Voice recognition method, device, storage medium, and program product

Abstract

The disclosed embodiment discloses a voice recognition method, device, storage medium, and program product. The method includes: collecting voice signals of multiple testers and performing feature extraction to obtain an MFCC feature matrix; obtaining the MFCC parameter set of each tester based on the MFCC feature matrix and a preset new indicator library; performing feature comparison between the MFCC parameter sets of all normal voice personnel and the MFCC parameter sets of all hoarseness patients to obtain the significance probability of each new indicator; determining the new indicator with a significance probability less than a preset probability threshold as an available indicator, and verifying the available indicator; after the verification is passed, constructing a voice recognition model based on the available indicator; and using the voice recognition model to identify the user's voice state. The method can explore and screen new indicators for voice recognition by quantifying the MFCC indicators, and realize accurate voice recognition based on the new indicators, which helps clinical physicians to better evaluate voice function.

Description

Technical Field

[0001] The present disclosure relates to the field of sound processing technology, and in particular to a voice recognition method, device, storage medium, and program product. Background Art

[0002] Hoarseness is a common symptom. It includes benign vocal cord diseases such as acute laryngitis, chronic laryngitis, various types of vocal cord abnormalities, vocal cord polyps and cysts, and malignant vocal cord diseases such as glottic cancer, which can also be collectively referred to as voice disorders. It seriously affects the quality of life of patients. Voice function assessment plays a key role in the diagnosis and treatment of these diseases.

[0003] Traditional voice quality assessment mainly relies on parameters such as average fundamental frequency, fundamental frequency perturbation and amplitude perturbation, while modern technologies such as MFCC have been widely used in fields such as speech recognition and audio analysis. Despite this, existing studies often ignore the in-depth interpretation and quantification of MFCC features when constructing voice disease prediction models, making it difficult for clinicians to fully trust the results of these models. In addition, existing voice assessment methods fail to fully utilize the potential of MFCC and lack quantifiable MFCC indicators, which limits their effectiveness in clinical applications. Summary of the invention

[0004] In view of this, the embodiments of the present disclosure provide a voice recognition method, device, storage medium, and program product, which can explore and screen new indicators for voice recognition by quantifying MFCC indicators, and achieve accurate voice recognition based on the new indicators, which helps clinical physicians to better evaluate voice function.

[0005] In a first aspect, the present disclosure provides a voice recognition method, which adopts the following technical solution:

[0006] Collecting speech signals of a plurality of test persons, wherein the plurality of test persons include persons with normal voices and patients with hoarseness;

[0007] Extracting features from the speech signal to obtain an MFCC feature matrix;

[0008] Based on the MFCC feature matrix and a preset new indicator library, obtaining the MFCC parameter set of each tester, wherein the new indicator library includes a plurality of new indicators;

[0009] Compare the MFCC parameter sets of all people with normal voices with the MFCC parameter sets of all patients with hoarseness to obtain the significance probability of each new indicator;

[0010] Determine the new indicator whose significance probability is less than the preset probability threshold as an available indicator, and verify the available indicator;

[0011] After verification, a sound recognition model is constructed based on the available indicators;

[0012] The voice recognition model is used to recognize the user's voice status.

[0013] Optionally, the feature comparison of the MFCC parameter sets of all persons with normal voices and the MFCC parameter sets of all patients with hoarseness to obtain the significance probability of each new indicator includes:

[0014] The MFCC parameter sets of all persons with normal voices are constructed as the first sample set;

[0015] The MFCC parameter sets of all hoarseness patients are constructed as the second sample set;

[0016] Based on the first sample set and the second sample set, obtaining distribution statistics and degrees of freedom of each new indicator;

[0017] Based on the distribution statistics and the degrees of freedom, obtaining the significance probability of each new indicator;

[0018] The calculation formula of the distribution statistic is as follows:

[0019]

[0020] Among them, t i is the distribution statistic of the i-th new indicator; is the mean value of the MFCC parameters of the i-th new indicator in the first sample set; is the mean value of the MFCC parameters of the i-th new indicator in the second sample set; is the combined variance of the i-th new indicator, ; N1 is the total number of MFCC parameters of the i-th new indicator in the first sample set; N2 is the total number of MFCC parameters of the i-th new indicator in the second sample set; is the MFCC parameter variance of the i-th new indicator in the first sample set; is the MFCC parameter variance of the i-th new indicator in the second sample set;

[0021] The calculation formula of the degree of freedom is as follows:

[0022]

[0023] in, is the degrees of freedom of the i-th new indicator.

[0024] Optionally, the verifying the available indicator includes:

[0025] Building a traditional indicator library, wherein the traditional indicator library includes a plurality of traditional indicators;

[0026] Screening the traditional indicators in the traditional indicator library, deleting invalid indicators, and retaining valid indicators;

[0027] Using multiple machine learning algorithms, constructing multiple machine learning models based on the effective indicators and the available indicators;

[0028] Evaluate multiple machine learning models to obtain a first performance parameter of each machine learning model;

[0029] Based on the first performance parameter, select the machine learning algorithm used by the machine learning model with the best performance as an available algorithm;

[0030] Using the available algorithm, constructing a verification model based on the available indicators;

[0031] Evaluating the verification model to obtain a second performance parameter of the verification model;

[0032] Based on the first performance parameter and the second performance parameter of the best performing machine learning model, determine whether the available indicator has passed the verification.

[0033] Optionally, the verifying the available indicator includes:

[0034] Building a traditional indicator library, wherein the traditional indicator library includes a plurality of traditional indicators;

[0035] Screening the traditional indicators in the traditional indicator library, deleting invalid indicators, and retaining valid indicators;

[0036] Using a plurality of machine learning algorithms, constructing a plurality of machine learning models based on the effective indicators and the available indicators;

[0037] Evaluate multiple machine learning models to obtain a first performance parameter of each machine learning model;

[0038] Based on the first performance parameter, selecting the machine learning model with the best performance as an available model;

[0039] Rank the importance of the effective indicators and available indicators adopted by the available models;

[0040] It is determined whether the available indicator has passed the verification based on the ranking result.

[0041] Optionally, constructing a sound recognition model based on the available indicators includes:

[0042] Dividing the multiple available indicators into a necessary indicator set and a pending indicator set based on the ranking result;

[0043] Construct multiple different indicator combination sets based on the necessary indicator set and the pending indicator set;

[0044] Build multiple candidate models based on multiple machine learning algorithms and multiple different indicator combination sets;

[0045] Determining whether the candidate model is overfitting or underfitting;

[0046] If so, the candidate model is excluded;

[0047] If not, obtaining a plurality of performance evaluation parameters of the candidate model;

[0048] Based on the performance evaluation parameters, the preset objective function and the constraints, the optimal candidate model is selected as the sound recognition model.

[0049] Optionally, the calculation formula of the objective function is as follows:

[0050]

[0051] Among them, s is the sequence number of the machine learning algorithm used by the candidate model; v is the sequence number of the indicator combination set used by the selected model; U sv is the comprehensive performance score of the candidate model constructed using the sth machine learning algorithm and the vth indicator combination set; w Ac 、w Se 、w Sp 、w AUC 、w BU 、w CPU 、w CU is the preset weight; Ac sv is the accuracy of the candidate model; Se sv is the sensitivity of the candidate model; Sp sv is the specificity of the candidate model; AUC sv is the area under the ROC curve of the candidate model; BU sv is the bandwidth occupancy of the candidate model; CPU sv is the CPU usage of the candidate model; CU sv is the computational occupancy of the candidate model;

[0052] The preset constraints include bandwidth occupancy constraints, CPU occupancy constraints, and computing occupancy constraints;

[0053] The expression of the broadband occupancy constraint condition is as follows:

[0054]

[0055] Among them, Max BU is the maximum bandwidth occupancy;

[0056] The expression of the CPU occupancy constraint condition is as follows:

[0057]

[0058] Among them, Max CPU is the maximum CPU usage;

[0059] The expression for calculating the occupancy constraint condition is as follows:

[0060]

[0061] Among them, Max CU is the maximum computing occupancy.

[0062] Optionally, the new indicator library includes at least one new indicator of mean, variance, standard deviation, 25th percentile, 75th percentile, minimum value, maximum value, median, skewness, and kurtosis.

[0063] In a second aspect, the embodiment of the present disclosure further provides a voice recognition system, which adopts the following technical solution:

[0064] A voice collection module, used to collect voice signals of multiple test persons, wherein the multiple test persons include persons with normal voices and patients with hoarseness;

[0065] A feature extraction module is used to extract features from the speech signal and obtain an MFCC feature matrix;

[0066] A parameter acquisition module, used to acquire the MFCC parameter set of each tester based on the MFCC feature matrix and a preset new indicator library, wherein the new indicator library includes a plurality of new indicators;

[0067] The feature comparison module is used to compare the MFCC parameter sets of all normal voice people with the MFCC parameter sets of all hoarseness patients to obtain the significance probability of each new indicator;

[0068] An indicator verification module, used to determine the new indicator whose significance probability is less than a preset probability threshold as an available indicator, and verify the available indicator;

[0069] A model building module, used to build a sound recognition model based on the available indicators after verification;

[0070] The voice recognition module is used to recognize the user's voice status using the voice recognition model.

[0071] In a third aspect, the embodiments of the present disclosure further provide a computer device, which adopts the following technical solution:

[0072] The computer device comprises:

[0073] at least one processor; and,

[0074] a memory communicatively connected to the at least one processor; wherein,

[0075] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform any of the above-mentioned voice recognition methods.

[0076] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute any of the above-mentioned voice recognition methods.

[0077] In a fifth aspect, an embodiment of the present disclosure further provides a computer program product, including a computer program / instruction, which implements the steps of any of the above methods when executed by a processor.

[0078] The voice recognition method provided by the embodiment of the present disclosure collects the voice signals of people with normal voices and patients with hoarseness, and extracts the MFCC feature matrix to provide data support for the subsequent quantification of multiple new indicators. This method utilizes the wide application of MFCC in the field of speech recognition and audio analysis, so that it can efficiently capture the frequency characteristics of voice signals, and provides a solid foundation for subsequent quantification and analysis. On the basis of obtaining the MFCC feature matrix, multiple new indicators are further combined with it to generate a quantification result, that is, an MFCC parameter set. This method not only fully utilizes the advantages of MFCC in spectrum analysis, but also increases the diversity and discrimination of features by introducing new indicators. By comparing the features of the MFCC parameter sets of people with normal voices and patients with hoarseness, the significance probability of each new indicator is obtained, which helps to identify indicators that are of great significance to the diagnosis of voice disorders. Further, the new indicators with a significance probability less than a preset probability threshold are determined as available indicators and strictly verified. This step not only ensures the stability and reliability of the selected indicators, but also greatly improves the interpretability of voice evaluation. Through the verified available indicators, clinicians can more easily understand and trust the evaluation results. After verification, a sound recognition model is built based on the available indicators. This model can be used to identify the user's voice status, providing an automated and standardized voice assessment tool.

[0079] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following preferred embodiments are specifically cited and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0081] Figure 1 A flowchart of a voice recognition method provided by an embodiment of the present disclosure;

[0082] Figure 2 A schematic diagram of MFCC spectrograms of multiple test persons provided in an embodiment of the present disclosure;

[0083] Figure 3 A flowchart of a method for verifying available indicators provided in an embodiment of the present disclosure;

[0084] Figure 4 Another schematic diagram of a flow chart of a method for verifying available indicators provided by an embodiment of the present disclosure;

[0085] Figure 5 A schematic diagram of the average impact of effective indicators and available indicators on the magnitude of model output provided by the embodiments of the present disclosure;

[0086] Figure 6 A schematic diagram of the impact of effective indicators and available indicators on model output provided by the embodiment of the present disclosure;

[0087] Figure 7 A flow chart of a method for constructing a sound recognition model provided in an embodiment of the present disclosure;

[0088] Figure 8 A schematic diagram of a voice recognition model user interface provided by an embodiment of the present disclosure;

[0089] Fig. 9 A functional block diagram of a voice recognition system provided by an embodiment of the present disclosure;

[0090] Fig.10 A schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0091] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0092] It should be clear that the following embodiments of the present disclosure are described by specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other in the absence of conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0093] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0094] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0095] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0096] Reference Figure 1 The present disclosure provides a voice recognition method, comprising the following steps:

[0097] S1: Collect voice signals of multiple test persons, including those with normal voices and those with hoarseness;

[0098] S2: Extract features of speech signals and obtain MFCC feature matrix;

[0099] S3: Based on the MFCC feature matrix and a preset new indicator library, the MFCC parameter set of each tester is obtained, wherein the new indicator library includes multiple new indicators;

[0100] S4: Compare the MFCC parameter sets of all people with normal voices with the MFCC parameter sets of all patients with hoarseness to obtain the significance probability of each new indicator;

[0101] S5: Determine the new indicators whose significance probability is less than the preset probability threshold as available indicators, and verify the available indicators;

[0102] S6: After verification, a sound recognition model is built based on available indicators;

[0103] S7: Use a sound recognition model to identify the user's voice status.

[0104] The voice recognition method disclosed in the present invention collects voice signals of people with normal voices and patients with hoarseness, and extracts MFCC feature matrices to provide data support for subsequent quantification of multiple new indicators. This method utilizes the wide application of MFCC in the fields of speech recognition and audio analysis, so that it can efficiently capture the frequency characteristics of voice signals, and provides a solid foundation for subsequent quantification and analysis. On the basis of obtaining the MFCC feature matrix, multiple new indicators are further combined with it to generate a quantification result, that is, an MFCC parameter set. This method not only fully utilizes the advantages of MFCC in spectrum analysis, but also increases the diversity and discrimination of features by introducing new indicators. By comparing the features of the MFCC parameter sets of people with normal voices and patients with hoarseness, the significance probability of each new indicator is obtained, which helps to identify indicators that are of great significance to the diagnosis of voice disorders. Further, the new indicators with a significance probability less than a preset probability threshold are determined as available indicators and strictly verified. This step not only ensures the stability and reliability of the selected indicators, but also greatly improves the interpretability of voice evaluation. Through the verified available indicators, clinicians can more easily understand and trust the evaluation results. After verification, a voice recognition model is built based on the available indicators. This model can be used to identify the user's voice status (such as normal voice or hoarseness), providing an automated and standardized voice assessment tool.

[0105] In summary, this voice recognition method combines MFCC feature extraction and machine learning technology to explore new indicators that can be used for voice recognition, providing a new and effective tool for diagnosing voice disorders, which helps to improve the accuracy, interpretability and efficiency of voice recognition, and also provides a new direction for voice medicine research.

[0106] In S1, multiple testers are recruited, including those with normal voices (e.g., 69) and patients with hoarseness (e.g., 140). Recording equipment (e.g., USB microphones, multi-channel recording equipment, etc.) is used to collect several notes uttered by the testers, and a recording file is generated. The recording file contains the voice data of the testers from the beginning to the end of self-pronunciation, wherein the collected notes include at least one of / a: / , / i: / , and / æ / . The testers pronounce each note for approximately 2-3 seconds. The collection environment is required to be indoors, and the noise level must not exceed 40 decibels.

[0107] Capture any continuous 1-second recording data from the recording file and save it as a voice file in ".wav" format as the input data for the next step of analysis. There must be no interruption in the middle of the 1-second segment. If the overall acquisition quality is poor, such as interruption in the middle, it is necessary to re-acquire to ensure the quality of all recording data, thereby ensuring the stability and reliability of the data, which is convenient for subsequent analysis and processing.

[0108] In S2, feature extraction is performed on the speech signal in the speech file. The feature extraction includes enhancing high-frequency signals, dividing the signal into short time frames, using Hamming windows or other window functions to reduce frame edge effects, converting time domain signals into frequency domain signals, mapping spectral energy to Mel scale, taking logarithms to compress dynamic range, extracting cepstral coefficients, etc. By executing these steps, an MFCC (Mel-frequency cepstral coefficient) feature matrix is ​​obtained, or the MFCC feature matrix is ​​obtained by analyzing using the librosa speech processing package based on Python programming software version 3.9.

[0109] In S3, the preset new indicator library includes at least one of the following new indicators: mean (MFCC-Mean), variance (MFCC-Var), standard deviation (MFCC-Std), 25th percentile (MFCC-25%), 75th percentile (MFCC-75%), minimum (MFCC-Min), maximum (MFCC-Max), median (MFCC-Median), skewness (MFCC-skew) and kurtosis (MFCC-kurt). Among them, mean refers to the average value of MFCC features; variance refers to the variance of MFCC features; standard deviation refers to the standard deviation of MFCC features; 25th percentile refers to the 25th percentile of MFCC features; 75th percentile refers to the 75th percentile of MFCC features; minimum refers to the minimum value of MFCC features; maximum refers to the maximum value of MFCC features; median refers to the median of MFCC features; skewness refers to the value reflecting the direction and degree of skewness of MFCC feature distribution; kurtosis refers to the value reflecting the peak degree of MFCC feature distribution. These statistical values ​​represent the variability of a column of data and are derived from the statistics of statistical logarithms. In the past, these indicators were mainly used in the study of heart rate variability of electrocardiogram signals and the extraction of two-dimensional histogram features in imaging genomics. However, in the field of MFCC, there is almost no relevant research on these indicators. Therefore, these indicators are regarded as new indicators, aiming to analyze the performance of each new indicator in the field of voice recognition, in order to break through the bottleneck in this field.

[0110] The MFCC feature matrix is ​​two-dimensional data, in which each row represents the MFCC coefficient of a frame. The new indicators in the new indicator library are simpler to calculate in a single dimension, while it is more complicated to directly calculate these new indicators for a multi-dimensional matrix. Therefore, the MFCC feature matrix is ​​converted into one-dimensional data to obtain a feature sequence. A statistical algorithm is used to calculate these new indicators for the feature sequence to obtain multiple MFCC parameters. These parameters constitute the MFCC parameter set of the tester, in which each MFCC parameter is a specific value calculated according to different new indicators for the feature sequence, that is, the quantitative result of the new indicator.

[0111] These quantized MFCC parameters are obtained based on the statistical calculation of the MFCC feature matrix, providing a general description of the MFCC features.

[0112] Based on the extracted MFCC feature matrix, the MFCC spectrogram is drawn, which shows the distribution of MFCC coefficients over time. Figure 2Schematic diagrams of MFCC spectrograms of multiple test persons are displayed. The first three are MFCC spectrograms of three patients with hoarseness, and the last three are MFCC spectrograms of three people with normal voices. The MFCC spectrograms of the three patients with hoarseness are compared with the MFCC spectrograms of the three people with normal voices. The MFCC spectrograms of the hoarse patients are obviously coarser than those of the people with normal voices. It can be seen that some features of MFCC can clearly distinguish between normal and hoarse voices. In traditional technologies, features that can be used to distinguish between normal and hoarse voices have been identified, but some key features are still missed, which makes it difficult for clinicians to use MFCC for auxiliary assessment of voice function. Therefore, by collecting multiple new indicators and screening the new indicators, new features for distinguishing between normal and hoarse voices can be discovered, thereby establishing a new voice assessment system based on MFCC, which is helpful for clinicians to better assess voice function.

[0113] In S4, the MFCC parameter sets of all people with normal voices are constructed into a sample set, recorded as the first sample set, and the MFCC parameter sets of all patients with hoarseness are constructed into another sample set, recorded as the second sample set. The independent sample t-test algorithm is used to compare the differences between the first sample set and the second sample set in each new indicator to obtain the significance probability (P value) of each new indicator. Specifically, based on the first sample set and the second sample set, the distribution statistic (t value) and degrees of freedom of each new indicator are obtained, and based on the distribution statistic and the degrees of freedom, the significance probability (P value) of each new indicator is obtained, where the calculation formula of the distribution statistic is as follows:

[0114]

[0115] In the formula, i is the sequence number of the new indicator; t i is the distribution statistic of the i-th new indicator; is the mean value of the MFCC parameters of the i-th new indicator in the first sample set, specifically, the mean value of all MFCC parameters calculated according to the i-th new indicator in the first sample set; is the mean value of the MFCC parameters of the i-th new indicator in the second sample set, specifically, the mean value of all MFCC parameters calculated according to the i-th new indicator in the second sample set; is the combined variance of the i-th new indicator, ; N1 is the total number of MFCC parameters of the i-th new indicator in the first sample set, specifically the total number of all MFCC parameters calculated according to the i-th new indicator in the first sample set; N2 is the total number of MFCC parameters of the i-th new indicator in the second sample set, specifically the total number of all MFCC parameters calculated according to the i-th new indicator in the second sample set; is the MFCC parameter variance of the i-th new indicator in the first sample set, specifically refers to the variance of all MFCC parameters calculated according to the i-th new indicator in the first sample set; is the MFCC parameter variance of the i-th new indicator in the second sample set, specifically refers to the variance of all MFCC parameters calculated according to the i-th new indicator in the second sample set.

[0116] The degree of freedom is calculated as follows:

[0117]

[0118] In the formula, is the degrees of freedom of the i-th new indicator.

[0119] Based on the calculated t-value and degrees of freedom, look up a t-distribution table or use statistical software (such as Python's SciPy library) to calculate the P value.

[0120] In S5, when the significance probability is less than the preset probability threshold, the new indicator is determined to be an available indicator. When the significance probability is not less than the preset probability threshold, the new indicator is determined to be an unavailable indicator and deleted. The preset probability threshold is 0.05.

[0121] Refer to Table 1 for the significance probability distribution table of the new indicators:

[0122] Table 1

[0123]

[0124] Among them, the first column shows the names of 10 new indicators, the second column shows the mean and standard deviation of the MFCC parameters of people with normal voice under different new indicators, for example, "2.5±2.1" in the second row and second column, 2.5 represents the mean of the MFCC parameters of people with normal voice under the new indicator MFCC-Mean, 2.1 represents the standard deviation of the MFCC parameters of people with normal voice under the new indicator MFCC-Mean, the third column shows the mean and standard deviation of the MFCC parameters of patients with hoarseness under different new indicators, the fourth column shows the t-values ​​of different new indicators, and the fifth column shows the P-values ​​of different new indicators.

[0125] In a specific implementation, referring to Figure 3 A flowchart of a method for verifying available indicators is shown. The method for verifying available indicators includes the following steps:

[0126] S501: construct a traditional indicator library, which includes several traditional indicators;

[0127] S502: Screening traditional indicators in the traditional indicator library, deleting invalid indicators, and retaining valid indicators;

[0128] S503: Using multiple machine learning algorithms, build multiple machine learning models based on effective indicators and available indicators;

[0129] S504: Evaluate multiple machine learning models to obtain a first performance parameter of each machine learning model;

[0130] S505: Based on the first performance parameter, select a machine learning algorithm used by a machine learning model with the best performance as an available algorithm;

[0131] S506: Using available algorithms, constructing a verification model based on available indicators;

[0132] S507: Evaluate the verification model to obtain a second performance parameter of the verification model;

[0133] S508: Determine whether the available indicators have passed the verification based on the first performance parameter and the second performance parameter of the machine learning model with the best performance.

[0134] In S501, the traditional indicator library includes at least one traditional indicator of mean fundamental frequency (Mean F0), standard deviation of fundamental frequency (STD-F0), maximum fundamental frequency (Maximum F0), minimum fundamental frequency (Minimum F0), fundamental frequency perturbation (Jitter), amplitude perturbation (Shimmer), glottal noise excitation ratio (GNE Ratio), irregularity (Roughness), breath sound (Breathiness), and overall severity (Overall Severity). These traditional indicators can be collected using LingWaves software. Among them, the average fundamental frequency refers to the average value of the fundamental frequency in the sound signal, reflecting the average pitch of the sound; the fundamental frequency standard deviation refers to the standard deviation of the fundamental frequency value, reflecting the stability or volatility of the sound pitch; the maximum fundamental frequency refers to the maximum value of the fundamental frequency in the sound signal, reflecting the highest pitch of the sound; the minimum fundamental frequency refers to the minimum value of the fundamental frequency in the sound signal, reflecting the lowest pitch of the sound; the fundamental frequency perturbation refers to the relative change between the fundamental frequency cycles, reflecting the irregularity of the sound; the amplitude perturbation refers to the relative change in the sound amplitude, reflecting the instability of the sound; the glottal noise excitation ratio refers to the proportion of the noise component in the glottal excitation, reflecting the roughness and hoarseness of the sound; irregularity refers to the roughness or hoarseness of the sound, which can be obtained through the spectrum analysis of the sound; breath sound refers to the proportion of the breathing component in the sound, reflecting the lightness or tension of the sound; the overall severity refers to the overall evaluation of the sound quality obtained after comprehensively considering multiple sound indicators, which can be used to assess the severity of voice diseases.

[0135] In S502, an independent sample t-test algorithm is used to screen the traditional indicators in the traditional indicator library, and invalid indicators with P values ​​not less than the preset probability threshold are deleted, and valid indicators with P values ​​less than the preset probability threshold are retained. The principle of the method for screening out valid indicators from the traditional indicator library here is the same as the principle of the method for screening out available indicators from the new indicator library mentioned above, and will not be repeated here.

[0136] In S503, multiple machine learning algorithms are preset, including Random Forest, Logistic Regression, K-nearest neighbor, Decision Tree, AdaBoost, XGBoost, LGBMClassifier, Linear Discriminant Analysis, Multi-layer perceptron and Support Vector Machine (SVM).

[0137] Different machine learning models are built through each machine learning algorithm, and effective indicators and available indicators are used as input features of each machine learning model.

[0138] In S504 and S505, each machine learning model is evaluated using a preset evaluation index to obtain the accuracy, sensitivity, specificity, and area under the ROC curve (AUC) of each machine learning model. These calculated performance parameters are recorded as first performance parameters. By comparing the first performance parameters of each machine learning model, the machine learning model with the best performance is selected, and the machine learning algorithm used by the model is determined to be an available algorithm. Among them, the accuracy, sensitivity, and specificity are obtained by calculating the confusion matrix, and the area under the ROC curve is obtained by calculating the ROC curve.

[0139] Refer to Table 2 for the machine learning algorithm modeling evaluation results:

[0140] Table 2

[0141]

[0142] The table shows the accuracy, sensitivity, specificity, and area under the ROC curve of machine learning models built using different machine learning algorithms. It is obvious that the performance parameters of the machine learning model built using the logistic regression algorithm are significantly higher than others. Therefore, the logistic regression algorithm is judged to be an available algorithm.

[0143] In S506, an available algorithm is used to reconstruct a verification model, where the model uses available indicators as input features.

[0144] In S507, the verification model is evaluated using preset evaluation indicators to obtain the accuracy, sensitivity, specificity, and area under the ROC curve of the verification model, and these calculated performance parameters are recorded as second performance parameters.

[0145] In S508, in a feasible specific solution, the accuracy, sensitivity, specificity, and area under the ROC curve of the best performing machine learning model are subtracted from the accuracy, sensitivity, specificity, and area under the ROC curve of the validation model to obtain the accuracy difference, sensitivity difference, specificity difference, and area under the ROC curve difference. Among these four differences, if the absolute values ​​of L differences are less than the preset difference threshold, it is determined that the available indicator verification has passed, otherwise, it is determined that the available indicator verification has not passed, where 1≤L≤4.

[0146] In another feasible specific scheme, the accuracy, sensitivity, specificity, and area under the ROC curve of the best performing machine learning model are weighted and summed to obtain a first performance score, and the accuracy, sensitivity, specificity, and area under the ROC curve of the verification model are weighted and summed to obtain a second performance score. If the absolute value of the difference between the first performance score and the second performance score is less than the preset difference threshold, it is determined that the available indicator verification has passed. If the absolute value of the difference between the first performance score and the second performance score is not less than the preset difference threshold, it is determined that the available indicator verification has not passed, and the preset probability threshold is increased to obtain multiple new indicators, and the new indicators are continuously verified until the verification is successful.

[0147] In another specific implementation, referring to Figure 4 Another flowchart of the method for verifying the available indicators is shown, and the method for verifying the available indicators includes the following steps:

[0148] S511: Build a traditional indicator library, which includes several traditional indicators;

[0149] S512: Screening traditional indicators in the traditional indicator library, deleting invalid indicators, and retaining valid indicators;

[0150] S513: Use multiple machine learning algorithms to build multiple machine learning models based on effective and available indicators;

[0151] S514: Evaluate multiple machine learning models to obtain a first performance parameter of each machine learning model;

[0152] S515: Based on the first performance parameter, select the machine learning model with the best performance as the available model;

[0153] S516: Rank the importance of effective indicators and available indicators adopted by available models;

[0154] S517: Determine whether the available indicators have passed verification based on the ranking results.

[0155] The above-mentioned methods S511-S514 and S501-S504 have the same principles and will not be described in detail here.

[0156] In S515, the machine learning model with the best performance is determined as an available model, referring to Figure 5 ,It is obvious that the performance parameters of the machine learning ,model constructed using the logistic regression algorithm are significantly higher than the others. ,Therefore, the machine learning model constructed using the logistic ,regression algorithm is judged as an available model.

[0157] In S516, the SHAP feature importance analysis algorithm is used to rank the importance of the effective indicators and available indicators used by the available models.

[0158] In S517, in a feasible specific solution, it is determined whether there is an available indicator whose importance is higher than the effective indicator according to the ranking result. If so, it is determined that the available indicator verification has passed; if not, it is determined that the available indicator verification has failed.

[0159] In another feasible specific solution, the average of the importance rankings of all valid indicators is obtained and recorded as the first-rank average, and the average of the importance rankings of all available indicators is obtained and recorded as the second-rank average. If the difference between the first-rank average and the second-rank average is less than the preset difference threshold, it is determined that the available indicator verification has passed; if the difference between the first-rank average and the second-rank average is not less than the preset difference threshold, it is determined that the available indicator verification has failed.

[0160] Reference Figure 5 Schematic diagram showing the average impact of effective and available indicators on the magnitude of model output (Mean(|SHAP value|), average SHAP absolute value), see Figure 6 A schematic diagram of the impact of effective indicators and available indicators on model output is shown (SHAP value, SHAP average value). The average SHAP absolute value and SHAP value of effective indicators and available indicators are weighted summed to obtain the importance evaluation values ​​of effective indicators and available indicators. The effective indicators and available indicators are ranked in order of importance evaluation value from large to small. It can be seen that the three available indicators of mean, variance and standard deviation are ranked high, even surpassing the effective indicators, and the ranking of other available indicators is also relatively high. Therefore, the available indicators are verified.

[0161] In S6, when the available indicators are verified, a sound recognition model is constructed based on the available indicators, and the model can accurately identify the user's voice status.

[0162] In a first specific implementation scheme, the verification model is trained and fine-tuned until the verification model achieves the desired effect, and the verification model is used as a sound recognition model for voice recognition.

[0163] In a second specific embodiment, referring to Figure 7 A flowchart of a method for building a sound recognition model is shown. The method for building a sound recognition model based on available indicators includes the following steps:

[0164] S61: dividing the multiple available indicators into a necessary indicator set and a pending indicator set based on the ranking result;

[0165] S62: construct multiple different indicator combination sets based on the necessary indicator set and the pending indicator set;

[0166] S63: Construct multiple candidate models based on multiple machine learning algorithms and multiple different indicator combination sets;

[0167] S64: Determine whether the candidate model is overfitting or underfitting; if so, execute S65; if not, execute S66;

[0168] S65: Eliminate candidate models;

[0169] S66: Obtain multiple performance evaluation parameters of the candidate model, and execute S67;

[0170] S67: Based on the performance evaluation parameters, the preset objective function and the constraints, the optimal candidate model is selected as the sound recognition model.

[0171] In S61, the importance ranking of the available indicators is obtained based on the ranking results, and the ranking percentage of the available indicators is obtained based on the importance ranking. If the ranking percentage of the available indicators is greater than the preset percentage threshold, the available indicators are added to the necessary indicator set. If the ranking percentage of the available indicators is not greater than the preset percentage threshold, the available indicators are added to the pending indicator set. Among them, the necessary indicator set and the pending indicator set are initially empty, and the calculation formula for the ranking percentage is as follows:

[0172]

[0173] In the formula, j is the sequence number of the available indicators; H j is the ranking percentage of the jth available indicator; h j is the importance ranking of the jth available indicator; F is the total number of valid indicators and available indicators participating in the importance ranking.

[0174] In S62, all available indicators in the necessary indicator set are key indicators, and all combinations contain all available indicators in the necessary indicator set. Therefore, all possible initial combinations are generated based on the pending indicator set. For example, if the pending indicator set contains 5 available indicators, the number of initial combinations is 31. All available indicators in the necessary indicator set are added to each initial combination to obtain multiple different indicator combination sets.

[0175] In S63, a plurality of different machine learning algorithms are paired with a plurality of different indicator combination sets to construct a plurality of candidate models. For example, if there are 31 indicator combination sets and 10 machine learning algorithms, then 310 candidate models are constructed.

[0176] In S64-S67, whether the candidate model is overfitting or underfitting is determined by the performance difference between the training set and the test set, the learning curve, cross-validation, etc. If the candidate model is overfitting or underfitting, the candidate model is excluded. If the candidate model is neither overfitting nor underfitting, multiple performance evaluation parameters of the candidate model are obtained, and based on the performance evaluation parameters, the preset objective function and the constraint conditions, the optimal candidate model is selected as the sound recognition model. Among them, the performance evaluation parameters include at least one of the accuracy, sensitivity, specificity, area under the ROC curve, bandwidth occupancy, CPU occupancy, and computing occupancy of the candidate model.

[0177] The expression of the objective function is as follows:

[0178]

[0179] In the formula, s is the sequence number of the machine learning algorithm used by the candidate model; v is the sequence number of the indicator combination set used by the selected model; U sv is the comprehensive performance score of the candidate model (specifically, the model constructed using the sth machine learning algorithm and the vth indicator combination set); w Ac 、w Se 、w Sp 、w AUC 、w BU 、w CPU 、w CU is the preset weight; Ac sv is the accuracy of the candidate model, , TP is a true positive example, which refers to the number of samples predicted to be positive and actually positive, TN is a true negative example, which refers to the number of samples predicted to be negative and actually negative, FP is a false positive example, which refers to the number of samples predicted to be positive but actually negative, and FN is a false negative example, which refers to the number of samples predicted to be negative but actually positive; Se sv is the sensitivity of the candidate model, ;Sp sv is the specificity of the candidate model, ; AUC sv is the area under the ROC curve of the candidate model, , R + is the number of positive samples, R + = TP + FN, R - is the number of negative samples, R - =TN + FP,y m is the predicted score of the mth positive sample, y k is the prediction score of the kth negative class sample, is the indicator function, when y m >y k hour, , when ym ≤y k hour, BU sv is the bandwidth occupancy of the candidate model, , TDS is the total data volume of the candidate model during the transmission process, IT is the time used to transmit the candidate model data; CPU sv is the CPU usage of the candidate model, , CTU is the time actually used by the CPU to execute the candidate model task during the observation time (including the time spent by the CPU in executing candidate model calculation, data processing and other tasks), TOT is the observation time; CU sv is the computational occupancy of the candidate model, , TCO is the total number of computing operations performed during the operation of the candidate model, and COT is the time taken to execute all computing operations of the candidate model.

[0180] The preset constraints include bandwidth occupancy constraints, CPU occupancy constraints, and computing occupancy constraints.

[0181] The expression of the broadband occupancy constraint is as follows:

[0182]

[0183] Where Max BU The maximum bandwidth usage.

[0184] The expression of CPU usage constraint is as follows:

[0185]

[0186] Where Max CPU The maximum CPU usage.

[0187] The expression for calculating the occupancy constraint is as follows:

[0188]

[0189] Where Max CU is the maximum computing occupancy.

[0190] In S7, a voice recognition model is used to recognize the voice state of the user, referring to Figure 8 The diagram shows the interface for using the voice recognition model. Users record voices online, and the system automatically fills in the indicators required for the voice recognition model. Alternatively, users can enter the relevant indicators themselves and click the Submit button, and the system automatically displays the recognition results of the voice status.

[0191] Reference Fig. 9 The present disclosure provides a voice recognition system, comprising:

[0192] The voice collection module 101 is used to collect voice signals of multiple test persons, wherein the multiple test persons include persons with normal voices and patients with hoarseness;

[0193] The feature extraction module 102 is used to extract features from the speech signal and obtain the MFCC feature matrix;

[0194] The parameter acquisition module 103 is used to acquire the MFCC parameter set of each tester based on the MFCC feature matrix and a preset new indicator library, wherein the new indicator library includes a plurality of new indicators;

[0195] A feature comparison module 104 is used to compare the MFCC parameter sets of all normal voice people with the MFCC parameter sets of all hoarseness patients to obtain the significance probability of each new indicator;

[0196] The indicator verification module 105 is used to determine the new indicator whose significance probability is less than the preset probability threshold as an available indicator, and verify the available indicator;

[0197] A model building module 106 is used to build a sound recognition model based on available indicators after verification;

[0198] The voice recognition module 107 is used to recognize the user's voice status using a voice recognition model.

[0199] The various variations and specific examples of the voice recognition method provided above are also applicable to the voice recognition system provided in the present disclosure. Through the above detailed description of the voice recognition method, those skilled in the art can clearly understand the implementation method of the voice recognition system. For the sake of brevity of the specification, it will not be described in detail here.

[0200] The computer device according to the embodiment of the present disclosure includes a memory and a processor. The memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc.

[0201] The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of the present disclosure, the processor is used to run the computer-readable instructions stored in the memory, so that the computer device performs all or part of the steps of the voice recognition method of each embodiment of the present disclosure.

[0202] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present disclosure.

[0203] like Fig.10 A schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure is shown, which is a schematic diagram of the structure of a computer device suitable for implementing the embodiment of the present disclosure. Fig.10 The computer device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0204] like Fig.10 As shown, the computer device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage device into a random access memory (RAM). In the RAM, various programs and data required for the operation of the computer device are also stored. The processor, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0205] Typically, the following devices can be connected to the I / O interface: input devices such as sensors or visual information acquisition devices; output devices such as display screens; storage devices such as tapes, hard disks, etc.; and communication devices. The communication device can allow the computer device to communicate with other devices (such as edge computing devices) wirelessly or by wire to exchange data. Fig.10 A computer device having various devices is shown, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0206] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the voice recognition method of the embodiment of the present disclosure are executed.

[0207] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0208] According to the computer-readable storage medium of the embodiment of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the voice recognition method of each embodiment of the present disclosure are executed.

[0209] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0210] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0211] The basic principles of the present disclosure are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present disclosure are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present disclosure. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, and are not limitations. The above details do not limit the present disclosure to the necessity of adopting the above specific details to be implemented.

[0212] In the present disclosure, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The block diagrams of the devices, devices, equipment, and systems involved in the present disclosure are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0213] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a separate list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0214] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0215] Various changes, substitutions, and modifications of the techniques described herein may be made without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of the present disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of events, means, methods, and actions described above. Currently existing or later to be developed processes, machines, manufactures, compositions of events, means, methods, or actions that perform substantially the same functions or achieve substantially the same results as the corresponding aspects described herein may be utilized. Thus, the appended claims include such processes, machines, manufactures, compositions of events, means, methods, or actions within their scope.

[0216] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0217] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A voice recognition method, characterized in that: include: Collecting voice signals of multiple test persons, wherein the multiple test persons include persons with normal voices and persons with hoarse voices; Extracting features from the speech signal to obtain an MFCC feature matrix; Based on the MFCC feature matrix and a preset new indicator library, obtaining the MFCC parameter set of each tester, wherein the new indicator library includes a plurality of new indicators; The MFCC parameter sets of all persons with normal voices are constructed as the first sample set; The MFCC parameter sets of all hoarse persons are constructed as the second sample set; Based on the first sample set and the second sample set, obtaining distribution statistics and degrees of freedom of each new indicator; Based on the distribution statistics and the degrees of freedom, obtaining the significance probability of each new indicator; Determine the new indicator whose significance probability is less than the preset probability threshold as an available indicator, and verify the available indicator; After verification, a sound recognition model is constructed based on the available indicators; Using the voice recognition model to recognize the user's voice; The calculation formula of the distribution statistic is as follows: Among them, t i is the distribution statistic of the i-th new indicator; is the mean value of the MFCC parameters of the i-th new indicator in the first sample set; is the mean value of the MFCC parameters of the i-th new indicator in the second sample set; is the combined variance of the i-th new indicator, ; N1 is the total number of MFCC parameters of the i-th new indicator in the first sample set; N2 is the total number of MFCC parameters of the i-th new indicator in the second sample set; is the MFCC parameter variance of the i-th new indicator in the first sample set; is the MFCC parameter variance of the i-th new indicator in the second sample set; The calculation formula of the degree of freedom is as follows: in, is the degrees of freedom of the i-th new indicator.

2. The voice recognition method according to claim 1, characterized in that: The verifying of the available indicators includes: Building a traditional indicator library, wherein the traditional indicator library includes a plurality of traditional indicators; Screening the traditional indicators in the traditional indicator library, deleting invalid indicators, and retaining valid indicators; Using a plurality of machine learning algorithms, constructing a plurality of machine learning models based on the effective indicators and the available indicators; Evaluate multiple machine learning models to obtain a first performance parameter of each machine learning model; Based on the first performance parameter, select the machine learning algorithm used by the machine learning model with the best performance as an available algorithm; Using the available algorithm, constructing a verification model based on the available indicators; Evaluating the verification model to obtain a second performance parameter of the verification model; Based on the first performance parameter and the second performance parameter of the best performing machine learning model, determine whether the available indicator has passed the verification.

3. The voice recognition method according to claim 1, characterized in that: The verifying of the available indicators includes: Building a traditional indicator library, wherein the traditional indicator library includes a plurality of traditional indicators; Screening the traditional indicators in the traditional indicator library, deleting invalid indicators, and retaining valid indicators; Using a plurality of machine learning algorithms, constructing a plurality of machine learning models based on the effective indicators and the available indicators; Evaluate multiple machine learning models to obtain a first performance parameter of each machine learning model; Based on the first performance parameter, selecting the machine learning model with the best performance as an available model; Rank the importance of the effective indicators and available indicators adopted by the available models; It is determined whether the available indicator has passed the verification based on the ranking result.

4. The voice recognition method according to claim 3, characterized in that: The step of constructing a sound recognition model based on the available indicators includes: Dividing the multiple available indicators into a necessary indicator set and a pending indicator set based on the ranking result; Construct multiple different indicator combination sets based on the necessary indicator set and the pending indicator set; Build multiple candidate models based on multiple machine learning algorithms and multiple different indicator combination sets; Determining whether the candidate model is overfitting or underfitting; If so, the candidate model is excluded; If not, obtaining a plurality of performance evaluation parameters of the candidate model; Based on the performance evaluation parameters, the preset objective function and the constraints, the optimal candidate model is selected as the sound recognition model.

5. The voice recognition method according to claim 4, characterized in that: The calculation formula of the objective function is as follows: Among them, s is the sequence number of the machine learning algorithm used by the candidate model; v is the sequence number of the indicator combination set used by the selected model; U sv is the comprehensive performance score of the candidate model constructed using the sth machine learning algorithm and the vth indicator combination set; w Ac 、w Se 、w Sp 、w AUC 、w BU 、w CPU 、w CU is the preset weight; Ac sv is the accuracy of the candidate model; Se sv is the sensitivity of the candidate model; Sp sv is the specificity of the candidate model; AUC sv is the area under the ROC curve of the candidate model; BU sv is the bandwidth occupancy of the candidate model; CPU sv is the CPU usage of the candidate model; CU sv is the computational occupancy of the candidate model; The preset constraints include bandwidth occupancy constraints, CPU occupancy constraints, and computing occupancy constraints; The expression of the broadband occupancy constraint condition is as follows: Among them, Max BU is the maximum bandwidth occupancy; The expression of the CPU occupancy constraint condition is as follows: Among them, Max CPU is the maximum CPU usage; The expression for calculating the occupancy constraint condition is as follows: Among them, Max CU is the maximum computing occupancy.

6. The voice recognition method according to claim 1, characterized in that: The novel index library includes at least one novel index of mean, variance, standard deviation, 25th percentile, 75th percentile, minimum value, maximum value, median, skewness and kurtosis.

7. A computer device, characterized in that: The computer device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the voice recognition method according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the voice recognition method described in any one of claims 1-6.

9. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images