CN113343384A - Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission - Google Patents

Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission Download PDF

Info

Publication number
CN113343384A
CN113343384A CN202110609285.9A CN202110609285A CN113343384A CN 113343384 A CN113343384 A CN 113343384A CN 202110609285 A CN202110609285 A CN 202110609285A CN 113343384 A CN113343384 A CN 113343384A
Authority
CN
China
Prior art keywords
subjective
listening
samples
sound quality
transmission
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110609285.9A
Other languages
Chinese (zh)
Other versions
CN113343384B (en
Inventor
王琇峰
杜磊磊
罗堃
区瑞坚
郭美娜
王继承
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou Veizu Equipment Diagnosis Technology Co ltd
Xian Jiaotong University
Original Assignee
Suzhou Veizu Equipment Diagnosis Technology Co ltd
Xian Jiaotong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Suzhou Veizu Equipment Diagnosis Technology Co ltd, Xian Jiaotong University filed Critical Suzhou Veizu Equipment Diagnosis Technology Co ltd
Priority to CN202110609285.9A priority Critical patent/CN113343384B/en
Publication of CN113343384A publication Critical patent/CN113343384A/en
Application granted granted Critical
Publication of CN113343384B publication Critical patent/CN113343384B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • G06F30/27Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01MTESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00Testing of machine parts
    • G01M13/02Gearings; Transmission mechanisms
    • G01M13/028Acoustic or vibration analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/23Clustering techniques
    • G06F18/232Non-hierarchical techniques
    • G06F18/2321Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/10Geometric CAD
    • G06F30/17Mechanical parametric or variational design
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/06Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063Operations research, analysis or management
    • G06Q10/0639Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06395Quality analysis or management
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/10Noise analysis or noise optimisation
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02TCLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • Evolutionary Computation (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Strategic Management (AREA)
  • Evolutionary Biology (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Educational Administration (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Hardware Design (AREA)
  • Development Economics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Game Theory and Decision Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

A sound quality subjective and objective evaluation method under the variable-speed working condition of a transmission comprises the steps of firstly collecting noise and a rotating speed signal under the variable-speed working condition of the transmission, carrying out sliding window interception on the noise signal to generate a preliminary listening sample set, carrying out sound quality psychoacoustic index calculation on the sample, and carrying out cluster analysis; then, forming a listening sample set according to the clustering analysis result, and adding an expert to manually select samples to form a formal listening sample group; grouping the appraisers, endowing group weights, and selecting appraising words by experts; then, the formal listening sample group is subjectively evaluated, the evaluation data is subjected to statistical test and added with weight, and statistical analysis is carried out to obtain a subjective evaluation label; then establishing a transmission sound quality subjective and objective evaluation model, and verifying the model precision; according to the method, the sound sample selection is not careless, the listening sample size is minimized, the reflected sound quality attribute difference is maximized, the subjective evaluation result accords with the statistical rule, and the objective evaluation reflects the sound quality characteristic more accurately.

Description

Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission
Technical Field
The invention belongs to the technical field of sound quality evaluation of transmissions, and particularly relates to an objective sound quality evaluation method under a variable-speed working condition of a transmission.
Background
With the continuous development of the vehicle industry, the research of the automobile technology gradually changes from performance requirement to quality development, and the vehicle is not only a transportation tool, but also develops towards living space. Noise is one of main factors influencing the comfort of a driver, and the development of sound quality psychoacoustics ensures that the sound quality characteristic of the noise is also adjusted while the sound pressure level of the noise is reduced, so that the subjective feeling of the driver is more comfortable.
Under the technical background of popularization of electric vehicles, effective control of vehicle noise sources such as engines and the like and high speed and light weight at present, the problem of noise of a transmission becomes prominent and becomes one of main links influencing the sound quality of the whole vehicle. The evaluation method taking the A weighting sound pressure level as the main evaluation standard has the advantage that the influence of the A weighting sound pressure level on the auditory sense of human ears is obviously reduced after the means for testing and controlling the A weighting sound pressure level is improved. With the gradual improvement of living standard, the requirement on the comfort of automobile driving is higher and higher, so that the sound control is no longer as small as possible, and the subjective perception is comfortable and pleasant.
The transmission sound quality evaluation comprises a subjective evaluation part and an objective evaluation part, wherein the subjective evaluation method is that an evaluator carries out a subjective evaluation experiment on transmission sound, and then a statistical method is applied to obtain a subjective evaluation label. The objective evaluation method is to fit the subjective evaluation label by using the acoustic quality psychoacoustic index, and the objective evaluation method is used for evaluating the acoustic quality by using a computer instead of human ears.
The existing transmission sound quality subjective and objective evaluation method has the following defects:
1. the interception of the sound quality evaluation sample of the variable-working-condition transmission has no scientific basis, and if the acquired noise signal is completely intercepted without omission, the audition sample amount is too much, and the subjective evaluation cannot be realized; if the signals are intercepted at intervals, noise signals in part of time periods can be omitted, so that the listening samples cannot cover the conditions of different sound qualities of the variable-speed transmission;
2. the listening samples are not effectively simplified, the sound evaluation times are repeated when the listening samples with the same sound quality are listened, the listening experiment is complicated, and the listening time is too long;
3. the evaluation index of subjective evaluators for evaluating the data validity is single, and the evaluation data validity detection method is incomplete;
4. the subjective evaluation dimension is single, and various sound characteristics of the sound quality of the transmission cannot be reflected;
5. the objective evaluation index is inaccurate in the subjective evaluation fitting result of the variable-condition transmission sound quality.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide the subjective and objective evaluation method of the sound quality under the variable-speed working condition of the transmission.
In order to achieve the purpose, the invention adopts the technical scheme that:
a sound quality subjective and objective evaluation method under the variable-speed working condition of a transmission comprises the following steps:
step 1: collecting noise signals of the transmission under the condition of changing the rotating speed of M gears by using a microphone, and collecting rotating speed signals of the transmission by using a photoelectric sensor;
step 2: performing sliding window interception on noise signals under M gears of the transmission to generate preliminary listening sample sets of the M gears, wherein each preliminary listening sample set comprises E listening samples, and the total number of the preliminary listening samples is M multiplied by E listening samples;
and step 3: respectively calculating sound quality psychoacoustic indexes of the E listening samples contained in each preliminary listening sample set in the step 2, and performing cluster analysis according to the calculation results of the sound quality psychoacoustic indexes to finally form M x c preliminary listening samples;
the acoustic quality psychoacoustic indexes include loudness, sharpness, roughness and jitter degree, but are not limited to the four acoustic quality psychoacoustic indexes; the clustering analysis method adopts a K-means clustering analysis method;
and 4, step 4: for the M × c preliminary listening samples in the step 3, reserving listening samples closest to the clustering center point in each category to form a listening sample set with the sample capacity of M × c;
and 5: adding H samples which are different in the two groups and are the same in the groups after the 1 st group and the M group of the listening sample set in the step 4 to serve as evaluation personnel reliability evaluation samples, and adding samples manually selected by NVH experts from the 2 nd group to the M-1 st group to keep the number of listening samples of each group consistent; the listening samples are arranged in sequence according to the corresponding actual rotating speed value of the transmission to form a formal listening sample group with the sample capacity of Mx (c + H);
step 6: grouping the evaluators, and adding different group weights between each group;
and 7: selecting semantic antisense words capable of reflecting the sound characteristics of the transmission to be evaluated from a transmission sound quality semantic word bank by NVH experts;
and 8: subjective evaluation is carried out on the formal listening sample group in the step 5, the semantic antisense words are adopted as the subjective evaluation words in the step 7, and after the subjective evaluation, the subjective evaluation result of each person on the listening samples in the formal listening sample group is obtained;
and step 9: carrying out listening reliability analysis on the appraisers, and adding reliability weight to subjective evaluation results of the appraisers;
step 10: performing statistical analysis on the Spearman correlation coefficient and Euclidean distance indexes on the subjective evaluation result in the step 8, eliminating evaluation personnel with inaccurate scores, adding the group weight in the step 6 and the reliability weight in the step 9 to the score data of the rest personnel after elimination, and calculating to obtain a subjective evaluation label;
step 11: establishing a transmission sound quality subjective and objective evaluation model;
calculating the acoustic quality psychoacoustic indexes of the formal listening sample component frames in the step 5, dividing a test set and a training set, and establishing a transmission acoustic quality subjective and objective evaluation model by adopting a support vector regression method according to the acoustic quality psychoacoustic index calculation results of the listening samples in the training set and the subjective evaluation labels in the step 10;
step 12: and (3) verifying the transmission sound quality subjective and objective evaluation model:
and (3) verifying the transmission sound quality subjective and objective evaluation model established in the step 11 by adopting the test set in the step 11, taking a Pearson correlation coefficient and an average absolute error as evaluation indexes, if the Pearson correlation coefficient is greater than 0.9, and the average absolute error is less than 10% of the maximum value of the grading interval, indicating that the transmission sound quality subjective and objective evaluation model is well predicted, and if the Pearson correlation coefficient is not greater than 0.9, returning to the step 6, and re-performing the sound quality subjective and objective evaluation.
In the step 1, a microphone is used for collecting noise signals under the variable-speed working condition of M gears of the transmission, the collection environment is a semi-anechoic chamber, the microphone is arranged at a position 1M away from the center of the transmission from a loading end to the left side, the position is a sound field far field, no interference effect exists among sound waves, the sound source is approximately regarded as a point sound source, the actual noise evaluation requirement is met, the sampling frequency is not lower than 40960Hz, the analysis frequency is not lower than 20480Hz and higher than the upper limit frequency of human hearing according to the Nyquist theorem, and each gear of the transmission is driven to rotate at the highest speed rpm under the maximum loadmaxUniformly decelerated to the lowest rpmminAnd collecting noise signals, wherein the collection time of each gear is S seconds.
In the step 2, noise signals under M gears of the transmission are intercepted by a sliding window, a window function of the sliding window is selected as a rectangular window, the length of the noise signal under each gear is set to be N, the window length is wlen, the sliding displacement of the latter window relative to the former window is wst, wherein the time length range of wlen is 5s or more and less than wlen and less than or equal to 10s, the time length range of wst is 0s or more and less than wst or less than or equal to wlen, and the calculation formula of the number N of sample sections intercepted by the sliding window is as follows:
Figure BDA0003095335920000051
the method for calculating the psychoacoustic index of sound quality in the step 3 comprises the following steps:
3.1) loudness calculation:
the specific loudness is calculated using the following formula:
Figure BDA0003095335920000052
of formula (II) to (III)'0To reference specific loudness, ETQFor the corresponding excitation in the quiet state, srIs the ratio of the sound intensity of a just-audible test tone to that of a broadband noise at the same critical band, E0Is sound intensity I0=10-12W/m2Corresponding reference excitation value, EsIs the excitation to which the sound corresponds, when N'0When 0.065, 0.25 s is selected as θr0.25; is N'0When equal to 0.08, take θ equal to 0.23, sr0.5; the Bark band division standard adopts a Zwicker model Bark band division standard;
the total loudness is obtained by integrating the specific loudness over the 0-24Bark scale:
Figure BDA0003095335920000061
3.2) sharpness calculation:
the Zwicker sharpness model is based on a loudness model, and the mathematical model is as follows:
Figure BDA0003095335920000062
in the formula, K is a weighting coefficient, and K is 0.11; ssharpnessRepresenting sharpness, and N' (z) representing specific loudness in Bark domain z, where g (z) is the weight coefficient of the sound signal in different Bark domains, expressed as:
Figure BDA0003095335920000063
3.3) roughness calculation:
the roughness model after Zwicker improvement is based on a loudness model, and the mathematical model is as follows:
Figure BDA0003095335920000064
wherein Rou is the calculated roughness, fmodTo modulate frequency, Δ LEFor the sound pressure variation amplitude in each critical frequency band, the following is defined:
Figure BDA0003095335920000065
in the formula, Nmax′(z) and Nmin′(z) represents the maximum and minimum values of the characteristic loudness in the Zwicker loudness model, respectively;
3.4) jitter degree calculation:
the Zwicker jitter model is based on a loudness model, and the mathematical model is as follows:
Figure BDA0003095335920000066
in the formula,. DELTA.LEThe sound pressure change amplitude in each critical frequency band is taken as the sound pressure change amplitude; f. ofmodIs the modulation frequency; f. of0Is to modulate the fundamental frequency of the signal,f0=4Hz。
the clustering analysis and calculation process in the step 3 is as follows:
(1) carrying out sliding window interception according to the noise signal under each gear, wherein the number of the sound sample sections intercepted by each gear sliding window is n, and an input sample is Q ═ x1,x2,...,xnRandomly selecting c sound samples from Q as centroid samples u1,u2,...,ucThe number c of the centroids of the clusters is determined by the maximum value of the scoring interval, and if the scoring is 0-10, c is 10;
(2) for input sample x1,x2,...,xnMeasuring the distance to the centroid and classifying it as a class to the nearest centroid;
(3) updating the center of each class to be the mean of all samples belonging to the class;
(4) repeating steps (2) and (3) until less than 3 samples are reassigned to different clusters;
(5) and after clustering is finished, the noise signals intercepted by the sliding windows of each gear are gathered into c-type sound samples.
And in the step 5, the chrominance evaluation sample is a sound sample selected by a large amount of auditions of an expert, wherein the number of the large amount of auditions is not less than the number of the centralized audition samples of the audition samples in the step 4.
In the step 5, the acoustic samples are arranged in sequence according to the corresponding actual rotating speed value of the gearbox, and the sequence of the acoustic samples is as follows: the first sample of each group is the intercepted sample corresponding to the highest rotating speed section, the second sample is the intercepted sample corresponding to the lowest rotating speed section, other samples are arranged according to the sequence of the corresponding rotating speed sections from large to small or from small to large, a 5s blank space is reserved between the two listening samples, and time is scored for evaluators.
In the step 6, the evaluators are grouped, and different group weights are added between each group, namely the evaluators with different listening levels are divided into expert groups, experience groups and common groups, and different group weights W are respectively given to the evaluatorsG
In the step 9, listening reliability analysis is carried out on the appraisers, and the appraisers are subjected to listening reliability analysisAttaching a reliability weight W to the subjective evaluation result of (1)TThe method comprises the following specific steps:
9.1) extracting the scoring data of 2 XH credibility evaluation samples by an evaluator, if all samples are scored consistently, the misjudgment rate is 0, if the samples with inconsistent scoring have delta, and the difference between the two scoring is not more than 10% of the maximum value of the scoring interval, calculating the acceptable misjudgment rate as
Figure BDA0003095335920000081
If b samples with inconsistent scores exist and the difference between the two scores exceeds 10% of the maximum value of the scoring interval, calculating the unacceptable misjudgment rate as
Figure BDA0003095335920000082
9.2) according to the acceptable false positive rate PyAnd unacceptable false positive rate PnForming a confidence weight WTThe calculation formula is as follows:
WT=1-Pn-λPy
wherein, the value of lambda is between 0 and 1, if the acceptable misjudgment is not allowed, the lambda is 1; if the acceptable misjudgment is completely allowed, λ is 0.
The statistical analysis and calculation steps of the Spearman correlation coefficient and the Euclidean distance in the step 10 are as follows:
10.1) calculating the Spearman correlation coefficient between each two evaluators:
Figure BDA0003095335920000083
in the formula: d is the difference between the subjective evaluation result grades of the two lines; r is the length of two lines of subjective evaluation results; ri,jA Spearman correlation coefficient representing the subjective evaluation result of the ith evaluator to the subjective evaluation result of the jth evaluator;
10.2) taking the average correlation coefficient:
Figure BDA0003095335920000091
in the formula: k is the number of evaluators;
Figure BDA0003095335920000092
the correlation coefficient between the ith evaluator and the jth evaluator is obtained; riThe average correlation coefficient of the ith evaluator relative to the other evaluators in the panel; set RiThe threshold value is 0.75 if RiIf the correlation is less than 0.75, the correlation of the evaluator is not high relative to other evaluators, the subjective listening sensation is greatly deviated, and the scoring data of the evaluator is removed;
10.3) averaging the scoring data of the evaluators retained after the rejection, wherein the average scoring value of each sample is obtained by the following calculation formula;
Figure BDA0003095335920000093
in the formula: k' is the number of remaining evaluators after rejection, Vi,aSubjective scoring value of the ith evaluator on the a-th sample; vaAverage scores for all raters for a samples;
10.4) score data V for each of the remaining ratersi,aAnd VaThe Euclidean distance statistical analysis and calculation method comprises the following steps:
Figure BDA0003095335920000094
wherein i is the number of evaluators;
eliminating residual evaluator correspondence D (V)i,a,Va) The larger evaluation data requires that the Spearman correlation coefficient and Euclidean distance statistical analysis rejecting population is not more than 20% of the total evaluation population.
The step 11 of establishing the transmission sound quality subjective and objective evaluation model comprises the following specific steps:
11.1) taking 70% of the formal listening samples of the listening sample group as a training set for training the transmission sound quality subjective and objective evaluation model; 30% of the test set is used for testing the transmission sound quality subjective and objective evaluation model;
11.2) setting frame length and frame shift for all listening samples, wherein the frame length time length interval is (0,1] second, the frame shift time length interval is (0,1] second, and the frames are divided into f sections;
11.3) calculating the psychoacoustic index of sound quality of each frame of the listening sample;
11.4) using the acoustic quality psychoacoustic indexes calculated by the training set in frames and the subjective evaluation labels in the step 10 as parameters for training an acoustic quality subjective and objective evaluation model of the transmission;
11.5) training a transmission sound quality subjective and objective evaluation model, and establishing the transmission sound quality subjective and objective evaluation model;
in the step 11, a support vector regression method is used for carrying out transmission sound quality subjective and objective evaluation model fitting, an RBF kernel function is selected, an insensitive loss function takes epsilon as 0.01, an optimal penalty parameter e and a kernel function parameter g are selected by adopting a K-CV (K-fold Cross Validation) Cross Validation method, the lowest mean square error mse in the Cross Validation process is taken as an optimization target function, the Cross Validation parameter v is selected as 3, and the mse formula is calculated as follows:
Figure BDA0003095335920000101
in the formula: v is the number of cross validation groups, nl is the number of cross validation groups, yijIn order to obtain the true label of the sample,
Figure BDA0003095335920000102
a transmission sound quality subjective and objective evaluation model prediction label;
using a grid search method, a rough selection is first performed, taking a log base 22e、log2g, the value ranges are [ -8,8 respectively]、[-8,8]The step size of the penalty parameter e and the kernel function parameter g is both 1; fine selection is carried out again according to the coarse selection result, and the log with the base 2 is continuously taken2e、log2g,The value ranges are [ -4,4 respectively]、[-4,4]The step sizes of the penalty parameter e and the kernel function parameter g are both 0.1.
The calculation formula of the Pearson correlation coefficient ρ in step 12 is as follows:
Figure BDA0003095335920000111
in the formula: y is the test set subjective rating label,
Figure BDA0003095335920000112
subjective evaluation label prediction values of the transmission sound quality subjective and objective evaluation model on the test set,
Figure BDA0003095335920000113
is y and
Figure BDA0003095335920000114
covariance between, σyIs the mean square error of y, μyIs the mean value of y, EλRepresents a mathematical expectation;
mean absolute error MAE is less than maximum value L of scoring interval max10% of MAE, the MAE calculation formula is as follows:
Figure BDA0003095335920000115
in the formula:
Figure BDA0003095335920000116
is the predicted value of the ith listening sample, yiIs a subjective evaluation label, nsamplesIs the total number of listening samples;
percentage of error WmaeThe calculation formula is as follows:
Figure BDA0003095335920000117
in the formula:
Figure BDA0003095335920000118
is the average absolute error between the predicted value of the subjective evaluation label and the subjective evaluation label of the test set, LmaxIs the maximum value of the scoring interval.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention intercepts the sound quality evaluation sample under the variable working condition of the transmission, adopts a sliding window intercepting method, and intercepts all time periods, so that the sound sample under each working condition of the transmission is selected without leakage.
2. In the invention, in the selection of the subjective sound quality evaluation sample, the sample with large difference of sound quality psychoacoustic indexes is selected from a large number of samples generated after the interception of the sliding window as the subjective evaluation sample, so that the audition sample amount is minimized and the reflection of the sound quality attribute difference is maximized.
3. The invention provides a subjective appraiser weight assignment method, which comprises appraiser reliability weight and group weight, so that the scoring data of professional appraisers with high appraisal reliability account for a larger proportion.
4. The invention provides a subjective evaluation result statistical method, which enables subjective evaluation results to better accord with statistical rules.
5. The method carries out frame interception on the listening samples, calculates the psychoacoustic index of sound quality of each frame, and fits the subjective evaluation label with the calculation result by using the support vector regression model, and the support vector regression model has high generalization capability and can extract the nonlinear relation between the frames, so that objective evaluation and reflection of sound quality characteristics are more accurate.
6. The invention provides a transmission sound quality subjective and objective evaluation model inspection method, which enables the transmission sound quality subjective and objective evaluation model to have scientific judgment basis for the fitting result precision of subjective and objective evaluation.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a schematic diagram of a sliding window intercepting a noise signal of each gear according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of a formal listening sample according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of calculating a psychoacoustic index of sound quality in frames according to an embodiment of the present invention.
Fig. 5 is a diagram illustrating prediction error of a transmission sound quality subjective and objective evaluation model established according to an embodiment of the present invention with respect to a test set subjective evaluation label.
Fig. 6 is a comparison graph of the transmission sound quality subjective and objective evaluation model established according to the embodiment of the invention, the predicted values of the test set and subjective evaluation labels.
Detailed Description
The present invention will be described in further detail with reference to the following drawings and examples.
Referring to fig. 1, the subjective and objective evaluation method for sound quality under the variable-speed working condition of the transmission comprises the following steps:
step 1: the method comprises the steps of collecting variable-rotating-speed noise signals under 6 gears of a transmission by using a microphone, wherein the collection environment is a semi-anechoic chamber, the microphone is arranged at a position 1 m on the left side (observed from a loading end) away from the center height of the transmission, the position is a sound field far field, no interference effect exists among sound waves, the sound source can be approximately regarded as a point sound source, the actual noise evaluation requirement is met, the sampling frequency is not lower than 40960Hz, the analysis frequency is not lower than 20480Hz and higher than the human hearing upper limit frequency according to the Nyquist theorem, the transmission has 6 gears in total, and the transmission can rotate at the highest rotating speed rpm under the maximum loadmax2600rpm, uniformly decelerated to the lowest rpmmin500rpm, and the acquisition time of each gear is 257 s;
step 2: performing sliding window interception on noise signals under 6 gears of the transmission, as shown in fig. 2, selecting a window function of a sliding window as a rectangular window, acquiring a noise signal under each gear at a length of 257s, acquiring a window length of 5s, and generating a preliminary listening sample set for each gear, where a sliding displacement of a subsequent window relative to a previous window is 2.5s, and a calculation formula of n, which is the number of listening sample sections intercepted by the sliding window, is as follows:
Figure BDA0003095335920000131
a total of 6 preliminary listening sample sets, each preliminary listening sample set comprising 101 listening samples, a total of 6 × 101 listening samples;
and step 3: performing cluster analysis on the sound quality psychoacoustic indexes, such as loudness, sharpness, roughness, jitter and the like, of each listening sample of 101 listening samples contained in each preliminary listening sample set in the step 2 according to the calculated sound quality psychoacoustic indexes;
each preliminary listening sample set divides 101 listening samples into 10 classes, and 6 preliminary listening sample sets are formed in total, and finally, 6 x 10 classes of preliminary listening samples are formed;
the method for calculating the psychoacoustic index of sound quality comprises the following steps:
3.1) loudness calculation:
the specific loudness is calculated using the following formula:
Figure BDA0003095335920000141
of formula (II) to (III)'0To reference specific loudness, ETQFor the corresponding excitation in the quiet state, srIs the ratio of the sound intensity of a just-audible test tone to that of a broadband noise at the same critical band, E0Is sound intensity I0=10-12W/m2Corresponding reference excitation value, EsIs the excitation to which the sound corresponds, when N'0When 0.065, 0.25 s is selected as θr0.25; is N'0When equal to 0.08, take θ equal to 0.23, sr0.5; the Bark band division standard adopts a Zwicker model Bark band division standard;
the total loudness is obtained by integrating the specific loudness over the 0-24Bark scale:
Figure BDA0003095335920000142
3.2) sharpness calculation:
the Zwicker sharpness model is based on a loudness model, and the mathematical model is as follows:
Figure BDA0003095335920000143
in the formula, K is a weighting coefficient, and K is 0.11; ssharpnessRepresenting sharpness, and N' (z) representing specific loudness in Bark domain z, where g (z) is the weight coefficient of the sound signal in different Bark domains, expressed as:
Figure BDA0003095335920000151
3.3) roughness calculation:
the roughness model after Zwicker improvement is based on a loudness model, and the mathematical model is as follows:
Figure BDA0003095335920000152
wherein Rou is the calculated roughness, fmodTo modulate frequency, Δ LEFor the sound pressure variation amplitude in each critical frequency band, the following is defined:
Figure BDA0003095335920000153
in the formula, Nmax′(z) and Nmin′(z) represents the maximum and minimum values of the characteristic loudness in the Zwicker loudness model, respectively;
3.4) jitter degree calculation:
the Zwicker jitter model is based on a loudness model, and the mathematical model is as follows:
Figure BDA0003095335920000154
in the formula,. DELTA.LEThe sound pressure change amplitude in each critical frequency band is taken as the sound pressure change amplitude; f. ofmodIs the modulation frequency; f. of0Is modulating the fundamental frequency, f0=4Hz;
The clustering analysis process is as follows:
(1) carrying out sliding window interception according to the noise signal under each gear, wherein the number of the intercepted sound sample sections of each gear sliding window is 101, and an input sample is set to be Q ═ x1,x2,...,x101Randomly selecting 10 sound samples from Q as centroid samples u1,u2,...,u10The number of centroids of the cluster is 10;
(2) for input sample x1,x2,...,x101Measuring the distance to the centroid and classifying it as a class to the nearest centroid;
(3) updating the center of each class to be the mean of all samples belonging to the class;
(4) repeating steps (2) and (3) until less than 3 samples are reassigned to different clusters;
(5) after clustering is finished, the noise signals intercepted by the sliding windows of each gear are clustered into 10 types of sound samples;
and 4, step 4: for the 6 × 10 preliminary listening samples in the step 3, reserving the listening sample closest to the clustering center point in each category, and generating a listening sample set with the sample capacity of 6 × 10;
and 5: respectively adding 4 listening samples which are different in group and same in group behind the 1 st group and the 6 th group of the listening sample set in the step 4 to serve as evaluation personnel reliability evaluation samples, wherein the reliability evaluation samples are representative sound samples selected by a large number of listening samples of an expert, and the number of the large number of listening samples is not less than that of the listening sample set in the step 4;
adding 4 samples manually selected by NVH experts from the group 2 to the group 5 respectively to keep the number of listening samples in each group consistent, in order to prevent the evaluation staff from having no sense of integral range during scoring, arranging sound samples of a lowest rotating speed section and a highest rotating speed section in samples No. 1 and No. 2 of initial listening, arranging the listening samples according to rotating speed signals collected in the step 1 from large to small (a blank space of 5s is formed between the two listening samples and scoring time is carried out on the evaluation staff) to form a formal listening sample group with the sample capacity of 6 multiplied by 14, wherein the samples of the formal listening sample group form a sample group shown in figure 3;
step 6: the listening experiment evaluator in the embodiment comprises 8 NVH experts, 6 drivers with driving experience more than 7 years and 10 ordinary evaluators, and the evaluation personnel give a group weight WGRespectively 0.5, 0.3 and 0.2, and the information and specific distribution of the evaluators are shown in table 1;
TABLE 1 group of weights
Figure BDA0003095335920000171
And 7: selecting 14 pairs of semantic antisense words describing the noise sound quality characteristics of the transmission as a transmission sound quality semantic word library, and enabling an NVH (noise, vibration and harshness) expert to screen the 14 pairs of semantic antisense words; the semantic antisense words in the table 2 can be scored by a 0-10-level scoring method to reflect the importance degree of the sound quality characteristics of the transmission, so as to determine the semantic antisense words (0 represents extremely unimportant and 10 represents extremely important) which are finally used for evaluating the sound quality of the transmission, and the semantic antisense words describing the sound quality characteristics of the transmission are shown in the table 2;
TABLE 2 semantic antisense words describing transmission acoustic quality characteristics
Numbering Semantic antisense words Numbering Semantic antisense words
1 Gentle of wave motion 8 Sharp-smooth
2 Harsh-pleasing 9 Acutely-bass
3 Vibratile-stable 10 Smooth and abrupt
4 Vibrating the ear-gentle 11 Comfort-discomfort
5 Of impact-moderation 12 Quiet-noisy
6 Humming-calming 13 Undulating-smooth
7 Rolling-quiet 14 Acutely-dull
The results of the NVH expert in scoring the importance degree of the 14 semantic anti-sense words are shown in Table 3;
TABLE 3 expert word-selecting scoring results
Word order number Expert 1 Expert 2 Expert 3 …… …… Expert 8
1 8 8 9 …… …… 9
2 6 7 6 …… …… 5
…… …… …… …… …… …… ……
14 8 9 7 …… …… 7
The 8-bit NVH expert gives statistics to the scoring results of 14 semantic antisense words to be selected, and the statistics are shown in Table 4;
TABLE 4 expert word-selecting scoring statistics
Serial number Word and phrase Mean value of Standard deviation of
1 Comfort-discomfort 9.125 0.7806
2 Acutely-dull 8.75 0.9682
3 Vibrating the ear-gentle 7.5 1
4 Acutely-bass 7.25 0.8291
5 Harsh-pleasing 7.25 1.3919
6 Rolling-quiet 7.125 0.7806
7 Humming-calming 7 1.8027
8 Undulating-smooth 6.875 1.3635
9 Smooth and abrupt 6.875 1.3635
10 Quiet-noisy 6.875 2.5217
11 Sharp-smooth 6.875 1.6153
12 Vibratile-stable 6.625 1.4086
13 Of impact-moderation 6.625 1.4947
14 Gentle of wave motion 6.5 1.4142
Selecting the first A' semantic antisense words with relatively high scores according to evaluation requirements, and only selecting the semantic antisense words with the highest scores in the evaluation: comfort-discomfort as a semantic antisense to transmission sound quality assessment;
and 8: the evaluator officially listens, and carries out subjective evaluation according to a subjective evaluation scoring table shown in table 5; the evaluator subjectively evaluates the sound samples in the formal listening sample group, evaluates the words as comfortable and uncomfortable, and evaluates the score to 0-10, wherein the score is shown in a grade score table in a table 5 corresponding to the subjective feeling;
TABLE 5 subjective evaluation scoring table
Figure BDA0003095335920000191
The evaluation environment is indoor with good ventilation, and the environmental noise is less than 30 dB; the method is characterized in that a special earphone (MDR-Z1000/Q Sony high-fidelity noise-isolation and leakage-prevention recording room professional monitoring earphone and a headset) for the high-fidelity noise-isolation and leakage-prevention recording room is adopted, windows are uniformly used for carrying audio software, the volume is adjusted to be uniform (the loudspeaker is adjusted to be 15, and the audio software is adjusted to be maximum), the same type of earphone, the same type of computer and an onboard sound card are used, and in the same time period (morning), the mental state of an evaluator is required to be good before listening, no disease exists, and the listening process needs to be completed independently;
listening experiment time is 9 am every day, 3 groups of samples are listened for 20-30 minutes, and each group of samples of the sample group is listened for 2 minutes as rest time;
the method comprises the steps that before formal listening, an integral audio signal for testing the transmission is heard, the sound of the transmission is integrally known, sound samples are pre-scored in a training sheet (see the training sheet in a table 5), pre-scoring is only carried out on the sound samples by an evaluator to know a scoring flow and master sound characteristics, and the pre-scored scoring data are not subjected to statistical analysis;
explaining the scene and the characteristics of the acoustic event before listening evaluation, so that an evaluator can form the feeling of the acoustic event in advance in the evaluation process and imagine the scene of evaluated noise;
the evaluators officially listen and begin scoring against the official scoring statistics shown in table 5; a total of 28 evaluators, each evaluated 6X 14 sound samples, which were combined to form a scale V28×84As shown in table 6;
TABLE 6 evaluation personnel rating results
Figure BDA0003095335920000201
And step 9: confidence analysis is performed on confidence samples No. 11-14 and No. 81-84 according to each row of data in the table 6, and the calculation formula is as follows:
WT=1-Pn-λPy
if λ is 1 and no acceptable erroneous determination is allowed, then
Figure BDA0003095335920000202
NwThe numbers of the scores of the two groups of No. 11-14 and No. 81-84 are inconsistent, for example, if the scores of No. 11-14 of the evaluator A1B1 are 9, 1, 7, 6, and the scores of No. 81-84 are 9, 2, 6, 6, then two confidence evaluation sample scores are inconsistent, and the confidence weight of the evaluator A1B1 is WT(A1B1)=0.5;
The calculation results of all the evaluator confidence level weight values are shown in table 7;
TABLE 7 evaluation personnel reliability weighting Table
Evaluation personnel Confidence weights Evaluation personnel Confidence weights
A1
1 O1 1
B1 0.75 P1 0.5
…… ……
N1 1 A1B1 0.5
Step 10: the results of the subjective evaluations of 28 evaluators in table 6 were subjected to Spearman correlation coefficient analysis:
10.1) calculating the Spearman correlation coefficient between each two evaluators:
Figure BDA0003095335920000211
in the formula: d is the difference between the subjective evaluation result grades in the two rows in the table; r is the length of two lines of subjective evaluation results; ri,jThe Spearman correlation coefficient of the subjective evaluation result of the ith evaluator to the subjective evaluation result of the jth evaluator is shown, and the calculation results of the evaluators are shown in table 8;
TABLE 8 Spearman correlation coefficient Table between evaluators
Correlation coefficient A1 B1 C1 …… A1B1
A1 1.00 0.81 0.81 …… 0.91
B1 0.81 1.00 0.86 …… 0.78
C1 0.81 0.86 1.00 …… 0.67
…… …… …… …… …… ……
A1B1 0.91 0.78 0.67 …… 1.00
10.2) removing the correlation coefficient of which each row value is 1, averaging to obtain an average correlation coefficient, wherein a calculation formula is shown as follows;
Figure BDA0003095335920000221
in the formula: k is the number of evaluators; ri,jThe correlation coefficient between the ith evaluator and the jth evaluator is obtained; riThe average correlation coefficient of the ith evaluator relative to other evaluators;
the results of the average correlation coefficient calculation by the evaluators are shown in table 9;
TABLE 9 average correlation coefficient of evaluators
Figure BDA0003095335920000223
The average correlation coefficient of 28 evaluators in total, the scoring data of the evaluators with the correlation coefficient of more than 0.75 is kept, and the scoring data of the evaluators with the numbers of C1, E1 and O1 in the table 6 are removed;
10.3) scoring data V for the remaining 25 persons of Table 625×84Averaging each column to obtain the average score of each sample, wherein the calculation formula is as follows;
Figure BDA0003095335920000222
in the formula: k' is the number of remaining evaluators after rejection, Vi,aSubjective scoring value of the ith evaluator on the a-th sample; vaAverage scores for all raters for a samples;
after three appraisers C1, E1 and O1 are removed, the average subjective evaluation scores of the remaining 25 appraisers on 84 listening samples are shown in Table 10;
TABLE 10 sample subjective evaluation mean scores
Sample numbering 1 2 3 …… 84
Mean score Va 2.38 9.42 2.00 …… 3.05
10.4) removing three evaluators C1, E1 and O1 from Table 6, and remaining V25×84Score data of (1), score data per line Vi,84And VaThe formula for calculating the Euclidean distance is as follows:
Figure BDA0003095335920000231
wherein i is 25;
the calculation results are shown in table 11;
TABLE 11 Euclidean distance statistical analysis
Number of evaluator A1 B1 D1 F1 …… A1B1
Euclidean distance D 7.48 12.26 13.72 9.27 …… 21.25
Table 11 shows the first action, the second action, the 84 listening sample scoring data per evaluator, and the average score V of evaluatorsaThe Euclidean distance D between the evaluation personnel and the evaluation personnel is obtained by rejecting the scoring data of the two evaluation personnel with the largest Euclidean distance, wherein the scoring data are the scoring data of the evaluation personnel with the numbers of R1 and A1B1 in the table 11, and the rejection number is not more than 20% of the total number of the evaluation people in the statistical analysis of the whole subjective evaluation experiment;
spearman correlation coefficient and Euclidean distance statisticsAfter the scientific analysis, the scoring data of the evaluators numbered C1, E1, O1, R1 and A1B1 in the table 6 are rejected, and the scoring data scale of the rest evaluators is V23×84To V pair23×84And attaching the reliability weight and the group weight of corresponding appraisers to each row, and obtaining the subjective appraisal labels of 84 listening samples according to the following formula:
Figure BDA0003095335920000241
finally, effective scoring data V of 23 evaluators is reserved23×84Adding the reliability weight and the group weight of corresponding evaluators, and calculating to obtain the final subjective evaluation labels of 84 listening samples as shown in table 12;
TABLE 12 Scoring effectiveness data of listening samples and subjective evaluation labels
Figure BDA0003095335920000242
Step 11: establishing a transmission sound quality subjective and objective evaluation model, which comprises the following specific steps:
11.1) taking 70% of 84 listening sample data as a training set for training a transmission sound quality subjective and objective evaluation model, and taking 30% of the 84 listening sample data as a test set for testing the transmission sound quality subjective and objective evaluation model;
11.2) all listening samples are processed in a frame-dividing mode, the frame length is 1s, the frame shift is 1s, and one listening sample is divided into 5 sections in a frame-dividing mode;
11.3) calculating the psychoacoustic index of sound quality of each frame of each listening sample, wherein the calculation process is shown in FIG. 4, and the calculation result is shown in Table 13;
TABLE 13 calculation of psychoacoustic index of sound quality by listening to samples in frames
Figure BDA0003095335920000243
Figure BDA0003095335920000251
11.4) the transmission sound quality subjective and objective evaluation model is fitted by a support vector regression method, MATLAB software and libsvm program package are adopted for modeling, the psychoacoustic index of sound quality calculated by framing of a training set and a subjective evaluation label are used as input data for transmission sound quality subjective and objective evaluation model training, an RBF kernel function is selected, an insensitive loss function is taken as epsilon 0.01, an optimal penalty parameter e and a kernel function parameter g are selected by a K-CV (K-fold Cross Validation) Cross Validation method, the lowest mean square error mse in the Cross Validation process is taken as an optimization target function, a Cross Validation parameter v is selected as 3, and the mse formula is calculated as follows:
Figure BDA0003095335920000252
in the formula: v is the number of cross validation groups, and n is the number of each group of cross validation groups; y isijIn order to obtain the true label of the sample,
Figure BDA0003095335920000253
a transmission sound quality subjective and objective evaluation model prediction label;
using a grid search method, a rough selection is first performed, taking a log base 22e、log2g, the value ranges are [ -8,8 respectively]、[-8,8]The step size of the penalty parameter e and the kernel function parameter g is both 1; fine selection is carried out again according to the coarse selection result, and the log with the base 2 is continuously taken2e、log2g, the value ranges are [ -4,4 ] respectively]、[-4,4]Step sizes of the penalty parameter e and the kernel function parameter g are both 0.1, and finally the optimal cross validation parameter e is selected to be 2.7763, and g is selected to be 0.1147; training the transmission sound quality subjective and objective evaluation model by using 59 groups of listening sample data in the training set, and predicting 25 groups of listening sample subjective evaluation labels in the test set by using the trained transmission sound quality subjective and objective evaluation model to obtain the prediction result of the transmission sound quality subjective and objective evaluation model on the test set subjective evaluation labels, namely the test resultCollecting a predicted value;
11.5) establishing a transmission sound quality subjective and objective evaluation model;
step 12: verifying the transmission sound quality subjective and objective evaluation model established in the step 11, taking a Pearson correlation coefficient and an average absolute error as evaluation indexes, if the Pearson correlation coefficient is greater than 0.9, and the average absolute error is less than 10% of the maximum value of a grading interval, indicating that the transmission sound quality subjective and objective evaluation model is well predicted, and if the Pearson correlation coefficient is not greater than 0.9, returning to the step 6, and performing the sound quality subjective and objective evaluation again;
and the transmission sound quality subjective and objective evaluation model is used for evaluating the Pearson correlation coefficient between the test set predicted value and the test set subjective evaluation label:
Figure BDA0003095335920000261
in the formula: y is the test set subjective rating label,
Figure BDA0003095335920000262
subjective evaluation label prediction values of the transmission sound quality subjective and objective evaluation model on the test set,
Figure BDA0003095335920000263
is y and
Figure BDA0003095335920000264
covariance between, σyIs the mean square error of y, μyIs the mean value of y, EλRepresents a mathematical expectation;
the average absolute error MAE is calculated as follows:
Figure BDA0003095335920000265
in the formula:
Figure BDA0003095335920000266
is the predicted value of the ith listening sample, yiIs a subjective evaluationPrice tag, nsamplesIs the total number of listening samples;
percentage of error WmaeThe calculation formula is as follows:
Figure BDA0003095335920000267
in the formula:
Figure BDA0003095335920000271
is the average absolute error between the predicted value of the subjective evaluation label and the subjective evaluation label of the test set, LmaxIs the maximum value of the scoring interval.
Therefore, the prediction accuracy of the transmission sound quality subjective and objective evaluation model meets the requirement, the comparison graph of the model prediction value and the test set subjective evaluation label is shown in fig. 5, and it can be seen from fig. 5 that the error value of the model prediction value and the test set subjective evaluation label of only a few samples is greater than 1, the majority is less than 0.5, and the average error value of all samples is 0.557; the test set predicted value and the test set subjective evaluation label error graph are shown in fig. 6, and it can be seen from fig. 6 that the test set subjective evaluation label and the test set model predicted value have basically consistent trend, high correlation and better transmission sound quality subjective and objective evaluation model prediction accuracy.

Claims (11)

1. The subjective and objective evaluation method for the sound quality under the variable-speed working condition of the transmission is characterized by comprising the following steps of:
step 1: collecting noise signals of the transmission under the condition of changing the rotating speed of M gears by using a microphone, and collecting rotating speed signals of the transmission by using a photoelectric sensor;
step 2: performing sliding window interception on noise signals under M gears of the transmission to generate preliminary listening sample sets of the M gears, wherein each preliminary listening sample set comprises E listening samples, and the total number of the preliminary listening samples is M multiplied by E listening samples;
and step 3: respectively calculating sound quality psychoacoustic indexes of the E listening samples contained in each preliminary listening sample set in the step 2, and performing cluster analysis according to the calculation results of the sound quality psychoacoustic indexes to finally form M x c preliminary listening samples;
the acoustic quality psychoacoustic indexes include loudness, sharpness, roughness and jitter degree, but are not limited to the four acoustic quality psychoacoustic indexes; the clustering analysis method adopts a K-means clustering analysis method;
and 4, step 4: for the M × c preliminary listening samples in the step 3, reserving listening samples closest to the clustering center point in each category to form a listening sample set with the sample capacity of M × c;
and 5: adding H samples which are different in the two groups and are the same in the groups after the 1 st group and the M group of the listening sample set in the step 4 to serve as evaluation personnel reliability evaluation samples, and adding samples manually selected by NVH experts from the 2 nd group to the M-1 st group to keep the number of listening samples of each group consistent; the listening samples are arranged in sequence according to the corresponding actual rotating speed value of the transmission to form a formal listening sample group with the sample capacity of Mx (c + H);
step 6: grouping the evaluators, and adding different group weights between each group;
and 7: selecting semantic antisense words capable of reflecting the sound characteristics of the transmission to be evaluated from a transmission sound quality semantic word bank by NVH experts;
and 8: subjective evaluation is carried out on the formal listening sample group in the step 5, the semantic antisense words are adopted as the subjective evaluation words in the step 7, and after the subjective evaluation, the subjective evaluation result of each person on the listening samples in the formal listening sample group is obtained;
and step 9: carrying out listening reliability analysis on the appraisers, and adding reliability weight to subjective evaluation results of the appraisers;
step 10: performing statistical analysis on the Spearman correlation coefficient and Euclidean distance indexes on the subjective evaluation result in the step 8, eliminating evaluation personnel with inaccurate scores, adding the group weight in the step 6 and the reliability weight in the step 9 to the score data of the rest personnel after elimination, and calculating to obtain a subjective evaluation label;
step 11: establishing a transmission sound quality subjective and objective evaluation model;
calculating the acoustic quality psychoacoustic indexes of the formal listening sample component frames in the step 5, dividing a test set and a training set, and establishing a transmission acoustic quality subjective and objective evaluation model by adopting a support vector regression method according to the acoustic quality psychoacoustic index calculation results of the listening samples in the training set and the subjective evaluation labels in the step 10;
step 12: and (3) verifying the transmission sound quality subjective and objective evaluation model:
and (3) verifying the transmission sound quality subjective and objective evaluation model established in the step 11 by adopting the test set in the step 11, taking a Pearson correlation coefficient and an average absolute error as evaluation indexes, if the Pearson correlation coefficient is greater than 0.9, and the average absolute error is less than 10% of the maximum value of the grading interval, indicating that the transmission sound quality subjective and objective evaluation model is well predicted, and if the Pearson correlation coefficient is not greater than 0.9, returning to the step 6, and re-performing the sound quality subjective and objective evaluation.
2. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: in the step 1, a microphone is used for collecting noise signals under the variable-speed working condition of M gears of the transmission, the collection environment is a semi-anechoic chamber, the microphone is arranged at a position 1M away from the center of the transmission from a loading end to the left side, the position is a sound field far field, no interference effect exists among sound waves, the sound source is approximately regarded as a point sound source, the actual noise evaluation requirement is met, the sampling frequency is not lower than 40960Hz, the analysis frequency is not lower than 20480Hz and higher than the upper limit frequency of human hearing according to the Nyquist theorem, and each gear of the transmission is driven to rotate at the highest speed rpm under the maximum loadmaxUniformly decelerated to the lowest rpmminAnd collecting noise signals, wherein the collection time of each gear is S seconds.
3. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: in the step 2, noise signals under M gears of the transmission are intercepted by a sliding window, a window function of the sliding window is selected as a rectangular window, the length of the noise signal under each gear is set to be N, the window length is wlen, the sliding displacement of the latter window relative to the former window is wst, wherein the time length range of wlen is 5s or more and less than wlen and less than or equal to 10s, the time length range of wst is 0s or more and less than wst or less than or equal to wlen, and the calculation formula of the number N of sample sections intercepted by the sliding window is as follows:
Figure FDA0003095335910000031
4. the method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: the method for calculating the psychoacoustic index of sound quality in the step 3 comprises the following steps:
3.1) loudness calculation:
the specific loudness is calculated using the following formula:
Figure FDA0003095335910000041
of formula (II) to (III)'0To reference specific loudness, ETQFor the corresponding excitation in the quiet state, srIs the ratio of the sound intensity of a just-audible test tone to that of a broadband noise at the same critical band, E0Is sound intensity I0=10-12W/m2Corresponding reference excitation value, EsIs the excitation to which the sound corresponds, when N'0When 0.065, 0.25 s is selected as θr0.25; is N'0When equal to 0.08, take θ equal to 0.23, sr0.5; the Bark band division standard adopts a Zwicker model Bark band division standard;
the total loudness is obtained by integrating the specific loudness over the 0-24Bark scale:
Figure FDA0003095335910000042
3.2) sharpness calculation:
the Zwicker sharpness model is based on a loudness model, and the mathematical model is as follows:
Figure FDA0003095335910000043
in the formula, K is a weighting coefficient, and K is 0.11; ssharpnessRepresenting sharpness, and N' (z) representing specific loudness in Bark domain z, where g (z) is the weight coefficient of the sound signal in different Bark domains, expressed as:
Figure FDA0003095335910000044
3.3) roughness calculation:
the roughness model after Zwicker improvement is based on a loudness model, and the mathematical model is as follows:
Figure FDA0003095335910000045
wherein Rou is the calculated roughness, fmodTo modulate frequency, Δ LEFor the sound pressure variation amplitude in each critical frequency band, the following is defined:
Figure FDA0003095335910000051
in the formula, Nmax′(z) and Nmin′(z) represents the maximum and minimum values of the characteristic loudness in the Zwicker loudness model, respectively;
3.4) jitter degree calculation:
the Zwicker jitter model is based on a loudness model, and the mathematical model is as follows:
Figure FDA0003095335910000052
in the formula,. DELTA.LEFor the amplitude of the sound pressure variation within each critical frequency band;fmodIs the modulation frequency; f. of0Is modulating the fundamental frequency, f0=4Hz。
5. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: and in the step 5, the chrominance evaluation sample is a sound sample selected by a large amount of auditions of an expert, wherein the number of the large amount of auditions is not less than the number of the centralized audition samples of the audition samples in the step 4.
6. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: in the step 5, the acoustic samples are arranged in sequence according to the corresponding actual rotating speed value of the gearbox, and the sequence of the acoustic samples is as follows: the first sample of each group is the intercepted sample corresponding to the highest rotating speed section, the second sample is the intercepted sample corresponding to the lowest rotating speed section, other samples are arranged according to the sequence of the corresponding rotating speed sections from large to small or from small to large, a 5s blank space is reserved between the two listening samples, and time is scored for evaluators.
7. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: in the step 6, the evaluators are grouped, and different group weights are added between each group, namely the evaluators with different listening levels are divided into expert groups, experience groups and common groups, and different group weights W are respectively given to the evaluatorsG
8. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: in the step 9, listening reliability analysis is carried out on the appraisers, and reliability weight W is added to the subjective evaluation results of the appraisersTThe method comprises the following specific steps:
9.1) extracting the scoring data of 2 XH credibility evaluation samples by an evaluator, wherein if all samples are scored consistently, the misjudgment rate is 0, if the samples with inconsistent scoring have delta, and the difference between two scoring does not exceed the scoringThe maximum value of the interval is 10%, the acceptable error rate is calculated to be
Figure FDA0003095335910000061
If b samples with inconsistent scores exist and the difference between the two scores exceeds 10% of the maximum value of the scoring interval, calculating the unacceptable misjudgment rate as
Figure FDA0003095335910000062
9.2) according to the acceptable false positive rate PyAnd unacceptable false positive rate PnForming a confidence weight WTThe calculation formula is as follows:
WT=1-Pn-λPy
wherein, the value of lambda is between 0 and 1, if the acceptable misjudgment is not allowed, the lambda is 1; if the acceptable misjudgment is completely allowed, λ is 0.
9. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: the statistical analysis and calculation steps of the Spearman correlation coefficient and the Euclidean distance in the step 10 are as follows:
10.1) calculating the Spearman correlation coefficient between each two evaluators:
Figure FDA0003095335910000071
in the formula: d is the difference between the subjective evaluation result grades of the two lines; r is the length of two lines of subjective evaluation results; ri,jA Spearman correlation coefficient representing the subjective evaluation result of the ith evaluator to the subjective evaluation result of the jth evaluator;
10.2) taking the average correlation coefficient:
Figure FDA0003095335910000072
in the formula: k is an evaluatorCounting;
Figure FDA0003095335910000073
the correlation coefficient between the ith evaluator and the jth evaluator is obtained; riThe average correlation coefficient of the ith evaluator relative to the other evaluators in the panel; set RiThe threshold value is 0.75 if RiIf the correlation is less than 0.75, the correlation of the evaluator is not high relative to other evaluators, the subjective listening sensation is greatly deviated, and the scoring data of the evaluator is removed;
10.3) averaging the scoring data of the evaluators retained after the rejection, wherein the average scoring value of each sample is obtained by the following calculation formula;
Figure FDA0003095335910000074
in the formula: k' is the number of remaining evaluators after rejection, Vi,aSubjective scoring value of the ith evaluator on the a-th sample; vaAverage scores for all raters for a samples;
10.4) score data V for each of the remaining ratersi,aAnd VaThe Euclidean distance statistical analysis and calculation method comprises the following steps:
Figure FDA0003095335910000075
wherein i is the number of evaluators;
eliminating residual evaluator correspondence D (V)i,a,Va) The larger evaluation data requires that the Spearman correlation coefficient and Euclidean distance statistical analysis rejecting population is not more than 20% of the total evaluation population.
10. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: the step 11 of establishing the transmission sound quality subjective and objective evaluation model comprises the following specific steps:
11.1) taking 70% of the formal listening samples of the listening sample group as a training set for training the transmission sound quality subjective and objective evaluation model; 30% of the test set is used for testing the transmission sound quality subjective and objective evaluation model;
11.2) setting frame length and frame shift for all listening samples, wherein the frame length time length interval is (0,1] second, the frame shift time length interval is (0,1] second, and the frames are divided into f sections;
11.3) calculating the psychoacoustic index of sound quality of each frame of the listening sample;
11.4) using the acoustic quality psychoacoustic indexes calculated by the training set in frames and the subjective evaluation labels in the step 10 as parameters for training an acoustic quality subjective and objective evaluation model of the transmission;
11.5) training a transmission sound quality subjective and objective evaluation model, and establishing the transmission sound quality subjective and objective evaluation model;
in the step 11, a support vector regression method is used for carrying out transmission sound quality subjective and objective evaluation model fitting, an RBF kernel function is selected, an insensitive loss function takes epsilon as 0.01, an optimal penalty parameter e and a kernel function parameter g are selected by adopting a K-CV (K-fold Cross Validation) Cross Validation method, the lowest mean square error mse in the Cross Validation process is taken as an optimization target function, the Cross Validation parameter v is selected as 3, and the mse formula is calculated as follows:
Figure FDA0003095335910000091
in the formula: v is the number of cross validation groups, nl is the number of cross validation groups, yijIn order to obtain the true label of the sample,
Figure FDA0003095335910000092
a transmission sound quality subjective and objective evaluation model prediction label;
using a grid search method, a rough selection is first performed, taking a log base 22e、log2g, the value ranges are [ -8,8 respectively]、[-8,8]The step size of the penalty parameter e and the kernel function parameter g is both 1; according toThe coarse selection result is then selected finely, and the log with the base 2 is taken continuously2e、log2g, the value ranges are [ -4,4 ] respectively]、[-4,4]The step sizes of the penalty parameter e and the kernel function parameter g are both 0.1.
11. The method for objectively evaluating the sound quality under the variable-speed working condition of the transmission according to claim 1, wherein the method comprises the following steps: the calculation formula of the Pearson correlation coefficient ρ in step 12 is as follows:
Figure FDA0003095335910000093
in the formula: y is the test set subjective rating label,
Figure FDA0003095335910000094
subjective evaluation label prediction values of the transmission sound quality subjective and objective evaluation model on the test set,
Figure FDA0003095335910000095
is y and
Figure FDA0003095335910000096
covariance between, σyIs the mean square error of y, μyIs the mean value of y, EλRepresents a mathematical expectation;
mean absolute error MAE is less than maximum value L of scoring intervalmax10% of MAE, the MAE calculation formula is as follows:
Figure FDA0003095335910000097
in the formula:
Figure FDA0003095335910000098
is the predicted value of the ith listening sample, yiIs a subjective evaluation label, nsamplesIs the total number of listening samples;
percentage of error WmaeThe calculation formula is as follows:
Figure FDA0003095335910000101
in the formula:
Figure FDA0003095335910000102
is the average absolute error between the predicted value of the subjective evaluation label and the subjective evaluation label of the test set, LmaxIs the maximum value of the scoring interval.
CN202110609285.9A 2021-06-01 2021-06-01 Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission Active CN113343384B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110609285.9A CN113343384B (en) 2021-06-01 2021-06-01 Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110609285.9A CN113343384B (en) 2021-06-01 2021-06-01 Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission

Publications (2)

Publication Number Publication Date
CN113343384A true CN113343384A (en) 2021-09-03
CN113343384B CN113343384B (en) 2023-04-07

Family

ID=77472763

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110609285.9A Active CN113343384B (en) 2021-06-01 2021-06-01 Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission

Country Status (1)

Country Link
CN (1) CN113343384B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115406670A (en) * 2022-08-16 2022-11-29 中国第一汽车股份有限公司 Vehicle performance testing method and device, electronic equipment and vehicle
CN116434372A (en) * 2023-06-12 2023-07-14 昆明理工大学 Intelligent data acquisition system and working condition identification system for variable working condition equipment

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103471709A (en) * 2013-09-17 2013-12-25 吉林大学 Method for predicting noise quality of noise inside passenger vehicle
US10249293B1 (en) * 2018-06-11 2019-04-02 Capital One Services, Llc Listening devices for obtaining metrics from ambient noise
CN109668626A (en) * 2018-12-25 2019-04-23 东莞材料基因高等理工研究院 A kind of sound quality evaluation method based on human-computer interaction interface
CN110737970A (en) * 2019-09-24 2020-01-31 天津大学 engine acceleration sound quality evaluation method
CN112215469A (en) * 2020-09-15 2021-01-12 中国第一汽车股份有限公司 Sound design method for automobile key switch
CN112790756A (en) * 2020-10-28 2021-05-14 北京银河方圆科技有限公司 Task state functional magnetic resonance scanning data denoising method, device, equipment and medium

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103471709A (en) * 2013-09-17 2013-12-25 吉林大学 Method for predicting noise quality of noise inside passenger vehicle
US10249293B1 (en) * 2018-06-11 2019-04-02 Capital One Services, Llc Listening devices for obtaining metrics from ambient noise
CN109668626A (en) * 2018-12-25 2019-04-23 东莞材料基因高等理工研究院 A kind of sound quality evaluation method based on human-computer interaction interface
CN110737970A (en) * 2019-09-24 2020-01-31 天津大学 engine acceleration sound quality evaluation method
CN112215469A (en) * 2020-09-15 2021-01-12 中国第一汽车股份有限公司 Sound design method for automobile key switch
CN112790756A (en) * 2020-10-28 2021-05-14 北京银河方圆科技有限公司 Task state functional magnetic resonance scanning data denoising method, device, equipment and medium

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
K. OZAWA ET AL: "Compensation methods of sound quality for a car-audio equalizer", 《IEEE》 *
姜吉光等: "车内噪声品质偏好性主客观评价及相关性分析", 《汽车技术》 *
曹晓琳等: "车内噪声品质主客观评价探究式教学实验", 《物理实验》 *
梁杰等: "基于相关分析的车内声品质偏好性评价模型", 《吉林大学学报(工学版)》 *
欧阳祖琛: "空载工况重型变速箱声品质方法研究", 《中国优秀硕士学位论文全文数据库(电子期刊)工程科技II辑》 *
王岩松等: "车辆噪声声品质的心理声学评价系统", 《上海工程技术大学学报》 *
陈克等: "车内声品质主客观评价的相关性分析", 《沈阳理工大学学报》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115406670A (en) * 2022-08-16 2022-11-29 中国第一汽车股份有限公司 Vehicle performance testing method and device, electronic equipment and vehicle
CN116434372A (en) * 2023-06-12 2023-07-14 昆明理工大学 Intelligent data acquisition system and working condition identification system for variable working condition equipment
CN116434372B (en) * 2023-06-12 2023-08-18 昆明理工大学 Intelligent data acquisition system and working condition identification system for variable working condition equipment

Also Published As

Publication number Publication date
CN113343384B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
CN113343384B (en) Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission
CN106933146B (en) Pedestrian warning sound control method for electric car
CN111064327B (en) Driving motor optimization method and system based on sound quality
Phatak et al. Consonant confusions in white noise
CN101672690A (en) Method for objectively and quantifiably evaluating noise fret degree in vehicle based on auditory model
CN110688712A (en) Evaluation index for objective annoyance degree of automobile wind vibration noise sound quality and calculation method thereof
CN110261101B (en) Howling quantitative evaluation method based on orthogonal-pair comparison test and SVR
CN111982274B (en) Motor noise source identification method and system based on sound quality contribution coefficient
CN103471709A (en) Method for predicting noise quality of noise inside passenger vehicle
Park et al. A comparative study on subjective feeling of engine acceleration sound by automobile types
Hilkhuysen et al. Effects of noise suppression on intelligibility: Dependency on signal-to-noise ratios
CN108827452B (en) Noise evaluation method of permanent magnet synchronous hub motor
CN108694935B (en) New energy vehicle and active noise reduction method and system thereof
Lemaitre et al. A psychoacoustical study of wind buffeting noise
KR101314043B1 (en) Method for sound quality analysis of axle gear whine sound and apparatus thereof
CN112131662A (en) Passenger car wind noise subjective evaluation objective quantification method
CN115683318A (en) Method for evaluating sound quality and determining limit value of driving motor system
Park et al. Evaluation index for sporty engine sound reflecting evaluators’ tastes, developed using K-means cluster analysis
CN115017974A (en) Sound quality self-shaping design method and system integrating sound quality evaluation
Rimskaya-Korsakova et al. Evaluations of the annoyance effects of noise
Samardzic et al. In-vehicle application of common speech intelligibility metrics
CN116933620A (en) Noise quality evaluation and model building method
KR100925828B1 (en) Method of expressing the quality of the sound in vehicle as the quantitive equation and device thereof
CN113486448A (en) Method for evaluating transmission squeal based on masking effect
Susini et al. The effect of loudness on the perceptual representation of sounds with similar timbre

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant