in the standard pronunciation audio is (-∞, 0).

[0092] One issue is that the phoneme annotation set Q contains a large number of phonemes; for example, English phonemes typically contain 48. Calculating the LPP value of all phonemes in Q every time would be computationally expensive. To facilitate the implementation of the GOP algorithm, an approximation is needed in practice. The sum of the LPP values ​​of all phonemes is replaced with the phoneme with the highest posterior probability in the logarithmic phoneme set, thus saving computational overhead.

[0093] Step 203: Calculate the pronunciation score threshold based on the pronunciation score values ​​of each phoneme in the standard pronunciation audio.

[0094] Continue to refer to Figure 4 , Figure 4 yes Figure 2 A flowchart of a specific embodiment of step 203 shown includes:

[0095] Step 401: Based on the pronunciation score and average value algorithm formula of each phoneme in the standard pronunciation audio, calculate the average pronunciation score of all phonemes in the phoneme annotation set;

[0096] Step 402: Calculate the standard deviation of the pronunciation scores of all phonemes in the phoneme annotation set based on the pronunciation score values ​​of each phoneme in the standard pronunciation audio, the mean and standard deviation algorithm formula of the pronunciation scores;

[0097] Step 403: Obtain the pronunciation score threshold using the preset threshold calculation formula: T=μ+α*σ+β, where T represents the pronunciation score threshold, μ represents the mean pronunciation score of all phonemes in the phoneme annotation set, σ represents the standard deviation of the pronunciation scores of all phonemes in the phoneme annotation set, and α and β are preset empirical constants.

[0098] In order to satisfy T∈(-∞, 0], the T value calculated by the threshold calculation formula should generally be slightly less than 0. In this case, the value range (-∞, 0] is divided into two value intervals (-∞, T] and (T, 0] by the pronunciation score threshold. If the GOP pronunciation score of a certain phoneme to be tested is between (T, 0], it is preliminarily judged that the pronunciation is accurate. At the same time, the closer the GOP pronunciation score of the phoneme to be tested is to 0, the more correct the pronunciation is.

[0099] Specifically, after obtaining the mean and standard deviation of the pronunciation scores of all phonemes in the phoneme annotation set, the pronunciation score threshold is obtained by linearization through a preset threshold calculation formula, which facilitates threshold comparison during subsequent audio detection.

[0100] Step 204: Obtain the audio of the pronunciation to be tested.

[0101] In this embodiment, after performing the step of obtaining the audio to be tested, the method further includes: performing pre-emphasis, framing, and windowing operations on the audio to be tested to obtain a preprocessing result of the audio to be tested; obtaining the MFCC acoustic features of the audio to be tested based on the acoustic feature extraction component and the preprocessing result of the audio to be tested, wherein the parameters set when performing pre-emphasis, framing, and windowing operations on the audio to be tested are consistent with the parameters set when performing pre-emphasis, framing, and windowing operations on standard audio in the standard pronunciation library; and obtaining the logarithmic phoneme posterior probability of each phoneme in the audio to be tested based on the MFCC acoustic features of the audio to be tested and the Kaldi-based voiceprint recognition model.

[0102] In this embodiment, by ensuring that the parameters set when performing pre-emphasis, framing, and windowing operations on the audio to be tested are consistent with the parameters set when performing pre-emphasis, framing, and windowing operations on the standard audio in the standard pronunciation library, the consistency of preprocessing of the standard audio and the audio to be tested is guaranteed. This ensures that when using the Kaldi-based voiceprint recognition model to obtain the logarithmic phoneme posterior probability of each phoneme in the audio to be tested, if the pronunciation is accurate, the logarithmic phoneme posterior probability of each phoneme in the audio to be tested is consistent with the logarithmic phoneme posterior probability of each phoneme in the standard audio, or the logarithmic phoneme posterior probability of each phoneme in the audio to be tested satisfies the preset pronunciation score threshold.

[0103] Step 205: Perform pronunciation detection and scoring on each phoneme in the audio to be tested, and obtain the detection score value of each phoneme in the audio to be tested.

[0104] Continue to refer to Figure 5 , Figure 5 yes Figure 2 A flowchart of a specific embodiment of step 205 shown includes:

[0105] Step 501: Traverse all phonemes in the audio of the pronunciation to be tested, and take different phonemes in the audio of the pronunciation to be tested as phonemes in turn;

[0106] Step 502, according to the preset second algorithm formula: GOP(p2)=PP(p2)-∑ q1∈Q LPP(q1) performs pronunciation detection and scoring on each phoneme in the audio to be tested, and obtains the detection score value of each phoneme in the audio to be tested. Here, p2 represents the phoneme to be tested, GOP(p2) represents the detection score value of the phoneme p2, LPP(p2) represents the logarithmic phoneme posterior probability of the phoneme p2, Q represents all phonemes in the phoneme annotation set, q1 represents the phoneme with the highest pronunciation score in the phoneme annotation set, and LPP(q1) represents the logarithmic phoneme posterior probability of phoneme q1, which is the sum of the logarithmic phoneme posterior probabilities of all phonemes in the phoneme annotation set.

[0107] Step 503: Organize the detection score values ​​of each phoneme in the audio to be tested into a detection score value set.

[0108] In this embodiment, since the logarithmic phoneme posterior probability of phoneme q1 is the maximum value selected from the logarithmic phoneme posterior probabilities of each phoneme in the standard pronunciation audio, the range of the pronunciation score values ​​of each phoneme in the test pronunciation audio is also (-∞, 0).

[0109] Step 206: Identify the pronunciation accuracy of each phoneme in the audio to be tested by using preset detection rules, the detection score value of each phoneme in the audio to be tested, and the pronunciation score threshold.

[0110] Continue to refer to Figure 6 , Figure 6 yes Figure 2 A flowchart of a specific embodiment of step 206 shown includes:

[0111] Step 601: Sequentially obtain the elements in the set of detection score values;

[0112] Step 602: Compare each element in the set of detection score values ​​obtained sequentially with the pronunciation score threshold to obtain the comparison result;

[0113] Step 603: Based on the comparison results, determine the relationship between each element in the detection score set and the pronunciation score threshold;

[0114] Step 604: If the current element is not greater than the pronunciation score threshold, then the phoneme corresponding to the current element is identified as having inaccurate pronunciation.

[0115] Step 605: If the current element is greater than the pronunciation score threshold, then according to the similar phoneme table, the detection score value of the current phoneme is re-acquired, and the re-acquired detection score value of the current phoneme is compared with the pronunciation score threshold to obtain the comparison result. If the current element is not greater than the pronunciation score threshold, then the pronunciation of the phoneme corresponding to the current element is identified as inaccurate; if the current element is greater than the pronunciation score threshold, then the pronunciation of the phoneme corresponding to the current element is identified as accurate.

[0116] In this embodiment, the step of re-obtaining the detection score of the current phoneme based on the similar phoneme table specifically includes: identifying all similar phonemes corresponding to the current phoneme based on the similar phoneme table, wherein each similar phoneme in the similar phoneme table has a pre-set corresponding logarithmic phoneme posterior probability; obtaining the logarithmic phoneme posterior probabilities of all similar phonemes corresponding to the current phoneme, and filtering out the maximum value, setting it as the sum of the logarithmic phoneme posterior probabilities of all similar phonemes; according to the preset third algorithm formula: Calculate the detection score of the current phoneme, where p3 represents the current phoneme, GOP(p3) represents the detection score of the current phoneme p3, LPP(p3) represents the logarithmic posterior probability of the current phoneme p3, and Q... x q1 represents the set of all similar phonemes corresponding to the current phoneme, and q2 represents the set of similar phonemes Q. x The phoneme with the highest pronunciation score is represented by LPP(q2), which is the logarithmic phoneme posterior probability of phoneme q2, i.e., the sum of the logarithmic phoneme posterior probabilities of all similar phonemes.

[0117] The accuracy of pronunciation recognition is improved by optimizing the algorithm for similar phonemes. The overall solution uses a traditional Group of Pronouncers (GOP) scoring algorithm for initial screening. By setting a pronunciation score threshold, phonemes with inaccurate pronunciation are quickly eliminated. Prior knowledge from a similar phoneme table is then incorporated into the GOP algorithm for a second scoring, reducing the interference of similar phonemes on the final GOP score. This expands the applicability of the GOP algorithm and enhances its ability to distinguish similar phonemes. The overall algorithm flow is clear, concise, and easy to apply, further improving the accuracy of pronunciation recognition.

[0118] In this embodiment, each similar phoneme in the similar phoneme table is pre-set with a corresponding logarithmic phoneme posterior probability. Specifically, the logarithmic phoneme posterior probability of each similar phoneme can be obtained from historical standard pronunciation audio. Alternatively, if the preset first algorithm formula obtains the logarithmic phoneme posterior probability of each phoneme in the standard pronunciation audio and the phonemes in the standard pronunciation audio happen to contain some similar phonemes, the logarithmic phoneme posterior probability of these similar phonemes can be directly obtained as the set value.

[0119] This application obtains a pre-organized table of similar phonemes, collects standard pronunciation audio and its corresponding phoneme annotation set from a pre-set standard pronunciation library; obtains the pronunciation score value of each phoneme in the standard pronunciation audio according to a pre-set first algorithm formula, the standard pronunciation audio, and its corresponding phoneme annotation set; obtains a pronunciation score threshold based on the pronunciation score value of each phoneme in the standard pronunciation audio and a pre-set threshold calculation formula; obtains the pronunciation audio to be tested; performs pronunciation detection and scoring on each phoneme in the pronunciation audio to be tested according to a pre-set second algorithm formula, obtaining the detection score value of each phoneme in the pronunciation audio to be tested; and identifies the pronunciation accuracy of each phoneme in the pronunciation audio to be tested by using pre-set detection rules, the detection score value of each phoneme in the pronunciation audio to be tested, and the pronunciation score threshold. By optimizing the algorithm for similar phonemes, the accuracy of pronunciation recognition is improved. In terms of the overall solution, the traditional GOP scoring algorithm is used for initial screening. By setting a pronunciation score threshold, phonemes with inaccurate pronunciation are quickly eliminated. By introducing prior knowledge of the similar phoneme table into the GOP algorithm, a second scoring is performed to reduce the interference of similar phonemes on the final GOP score. This expands the applicability of the GOP algorithm and improves the algorithm's ability to distinguish similar phonemes. The overall algorithm process is clear, concise, and easy to apply, further improving the accuracy of pronunciation recognition.

[0120] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0121] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0122] In this embodiment, the accuracy of pronunciation recognition is improved by optimizing the algorithm for similar phonemes. The overall scheme uses a traditional GOP scoring algorithm for initial screening. By setting a pronunciation score threshold, phonemes with inaccurate pronunciation are quickly eliminated. Prior knowledge from a similar phoneme table is introduced into the GOP algorithm for a second scoring, reducing the interference of similar phonemes on the final GOP score. This expands the applicability of the GOP algorithm and enhances its ability to distinguish similar phonemes. The overall algorithm flow is clear, concise, and easy to apply, further improving the accuracy of pronunciation recognition.

[0123] Further reference Figure 7 As a response to the above Figure 2 The present application provides an embodiment of a pronunciation recognition device to implement the method shown, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0124] like Figure 7 As shown, the pronunciation recognition device 700 described in this embodiment includes: a first data acquisition module 701, a standard score calculation module 702, a score threshold calculation module 703, a second data acquisition module 704, a detection score calculation module 705, and an accuracy recognition module 706. Wherein:

[0125] The first data acquisition module 701 is used to acquire a pre-organized similar phoneme table, collect standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio from a preset standard pronunciation library, wherein the similar phoneme table contains all target phonemes with similar phonemes, and the similar phonemes corresponding to each target phoneme;

[0126] The standard score calculation module 702 is used to obtain the pronunciation score value of each phoneme in the standard pronunciation audio based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio.

[0127] The scoring threshold calculation module 703 is used to calculate the pronunciation score threshold based on the pronunciation score value of each phoneme in the standard pronunciation audio.

[0128] The second data acquisition module 704 is used to acquire the audio of the pronunciation to be tested;

[0129] The detection score calculation module 705 is used to perform pronunciation detection scoring on each phoneme in the audio to be tested, and to obtain the detection score value of each phoneme in the audio to be tested.

[0130] The accuracy recognition module 706 is used to identify the pronunciation accuracy of each phoneme in the audio to be tested by using preset detection rules, the detection score value of each phoneme in the audio to be tested, and the pronunciation score threshold.

[0131] In some specific embodiments of this application, the pronunciation recognition device 700 further includes: an MFCC acoustic feature extraction module, wherein the MFCC acoustic feature extraction module is used to perform pre-emphasis, framing, and windowing operations on the standard pronunciation audio in the standard pronunciation library to obtain the standard audio preprocessing result; it is also used to obtain the MFCC acoustic features of the standard pronunciation audio according to a preset acoustic feature extraction component and the standard audio preprocessing result; it is also used to perform pre-emphasis, framing, and windowing operations on the pronunciation audio to be tested to obtain the pronunciation audio preprocessing result; and it is also used to obtain the MFCC acoustic features of the pronunciation audio to be tested according to the acoustic feature extraction component and the pronunciation audio preprocessing result, wherein the parameters set when performing pre-emphasis, framing, and windowing operations on the pronunciation audio to be tested are consistent with the parameters set when performing pre-emphasis, framing, and windowing operations on the standard pronunciation audio in the standard pronunciation library.

[0132] In some specific embodiments of this application, the pronunciation recognition device 700 further includes: a logarithmic phoneme posterior probability recognition module, wherein the logarithmic phoneme posterior probability recognition module is used to obtain the logarithmic phoneme posterior probability of each phoneme in the standard pronunciation audio based on the MFCC acoustic features of the standard pronunciation audio and a preset voiceprint recognition model, wherein the preset voiceprint recognition model includes a Kaldi-based voiceprint recognition model, and is also used to obtain the logarithmic phoneme posterior probability of each phoneme in the pronunciation audio to be tested based on the MFCC acoustic features of the pronunciation audio to be tested and the Kaldi-based voiceprint recognition model.

[0133] Continue to refer to Figure 8 , Figure 8 yes Figure 7 The schematic diagram shows a specific embodiment of module 706. The accuracy recognition module 706 includes a detection element acquisition submodule 801, a score size first comparison submodule 802, a size relationship judgment submodule 803, an accuracy first recognition submodule 804, a detection score recalculation submodule 805, a score size second comparison submodule 806, an accuracy second recognition submodule 807, and an accuracy third recognition submodule 808, wherein:

[0134] The detection element acquisition submodule 801 is used to sequentially acquire elements in the detection score value set;

[0135] The first comparison submodule 802 is used to compare each element in the set of detection score values ​​obtained sequentially with the pronunciation score threshold to obtain the comparison result;

[0136] The size relationship judgment submodule 803 is used to determine the size relationship between each element in the detection score value set and the pronunciation score threshold based on the comparison result;

[0137] The accuracy first recognition submodule 804 is used to identify that the phoneme corresponding to the current element is not pronounced accurately if the current element is not greater than the pronunciation score threshold.

[0138] The detection score recalculation submodule 805 is used to re-obtain the detection score value of the current phoneme according to the similar phoneme table if the current element is greater than the pronunciation score threshold.

[0139] The second comparison submodule 806 is used to compare the detection score value of the reacquired current phoneme with the pronunciation score threshold to obtain a comparison result;

[0140] The second accuracy recognition submodule 807 is used to identify that the phoneme corresponding to the current element is not pronounced accurately if the current element is not greater than the pronunciation score threshold.

[0141] The accuracy third recognition submodule 808 is used to identify the accurate pronunciation of the phoneme corresponding to the current element if the current element is greater than the pronunciation score threshold.

[0142] This application obtains a pre-organized similar phoneme table, collects standard pronunciation audio and its corresponding phoneme annotation set from a preset standard pronunciation library; obtains the pronunciation score value of each phoneme in the standard pronunciation audio according to a preset first algorithm formula, the standard pronunciation audio, and its corresponding phoneme annotation set; obtains a pronunciation score threshold based on the pronunciation score value of each phoneme in the standard pronunciation audio and a preset threshold calculation formula; obtains the pronunciation audio to be tested; performs pronunciation detection and scoring on each phoneme in the pronunciation audio to be tested according to a preset second algorithm formula, obtaining the detection score value of each phoneme in the pronunciation audio to be tested; and identifies the pronunciation accuracy of each phoneme in the pronunciation audio to be tested by using preset detection rules, the detection score value of each phoneme in the pronunciation audio to be tested, and the pronunciation score threshold. By optimizing the accuracy recognition module, it is made capable of multiple recognitions, thereby improving the accuracy of pronunciation recognition. The traditional GOP scoring algorithm is used for initial screening. By setting a pronunciation score threshold, phonemes with inaccurate pronunciation are quickly eliminated. By introducing prior knowledge of the similar phoneme table into the GOP algorithm, a second scoring is performed to reduce the interference of similar phonemes on the final GOP score. This expands the applicability of the GOP algorithm and improves its ability to distinguish similar phonemes. The overall algorithm process is clear, concise, and easy to apply, further improving the accuracy of pronunciation recognition.

[0143] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0144] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0145] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 9 , Figure 9 This is a basic structural block diagram of the computer device in this embodiment.

[0146] The computer device 9 includes a memory 9a, a processor 9b, and a network interface 9c that are interconnected via a system bus. It should be noted that only the computer device 9 with components 9a-9c is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0147] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0148] The memory 9a includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 9a may be an internal storage unit of the computer device 9, such as the hard disk or memory of the computer device 9. In other embodiments, the memory 9a may also be an external storage device of the computer device 9, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 9. Of course, the memory 9a may also include both the internal storage unit and its external storage device of the computer device 9. In this embodiment, the memory 9a is typically used to store the operating system and various application software installed on the computer device 9, such as computer-readable instructions for speech recognition methods. In addition, the memory 9a can also be used to temporarily store various types of data that have been output or will be output.

[0149] In some embodiments, the processor 9b may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 9b is typically used to control the overall operation of the computer device 9. In this embodiment, the processor 9b is used to execute computer-readable instructions stored in the memory 9a or to process data, for example, to execute computer-readable instructions for the pronunciation recognition method.

[0150] The network interface 9c may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 9 and other electronic devices.

[0151] The computer device proposed in this embodiment belongs to the fields of artificial intelligence and financial technology, and is applied to the process of recognizing the accuracy of pronunciation in voice customer service in financial business. This application improves the accuracy of pronunciation recognition by optimizing the GOP algorithm for similar phonemes. In terms of the overall scheme, the traditional GOP scoring algorithm is used for initial screening. By setting a pronunciation score threshold, phonemes with inaccurate pronunciation are quickly eliminated. By introducing prior knowledge of the similar phoneme table into the GOP algorithm, a second scoring is performed to reduce the interference of similar phonemes on the final GOP score, thereby expanding the applicability of the GOP algorithm and improving the algorithm's ability to distinguish similar phonemes. The overall algorithm process is clear, concise, and easy to apply, further improving the accuracy of pronunciation recognition.

[0152] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by a processor to cause the processor to perform the steps of the pronunciation recognition method described above.

[0153] The computer-readable storage medium proposed in this embodiment belongs to the fields of artificial intelligence and financial technology, and is applied to the process of recognizing the accuracy of pronunciation in voice customer service in financial business. This application improves the accuracy of pronunciation recognition by optimizing the GOP algorithm based on similar phonemes. In terms of the overall scheme, the traditional GOP scoring algorithm is used for initial screening. By setting a pronunciation score threshold, phonemes with inaccurate pronunciation are quickly eliminated. By introducing prior knowledge of the similar phoneme table into the GOP algorithm, a second scoring is performed to reduce the interference of similar phonemes on the final GOP score, thereby expanding the applicability of the GOP algorithm and improving the algorithm's ability to distinguish similar phonemes. The overall algorithm process is clear, concise, and easy to apply, further improving the accuracy of pronunciation recognition.

[0154] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0155] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A pronunciation recognition method, characterized in that, Includes the following steps: Obtain a pre-organized similar phoneme table, collect standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio from a preset standard pronunciation library, wherein the similar phoneme table contains all target phonemes with similar phonemes, and the similar phonemes corresponding to each target phoneme; Based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio, obtain the pronunciation score value of each phoneme in the standard pronunciation audio; Calculate the pronunciation score threshold based on the pronunciation score value of each phoneme in the standard pronunciation audio; Obtain the audio of the pronunciation to be tested; The pronunciation of each phoneme in the audio to be tested is detected and scored to obtain the detection score value of each phoneme in the audio to be tested. The accuracy of pronunciation of each phoneme in the audio to be tested is identified by using preset detection rules, detection scores of each phoneme in the audio to be tested, and the pronunciation score threshold. Specifically, this includes: Retrieve elements from the detection score set sequentially; Each element in the set of detection scores obtained sequentially is compared with the pronunciation score threshold to obtain the comparison result; Based on the comparison results, determine the relationship between each element in the set of detection score values ​​and the pronunciation score threshold; If the current element is not greater than the pronunciation score threshold, then the phoneme corresponding to the current element is identified as being inaccurately pronounced; If the current element is greater than the pronunciation score threshold, then according to the similar phoneme table, the detection score value of the current phoneme is re-obtained. The re-obtained detection score value of the current phoneme is compared with the pronunciation score threshold to obtain the comparison result. If the current element is not greater than the pronunciation score threshold, then the pronunciation of the phoneme corresponding to the current element is identified as inaccurate. If the current element is greater than the pronunciation score threshold, then the pronunciation of the phoneme corresponding to the current element is identified as accurate. Specifically, the step of re-obtaining the detection score value of the current phoneme according to the similar phoneme table includes: Based on the similar phoneme table, all similar phonemes corresponding to the current phoneme are identified. Each similar phoneme in the similar phoneme table has a pre-set corresponding logarithmic phoneme posterior probability. Specifically, the logarithmic phoneme posterior probability of each similar phoneme is obtained based on historical standard pronunciation audio. Obtain the logarithmic phoneme posterior probability of all similar phonemes corresponding to the current phoneme, and filter out the maximum value, setting it as the sum of the logarithmic phoneme posterior probabilities of all similar phonemes; According to the preset third algorithm formula: Calculate the detection score of the current phoneme, where, Indicates the current phoneme, Indicates the current phoneme The detection score, Indicates the current phoneme The logarithmic posterior probability of phonemes This represents the set of all similar phonemes corresponding to the current phoneme. Represents the set of similar phonemes The phoneme with the highest pronunciation score in Chinese, Representing phonemes The logarithmic posterior probability of phonemes.

2. The pronunciation recognition method according to claim 1, characterized in that, Before performing the step of obtaining the pronunciation score value of each phoneme in the standard pronunciation audio based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio, the method further includes: The standard pronunciation audio in the standard pronunciation library is pre-emphasized, framed, and windowed to obtain the standard audio preprocessing result; Based on the preset acoustic feature extraction component and the standard audio preprocessing results, the MFCC acoustic features of the standard pronunciation audio are obtained; Based on the MFCC acoustic features of the standard pronunciation audio and the preset voiceprint recognition model, the logarithmic phoneme posterior probability of each phoneme in the standard pronunciation audio is obtained, wherein the preset voiceprint recognition model includes a Kaldi-based voiceprint recognition model. The step of obtaining the pronunciation score value of each phoneme in the standard pronunciation audio based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio specifically includes: The maximum value is selected from the logarithmic posterior probabilities of each phoneme in the standard pronunciation audio. The maximum logarithmic phoneme posterior probability is pre-set to the sum of the logarithmic phoneme posterior probabilities of all phonemes in the phoneme annotation set; Iterate through all phonemes in the phoneme annotation set, and take different phonemes in the phoneme annotation set as target phonemes in turn; According to the preset first algorithm formula: Obtain the pronunciation score value of each phoneme in the standard pronunciation audio, wherein, Indicates the target phoneme, Indicates the target phoneme The pronunciation score, Indicates the target phoneme The logarithmic posterior probability of phonemes This represents all phonemes in the phoneme annotation set. This indicates the phoneme with the highest pronunciation score in the phoneme annotation set. Representing phonemes The logarithmic posterior probability of phonemes.

3. The pronunciation recognition method according to claim 1 or 2, characterized in that, The step of calculating the pronunciation score threshold based on the pronunciation score values ​​of each phoneme in the standard pronunciation audio specifically includes: Based on the pronunciation score and average value algorithm formula of each phoneme in the standard pronunciation audio, the average pronunciation score of all phonemes in the phoneme annotation set is calculated; Based on the pronunciation score of each phoneme in the standard pronunciation audio, the mean and standard deviation algorithm formula of the pronunciation scores, the standard deviation of the pronunciation scores of all phonemes in the phoneme annotation set is calculated; Calculated using a preset threshold formula: Obtain the pronunciation score threshold, where, This represents the pronunciation score threshold. This represents the average pronunciation score of all phonemes in the phoneme annotation set. This represents the standard deviation of the pronunciation scores for all phonemes in the phoneme annotation set. and These are preset empirical constants.

4. The pronunciation recognition method according to claim 2, characterized in that, After performing the step of acquiring the audio of the pronunciation to be tested, the method further includes: The audio to be tested is subjected to pre-emphasis, framing, and windowing operations to obtain the preprocessing result of the audio to be tested; Based on the acoustic feature extraction component and the preprocessing result of the audio to be tested, the MFCC acoustic features of the audio to be tested are obtained, wherein the parameters set when performing pre-emphasis, framing and windowing operations on the audio to be tested are consistent with the parameters set when performing pre-emphasis, framing and windowing operations on the standard audio in the standard pronunciation library; Based on the MFCC acoustic features of the audio to be tested and the Kaldi-based voiceprint recognition model, the logarithmic posterior probability of each phoneme in the audio to be tested is obtained.

5. The pronunciation recognition method according to claim 4, characterized in that, The step of performing pronunciation detection and scoring on each phoneme in the audio to be tested, and obtaining the detection score value of each phoneme in the audio to be tested, specifically includes: Traverse all phonemes in the audio of the pronunciation to be tested, and take the different phonemes in the audio of the pronunciation to be tested as the phonemes to be tested in turn; According to the preset second algorithm formula: The pronunciation of each phoneme in the audio to be tested is detected and scored, and the detection score value of each phoneme in the audio to be tested is obtained. Indicates the phoneme to be tested. Indicates the phoneme to be tested The detection score, Indicates the phoneme to be tested The logarithmic posterior probability of phonemes This represents all phonemes in the phoneme annotation set. This indicates the phoneme with the highest pronunciation score in the phoneme annotation set. Representing phonemes The logarithmic posterior probability of a phoneme; The detection scores of each phoneme in the audio to be tested are organized into a set of detection scores.

6. A pronunciation recognition device, characterized in that, The pronunciation recognition device implements the steps of the pronunciation recognition method as described in any one of claims 1 to 5, and the pronunciation recognition device comprises: The first data acquisition module is used to acquire a pre-organized similar phoneme table, collect standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio from a preset standard pronunciation library, wherein the similar phoneme table contains all target phonemes with similar phonemes, and the similar phonemes corresponding to each target phoneme; The standard score calculation module is used to obtain the pronunciation score value of each phoneme in the standard pronunciation audio based on the standard pronunciation audio and the phoneme annotation set corresponding to the standard pronunciation audio. The scoring threshold calculation module is used to calculate the pronunciation score threshold based on the pronunciation score value of each phoneme in the standard pronunciation audio. The second data acquisition module is used to acquire the audio of the pronunciation to be tested; The detection score calculation module is used to perform pronunciation detection and scoring on each phoneme in the audio to be tested, and to obtain the detection score value of each phoneme in the audio to be tested. The accuracy recognition module is used to identify the pronunciation accuracy of each phoneme in the audio to be tested by using preset detection rules, the detection score value of each phoneme in the audio to be tested, and the pronunciation score threshold.

7. A computer device comprising a memory and a processor, the memory storing computer-readable instructions, wherein the processor, when executing the computer-readable instructions, implements the steps of the pronunciation recognition method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the pronunciation recognition method as described in any one of claims 1 to 5.

Images