A sound quality evaluation method based on CEEMD and CWD

Abstract

The present application provides a sound quality evaluation method based on CEEMD and CWD. The evaluation method performs CEEMD on each collected noise signal sample to obtain several intrinsic mode functions (IMFs) of each. The sample entropy value is used to identify and remove the false components in the IMFs. The remaining IMF components are respectively subjected to CWD analysis. According to the CWD analysis results, a new concept of sound quality evaluation index - sound quality evaluation index SQDF is proposed, and SQDF is used as the characteristic parameter of the corresponding noise signal sample. Based on the extreme learning machine ELM, a connection between the characteristic parameter and the subjective evaluation value is established, and the sound quality of the noise can be evaluated using this relationship. This method can be used for the sound quality evaluation of various noises, can reduce the subjective evaluation cost of the noise, and can avoid the influence caused by various adverse factors in the subjective evaluation process. The sound quality can also be evaluated in the design stage of the product, thereby improving the design quality and reducing the R & D cost.

Description

Technical Field

[0001] This application belongs to the technical field of noise analysis and evaluation, and particularly relates to a sound quality evaluation method based on CEEMD (Complete Ensemble Empirical Mode Decomposition) and CWD (Choi-Williams Distribution). Background Art

[0002] At present, with the development of the automotive industry, simple sound pressure level control can no longer meet the customers' preferences for sound and the manufacturers' requirements for improving and designing sound quality. Currently, loudness, sharpness, roughness, flutter, and speech intelligibility are often used to evaluate sound quality. However, these traditional sound quality evaluation parameters are easily affected by broadband and narrowband noise in signal characteristics, which results in no stable correlation between subjective feelings and psychological parameters when using traditional sound quality evaluation methods for subjective evaluation. Therefore, how to associate the subjective evaluation results of people with the relevant physical parameters represented by sound quality is a difficult problem that needs to be solved urgently. Summary of the Invention

[0003] The purpose of the present invention is to provide a sound quality evaluation method based on CEEMD and CWD. By performing CEEMD on each collected noise signal sample, several orders of Intrinsic Mode Functions (IMFs) are obtained respectively. The false components in the IMFs are identified and removed using sample entropy values. The remaining IMF components are respectively subjected to CWD analysis. According to the CWD analysis results, a new concept of sound quality evaluation index - Sound Quality Index Function (SQDF) is proposed, and SQDF is used as the characteristic parameter of the corresponding noise signal sample. Based on the Extreme Learning Machine (ELM), a connection is established between the characteristic parameter and the subjective evaluation value. Using this relationship, the sound quality of the noise can be evaluated. This method can reduce the subjective evaluation cost of noise and avoid the influence of various adverse factors in the subjective evaluation process. The sound quality can also be evaluated during the design stage of the product, thereby improving the design quality and reducing the R & D cost.

[0004] The present application specifically adopts the following technical solutions:

[0005] A sound quality evaluation method based on CEEMD and CWD, the method comprising the following steps:

[0006] S1: Collect n noise signal samples, denoted as xi (N) (i = 1, 2, …, n), and subjectively evaluate the noise signal samples respectively;

[0007] S2: Perform CEEMD decomposition on the noise signal samples to obtain several orders of IMF components of each noise signal sample, and denote the IMF components as set B i (i = 1, 2, …, n);

[0008] S3: Calculate the sample entropy values of each order of IMF in set B i respectively, screen the IMF components according to the sample entropy values, eliminate the false components, and retain the effective components, denoted as set B i '(i = 1, 2, …, n);

[0009] S4: Perform CWD analysis on each order of IMF in each set B i ';

[0010] S5: Calculate the SQDF values of each order of IMF in each set B i ' and the total SQDF values of each noise signal sample;

[0011] S6: Use the total SQDF value of each noise signal sample as the corresponding characteristic parameter of the noise signal sample, and establish the relationship between the characteristic parameter and the subjective evaluation value of each noise signal sample by using ELM;

[0012] S7: Calculate the SQDF value of the noise to be evaluated, and perform sound quality evaluation by using the corresponding relationship established in step S6.

[0013] As a further description of this application, step S2 specifically includes the following steps:

[0014] S21: Add a pair of positive and negative white noises to the noise signal sample x i (N)

[0015]

[0016] where n(N) is the white noise component, are the signals after adding positive and negative white noises to the noise sample signal respectively;

[0017] S22: Perform EMD decomposition on respectively, and the decomposition process is as follows:

[0018] S221: Let a certain time series be x(N). Determine all its maximum points and minimum points. Use cubic spline function interpolation for the maximum points and the minimum points respectively to form the upper envelope curve u1(N) and the lower envelope curve u2(N) of the data, and take the average value m1(N) of u1(N) and u2(N), that is

[0019] m1(N) = (u1(N) + u2(N)) / 2

[0020] S222: Let h1(N) = x(N) - m1(N). If h1(N) simultaneously satisfies two conditions of the IMF: ① In the entire time course, the number of times crossing the zero point is equal to or at most differs by 1 from the number of extreme points; ② It is locally symmetric about the time axis, that is, the mean value of the upper envelope line defined by local maxima and the lower envelope line defined by local minima is 0, then h1(N) is the first-order IMF. If not satisfied, regard h1(N) as the new x(N), and m 11 (N) is the mean value of its upper and lower envelope lines, and there is h 11 (N) = h1(N) - m 11 (N). If h 11 (N) still does not satisfy, repeat this process k times to get

[0021] h 1k (N) = h 1(k-1) (N) - m 1k (N)

[0022] If the standard deviation SD between h 1k (N) and h 1(k-1) (N) is between 0.2 - 0.3, then stop repeating this process. At this time, h 1k (N) is the first-order IMF of the signal x(N). Denote c1(N) = h 1k (N), where the standard deviation SD calculation formula is:

[0023]

[0024] In the formula, N is the sequence length of the signal.

[0025] S223: Let r1(N) = x(N) - c1(N). Perform step S222 on r1(N) to successively obtain c2(N), c3(N) …… until r i (N) is a monotonic function and can no longer be decomposed into IMF;

[0026] After EMD decomposition, it can be expressed as

[0027]

[0028] Where \(l\) is the total number of IMF components, are the noise signal samples \(x\) i (N) the \(j\)th IMF component after adding positive and negative white noises;

[0029] S23: Add \(M\) different white noises, and perform steps S21 and S22 each time. The noise sample signal \(x\) i (N) The final IMF component is

[0030]

[0031] Where IMF ij is the final \(j\)th IMF component of the noise signal sample \(x\) i (N), are respectively the \(j\)th IMF component of \(x\) i (N) after adding positive and negative white noises for the \(k\)th time;

[0032] Denote the final IMF component of \(x\) i (N) as set \(B\) i .

[0033] As a further description of this application, step S3 specifically includes the following steps:

[0034] S31: The calculation process of calculating the sample entropy value of each order IMF in the set \(B_i\) is as follows:

[0035] S311: Let a certain time series be \(E(N)=\{e(1),e(2),\cdots,e(N)\}\), and arrange the time series in the order of \(m\)-dimensional vectors to obtain the following vectors:

[0036] E m (i)=[e(i),e(i + 1),\cdots,e(i + m - 1)] (1\leq i\leq N - m + 1)

[0037] The vector represents \(m\) consecutive \(e\) values starting from point \(i\);

[0038] S312: Define the distance m between \(E\) m (i) and \(E\) as the absolute value of the maximum difference among the elements they contain, then

[0039]

[0040] where \(k = 0,1,2,\cdots,m - 1\);

[0041] S313: Count the distance m between \(E\) m (i) and \(E\) The number A less than or equal to R i , then

[0042]

[0043] In the formula, R represents a set threshold value;

[0044] S314: Obtain A i the mean value of The calculation formula is

[0045]

[0046] S315: When calculating to the (m + 1)-dimensional, repeat steps S311 - S313, then

[0047]

[0048] Then the sample entropy SE is

[0049]

[0050] Under actual conditions, N cannot be infinite, so

[0051]

[0052] m takes 2 or 1, and the value range of R is 0.1SD' to 0.25SD' (SD' is the standard deviation of the time series E(N) = {e(1), e(2),..., e(N)});

[0053] S32: Screen the IMF components according to the sample entropy value calculated in S31, eliminate the IMF components whose sample entropy values approach 0, and retain the remaining IMF components;

[0054] S33: Perform steps S31 and S32 on each order of IMF components in the set B i , and denote the IMF set after eliminating the false components as B i '.

[0055] As a further description of this application, the S4 specifically includes the following steps:

[0056] S41: Perform CWD analysis on each order of IMF components in the set B i ' respectively, and the calculation method is as follows:

[0057]

[0058] In the formula, σ is a constant, and the value of σ ranges from 0.1 to 10;

[0059] S42: Perform step S41 on the effective IMF components of each of the noise signal samples respectively.

[0060] As a further illustration of the present application, S5 specifically includes the following steps:

[0061] S51: Calculate the SQDF values of the IMF components of each order in the set B i ' respectively, and the calculation method is as follows:

[0062]

[0063] In the formula, SQDF IMF-g is the SQDF value of the g-th order IMF component, w p×q is the amplitude matrix after CWD, w p×q (a, b) is the element value of the a-th row and the b-th column of the amplitude matrix w p×q , p is half of the number of spectral lines of the noise signal sample, q is the number of time-domain sampling points, and f max is the maximum frequency for analysis;

[0064] S52: Calculate the total SQDF values of the noise signal samples respectively, and the calculation method is as follows:

[0065]

[0066] In the formula, SQDF total is the total SQDF value of the noise signal sample, and L is the number of its effective IMF components.

[0067] S53: Perform steps S51 and S52 on each of the noise signal samples respectively, and calculate the total SQDF values of each of the noise signal samples.

[0068] As a further illustration of the present application, S6 specifically includes the following steps:

[0069] S61: Randomly select a part of the noise signal samples as the samples for ELM training. Use the total SQDF value of the selected noise signal samples as the input layer and the subjective evaluation value as the output layer, and train using the ELM network to establish the relationship between the total SQDF value and the subjective evaluation value of each of the noise signal samples;

[0070] S62: Use the remaining noise signal samples to verify the trained ELM network. If the standard deviation SD” of the difference between the sound quality evaluation value and the actual subjective evaluation value of each of the noise signal samples is not greater than 0.001, it is considered that the ELM network in step S61 is qualified; otherwise, it is considered unqualified, and continue to perform step S61 until the requirements are met.

[0071] As a further illustration of the present application, the randomly selected sample size of the noise signal is 70%-80%.

[0072] Compared with the prior art, the present application has the following beneficial technical effects:

[0073] (1) The present invention performs CEEMD decomposition on the original noise signal, which not only well preserves the time-frequency characteristics of the signal but also avoids mode overlapping. The IMF components obtained by decomposition are screened using sample entropy values to eliminate false components, thereby improving the signal-to-noise ratio.

[0074] (2) The present invention proposes a brand-new sound quality evaluation index - the sound quality evaluation index.

[0075] (3) The present invention combines CEEMD, sample entropy, CWD analysis, and ELM network, and proposes a brand-new sound quality evaluation method, combining the respective advantages of these methods to achieve higher evaluation accuracy.

[0076] (4) The present invention correlates the objective information of the noise with the subjective evaluation, establishes the connection between the noise signal and the subjective evaluation value, and subsequently, the noise signal can be modulated according to the subjective evaluation value to improve its sound quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 is a flowchart of the sound quality evaluation method based on CEEMD and CWD provided by the present application;

[0078] Figure 2 is a flowchart of the signal EMD provided by the present application;

[0079] Figure 3 is a schematic diagram of the ELM network structure provided by the present application;

[0080] Figure 4 is the time-domain waveform diagram and frequency spectrum diagram of each order IMF component of Sample 1 in Embodiment 1 of the present application; wherein, (a) time-domain waveform diagram; (b) frequency spectrum diagram;

[0081] Figure 5 is the CWD analysis result of the 1st order IMF component of Sample 1 in the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0082] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some, but not all, of the embodiments of the present application.

[0083] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but merely represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts fall within the scope of protection of the present application.

[0084] The technical solution of the present application will be explained below in conjunction with specific embodiments.

[0085] Combined with Figure 1 , a sound quality evaluation method based on CEEMD and CWD is provided, including the following steps:

[0086] S1: Collect n noise signal samples, denoted as x i (N) (i = 1, 2,..., n), and subjectively evaluate the noise signal samples respectively;

[0087] S2: Perform CEEMD decomposition on the noise signal samples to obtain several orders of IMF components of each of the noise signal samples, and denote the IMF components as set B i (i = 1, 2,..., n);

[0088] The above S2 specifically includes the following steps:

[0089] S21: Add a pair of positive and negative white noises to the noise signal sample x i (N)

[0090]

[0091] In the formula, n(N) is the white noise component, are the signals after adding positive and negative white noises to the noise sample signal respectively;

[0092] S22: Perform EMD decomposition on respectively, and the decomposition process is as follows:

[0093] S221: Let a certain time series be x(N), determine all its maximum and minimum points, and use cubic spline functions to interpolate the maximum and minimum points to form the upper envelope curve u1(N) and the lower envelope curve u2(N) of the data respectively, and take the average value m1(N) of u1(N) and u2(N), that is

[0094] m1(N) = (u1(N) + u2(N)) / 2

[0095] S222: Let h1(N) = x(N) - m1(N). If h1(N) simultaneously satisfies the two conditions of the IMF: ① Throughout the time history, the number of times crossing zero is equal to or at most one different from the number of extreme points; ② Locally symmetric about the time axis, that is, the mean value of the upper envelope defined by local maxima and the lower envelope defined by local minima is 0, then h1(N) is the first-order IMF. If not satisfied, regard h1(N) as the new x(N), and m 11 (N) is the mean value of its upper and lower envelopes, and there is h 11 (N) = h1(N) - m 11 (N). If h 11 (N) still does not satisfy, repeat this process k times to obtain

[0096] h 1k (N) = h 1(k-1) (N) - m 1k (N)

[0097] If h 1k (N) and the standard deviation SD of h 1(k-1) (N) are between 0.2 and 0.3, then stop repeating this process. At this time, h 1k (N) is the first-order IMF of the signal x(N), denoted as c1(N) = h 1k (N), where the standard deviation SD calculation formula is:

[0098]

[0099] In the formula, N is the sequence length of the signal.

[0100] S223: Let r1(N) = x(N) - c1(N), perform step S222 on r1(N), and successively obtain c2(N), c3(N) …… until r i (N) is a monotonic function and can no longer be separated into IMFs;

[0101] After EMD decomposition, can be expressed as

[0102]

[0103] In the formula, l is the total number of IMF components, are respectively the jth IMF components after adding positive and negative white noises to the noise signal sample x i (N);

[0104] S23: Add M different white noises, and perform step S21 and step S22 each time. The final IMF component of the noise sample signal x i (N) is

[0105]

[0106] In the formula, IMF ij is the noise signal sample x i (N) The final j-th IMF component, are respectively x i (N) The j-th IMF component after adding positive and negative white noises for the k-th time;

[0107] Denote the final IMF component of x i (N) as set B i .

[0108] S3: Calculate the sample entropy values of each order of IMF in the set B i respectively, screen the IMF components according to the sample entropy values, eliminate the false components, and retain the effective components, denoted as set B i '(i = 1, 2,..., n);

[0109] The above S3 specifically includes the following steps:

[0110] S31: The calculation process of calculating the sample entropy values of each order of IMF in the set Bi is as follows:

[0111] S311: Let a certain time series be E(N) = {e(1), e(2),..., e(N)}, and arrange the time series in the order of m-dimensional vectors to obtain the following vectors:

[0112] E m (i) = [e(i), e(i + 1),..., e(i + m - 1)] (1 ≤ i ≤ N - m + 1)

[0113] The vector represents m consecutive e values starting from point i;

[0114] S312: Define the distance m between E m (i) and E as the absolute value of the maximum difference among the elements they contain, then

[0115]

[0116] where k = 0, 1, 2,..., m - 1;

[0117] S313: Count the number A m of the distances m between E less than or equal to R, then i

[0118]

[0119] In the formula, R represents a set threshold value;

[0120] S314: Obtain the mean value of A i The calculation formula is as follows

[0121]

[0122] S315: When calculating to the (m + 1)-dimensional, repeat steps S311 - S313, then

[0123]

[0124] Then the sample entropy SE is

[0125]

[0126] However, under actual conditions, N cannot be infinite, so

[0127]

[0128] m takes 2 or 1, and the value range of R is 0.1SD' - 0.25SD' (SD' is the standard deviation of the time series E(N) = {e(1), e(2),..., e(N)});

[0129] S32: Screen the IMF components according to the sample entropy value calculated in S31, remove the IMF components whose sample entropy value approaches 0, and retain the remaining IMF components;

[0130] S33: Perform steps S31 and S32 on each order of IMF component in the set B i The IMF set after removing the false components is respectively denoted as B i '.

[0131] S4: Perform CWD analysis on each order of IMF in each of the said sets Bi';

[0132] The above S4 specifically includes the following steps:

[0133] S41: Respectively perform CWD analysis on each order of IMF component in the set B i ', and the calculation method is as follows:

[0134]

[0135] In the formula, σ is a constant. If the amplitude and frequency of the signal change rapidly, σ should be taken slightly larger; conversely, if the amplitude and frequency of the signal are relatively gentle, σ should be slightly smaller. Generally, the value of σ should be between 0.1 and 10;

[0136] S42: Perform step S41 on the effective IMF components of each of the said noise signal samples respectively.

[0137] S5: Calculate the SQDF values of each order of IMF in each of the said sets B i ' and the total SQDF value of each of the said noise signal samples;

[0138] The above S5 specifically includes the following steps:

[0139] S51: Calculate the SQDF values of the IMF components of each order in the said set B i ' respectively, and the calculation method is as follows:

[0140]

[0141] In the formula, SQDF IMF-g is the SQDF value of the g-th order IMF component, w p×q is the amplitude matrix after CWD, w p×q (a, b) is the element value of the a-th row and the b-th column of the amplitude matrix w p×q , p is half of the number of spectral lines of the noise signal sample, q is the number of time-domain sampling points, and f max is the maximum frequency for analysis;

[0142] S52: Calculate the total SQDF values of the said noise signal samples respectively, and the calculation method is as follows:

[0143]

[0144] In the formula, SQDF total is the total SQDF value of the noise signal sample, and L is the number of its effective IMF components.

[0145] S53: Perform steps S51 and S52 on each of the said noise signal samples respectively, and calculate the total SQDF value of each of the said noise signal samples.

[0146] S6: Take the total SQDF value of each of the said noise signal samples as the characteristic parameter corresponding to the said noise signal sample, and establish the relationship between the characteristic parameter and the subjective evaluation value of each of the said noise signal samples by using ELM;

[0147] The above S6 specifically includes the following steps:

[0148] S61: Randomly select a part (recommended 70%-80%) of the said noise signal samples as the samples for ELM training, take the total SQDF value of the selected said noise signal samples as the input layer, and the subjective evaluation value as the output layer, and use the ELM network for training to establish the relationship between the total SQDF value and the subjective evaluation value of each of the said noise signal samples;

[0149] S62: Use the remaining noise signal samples to verify the trained ELM network. If the standard deviation SD” of the difference between the sound quality evaluation value and the actual subjective evaluation value of each noise signal sample is not greater than 0.001, it is considered that the ELM network in step S61 is qualified; otherwise, it is considered unqualified, and step S61 is continued until the requirements are met.

[0150] S7: Calculate the SQDF value of the noise to be evaluated, and conduct sound quality evaluation using the corresponding relationship established in step S6.

[0151] Example 1

[0152] This example provides a sound quality evaluation method based on CEEMD and CWD, including the following steps:

[0153] Step 1: Collect the intake noise of a certain long-nose truck. In this experiment, a total of 8 vehicles' right-side intake port noise data were collected. Three groups of data were collected for each sample vehicle, and finally a total of 24 noise signal samples were obtained. Twenty hearing-normal evaluators were selected to conduct sound quality evaluations on each noise signal sample. The grade scoring method was used for this evaluation. A 10-point scale was used to score according to the annoyance degree of the noise. Finally, the average value of the subjective evaluations of each noise signal sample was obtained as their respective final subjective evaluation scores. The subjective evaluation results of these 24 noise signal samples are shown in Table 1;

[0154] Table 1 Subjective evaluation results of noise signal samples

[0155] Sample number Subjective evaluation value Sample number Subjective evaluation value Sample number Subjective evaluation value 1 5.65 9 6.00 17 6.50 2 5.75 10 5.95 18 7.00 3 5.70 11 5.76 19 7.15 4 5.95 12 6.15 20 6.75 5 5.85 13 6.20 21 7.00 6 5.90 14 6.25 22 7.20 7 5.65 15 6.40 23 6.75 8 5.65 16 6.60 24 6.61

[0156] Step 2: Conduct CEEMD on the 24 noise signal samples respectively to obtain their respective several-order IMF components. Taking sample 1 as an example, 6-order IMF components were decomposed, Figure 4 The time-domain waveform diagrams and frequency spectra of these 6-order IMF components are given;

[0157] Step 3: Calculate the sample entropy values of the respective IMF components of the 24 noise signal samples respectively, and calculate the arithmetic mean of the sample entropy values of all IMF components of each noise signal sample respectively. Eliminate the IMF components whose sample entropy values are greater than the arithmetic mean, and retain the other IMF components as effective components. Taking sample 1 as an example, the sample entropy values of its 6-order IMF components and their arithmetic mean are shown in Table 2. Eliminate the last three-order IMF components and retain the first three-order IMF components;

[0158] Table 2 Sample entropy values of each order IMF components of sample 1

[0159]

[0160] Step 4: Perform CWD analysis on the effective IMF components of each noise signal sample to obtain its time-frequency distribution. Taking sample 1 as an example, the CWD analysis result of the 1st order IMF component is as Figure 5 shown;

[0161] Step 5: Calculate the total SQDF value of each noise signal sample. Taking sample 1 as an example, the SQDF values of each order of effective IMF components are shown in Table 3, and the total SQDF value of this sample is 44.92;

[0162] Table 3 SQDF values of effective IMF components of sample 1

[0163] Effective IMF component SQDF value Effective IMF component SQDF value First order 3.68E-4 Third order 40.17 Second order 4.75

[0164] Step 6: Randomly select a part (recommended 70%-80%) of the noise signal samples as the training samples of the ELM network. Take the total SQDF value of each training sample as the input layer and the subjective evaluation value as the output layer, and use the ELM network to establish the relationship between the total SQDF value and the subjective evaluation value. In this embodiment, 17 noise signal samples are selected as the training samples, and the selected sample numbers are 1, 2, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 16, 20, 23, 24;

[0165] Step 7: Randomly select 5 from the remaining 7 noise signal samples in this embodiment to verify the trained ELM network. If the requirements are not met, go to step six until the requirements are met. The finally randomly selected sample numbers are 3, 7, 13, 19, 22 respectively;

[0166] Step 8: In this embodiment, take the finally remaining 2 noise signal samples (i.e., sample 17 and sample 21) as the noises to be evaluated, calculate their total SQDF values respectively, and use the established relationship between the SQDF value and the subjective evaluation value to conduct sound quality evaluation on these two noise samples. The final evaluation results are shown in Table 4.

[0167] Table 4 Comparison table of errors between actual and predicted values of subjective evaluation

[0168] Sample number Actual value of subjective evaluation Predicted value of subjective evaluation Relative error 17 6.50 6.83 5.08％ 21 7.00 7.02 0.29％

[0169] The above-given embodiments are the better examples to implement the present application, and the present application is not limited to the above embodiments. Any non-essential addition or replacement made by those skilled in the art according to the technical features of the technical solution of the present application shall fall within the protection scope of the present application.

Claims

1. A sound quality evaluation method based on CEEMD and CWD, characterized in that, The method includes the following steps: S1: Collect n noise signal samples, denoted as x i (N) respectively, and subjectively evaluate the noise signal samples; S2: Perform CEEMD decomposition on the noise signal samples to obtain several orders of IMF components of each of the noise signal samples, and denote the IMF components as set B respectively i ; S3: Calculate the sample entropy values of each order IMF in the set B respectively, screen the IMF components according to the sample entropy values, eliminate the false components, and retain the effective components, denoted as set B i ; screen the IMF components according to the sample entropy values, eliminate the false components, and retain the effective components, denoted as set B i '; S4: Perform CWD analysis on each order IMF in each of the sets B i '; S5: Calculate each of the sets B i 's SQDF values of each order IMF and the total SQDF value of each of the noise signal samples; S6: Taking the total SQDF value of each of the noise signal samples as the characteristic parameter of the corresponding noise signal sample, and establishing the relationship between the characteristic parameter and the subjective evaluation value of each of the noise signal samples by using ELM; S7: Calculating the SQDF value of the noise to be evaluated, and performing sound quality evaluation by using the corresponding relationship established in step S6; In S1, S2, and S3, i = 1, 2,..., n; The specific steps of S5 are as follows: S51: Calculate each order IMF component's SQDF value in the set B i ' respectively, and the calculation method is as follows: where SQDF IMF-g is the SQDF value of the g-th order IMF component, w p×q is the amplitude matrix after CWD, w p×q (a, b) is the element value of the a-th row and the b-th column of the amplitude matrix w p×q , p is half of the number of sample spectral lines of the noise signal, q is the number of time-domain sampling points, f max is the maximum frequency for analysis; S52: Respectively calculating the total SQDF value of the noise signal samples, and the calculation method is as follows: where SQDF total is the total SQDF value of the noise signal samples, and L is the number of its effective IMF components; S53: Respectively performing steps S51 and S52 on each of the noise signal samples to calculate the total SQDF value of each of the noise signal samples.

2. The sound quality evaluation method based on CEEMD and CWD according to claim 1, characterized in that, The specific steps of S2 are as follows: S21: Add a pair of positive and negative white noises to the noise signal sample x i (N) where n(N) is the white noise component, are the signals after adding positive and negative white noises to the noise sample signal, respectively; S22: Respectively perform EMD decomposition on The decomposition process is as follows: S221: Let a certain time series be x(N), determine all its maximum points and minimum points, respectively form the upper envelope curve u1(N) and the lower envelope curve u2(N) of the data by interpolating the maximum points and the minimum points with a cubic spline function, and take the average value m1(N) of u1(N) and u2(N), that is m1(N) = (u1(N) + u2(N)) / 2 S222: Let h1(N) = x(N) - m1(N). If h1(N) simultaneously satisfies the two conditions of the IMF: ① within the entire time course, the number of times crossing the zero point is equal to or at most differs by 1 from the number of extreme points; ② it is locally symmetric about the time axis, that is, the mean value of the upper envelope defined by the local maximum value and the lower envelope defined by the local minimum value is 0, then h1(N) is the first-order IMF. If not satisfied, then regard h1(N) as the new x(N), and m 11 (N) is the mean value of its upper and lower envelope lines, and there is h 11 (N) = h1(N) - m 11 (N). If h 11 (N) still does not satisfy, repeat this process k times to obtain h 1k (N) = h 1(k-1) (N) - m 1k (N) If h 1k (N) and h 1(k-1) (N) have a standard deviation SD between 0.2 and 0.3, then stop repeating the process. At this time, h 1k (N) is the first-order IMF of the signal x(N), denoted as c1(N) = h 1k (N), where the standard deviation SD calculation formula is: In the formula, N is the sequence length of the signal; S223: Let \(r_1(N)=x(N)-c_1(N)\), and perform step S222 on \(r_1(N)\) to successively obtain \(c_2(N)\), \(c_3(N)\), \(\cdots\), until \(r i (N)\) is a monotonic function and no more IMFs can be separated; After EMD decomposition, can be expressed as where \(l\) is the total number of IMF components, are the noise signal samples \(x\) i (N) the \(j\)-th IMF component after adding positive and negative white noises; S23: Add white noise M times differently, and perform step S21 and step S22 each time. The noise sample signal x i (N) The final IMF components are where, IMF ij is the noise signal sample x i (N) the final j-th IMF component, are respectively x i (N) the j-th IMF component after adding positive and negative white noises for the k-th time, k = 1, 2, …, M; Let x i (N) The final IMF components are denoted as set B i .

3. The method for evaluating sound quality based on CEEMD and CWD according to claim 1, wherein, The specific steps of S3 are as follows: S31: The calculation process of respectively calculating the sample entropy value of each order IMF in the set Bi is as follows: S311: Let a certain time series be E(N) = {e(1), e(2),..., e(N)}, and arrange the time series in the order of m-dimensional vectors to obtain the following vectors: E m (i) = [e(i), e(i + 1), …, e(i + m - 1)] The vector represents m consecutive e values starting from the i-th point, where 1 ≤ i ≤ N - m + 1; S312: Define E m (i) The distance between m (j) and E is the absolute value of the maximum difference between the elements contained in both, then where k = 0, 1, 2,..., m - 1; S313: Count E m (i) The distance between m (j) and E Less than or equal to the number A of R i , then In the formula, R represents a set threshold; S314: Obtain the average value of A i The average value The calculation formula is S315: When calculating to the m + 1 dimension, repeat steps S311 - S313, then Then the sample entropy SE is And under actual conditions, N cannot be infinite, so m takes 2 or 1, and the value range of R is 0.1SD' to 0.25SD', where SD' is the standard deviation of the time series E(N) = {e(1), e(2),..., e(N)}; S32: Screening the IMF components according to the sample entropy value calculated in S31, removing the IMF components whose sample entropy values approach 0, and retaining the remaining IMF components; S33: For each IMF component in set B i perform steps S31 and S32, and denote the IMF sets with spurious components removed as B i '.

4. The method for evaluating sound quality based on CEEMD and CWD according to claim 1, wherein, The specific steps of S4 are as follows: S41: Perform CWD analysis on each order IMF component in the set B i ' respectively, and the calculation method is as follows: In the formula, σ is a constant, and the value of σ is between 0.1 and 10; S42: Respectively performing step S41 on the effective IMF components of each of the noise signal samples.

5. The method for evaluating sound quality based on CEEMD and CWD according to claim 1, wherein, The specific steps of S6 are as follows: S61: Randomly selecting a part of the noise signal samples as the samples for ELM training, taking the total SQDF value of the selected noise signal samples as the input layer and the subjective evaluation value as the output layer, and training by using the ELM network to establish the relationship between the total SQDF value and the subjective evaluation value of each of the noise signal samples. S62: Use the remaining noise signal samples to verify the trained ELM network. If the standard deviation SD” of the difference between the sound quality evaluation value of each noise signal sample and the actual subjective evaluation value is not greater than 0.001, it is considered that the ELM network in step S61 is qualified; otherwise, it is considered unqualified, and step S61 is continued until the requirements are met.

6. The method for evaluating sound quality based on CEEMD and CWD according to claim 5, wherein, The amount of the randomly selected noise signal samples is 70%-80%.

Images