CN110261101A - Quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR - Google Patents
Quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR Download PDFInfo
- Publication number
- CN110261101A CN110261101A CN201910513521.XA CN201910513521A CN110261101A CN 110261101 A CN110261101 A CN 110261101A CN 201910513521 A CN201910513521 A CN 201910513521A CN 110261101 A CN110261101 A CN 110261101A
- Authority
- CN
- China
- Prior art keywords
- orthogonal
- pairs
- marking
- sample
- frequency
- 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
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/021—Gearings
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/028—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Acoustics & Sound (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
Quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR, orthogonal test and pairs of comparative test are combined first, obtain the test sample for training pattern, next carries out the subjective assessment of actual measurement audio sample, different pure tones are carried out to quantify abnormal sound percentage contribution of uttering long and high-pitched sounds, in conjunction with exclusion-comparison thought and grade scoring method, more pure tones finally are constructed with support vector regression and are uttered long and high-pitched sounds Quantitative evaluation model, and have carried out validation checking；Orthogonal test has the characteristics that equilibrium disperses, is neat comparable, can greatly reduce experiment quantity；Comparison has the characteristics that easily operated, simple and fast in pairs, of less demanding to professional knowledge, can more reflect target signature；The present invention realizes the Quantitative evaluation of prominent frequency contribution in overall noise of uttering long and high-pitched sounds in audio signal.
Description
Technical field
It is the present invention relates to subtracting variable-speed motor to utter long and high-pitched sounds abnormal sound assessment technique field, in particular to a kind of to be based on orthogonal-pairs of comparison
The quantitative evaluation method of uttering long and high-pitched sounds of test and SVR.
Background technique
Subtracting variable-speed motor is that can be realized the output of different rotating speeds for changing the mechanism from engine speed and torque,
Have the advantages that high-efficient, simple structure, easy to use, is widely used in the fields such as ship, vehicle.Subtract the gear of variable-speed motor
System is usually to, gear meshing gear pair and idle pulley meshing gear by constant mesh gear to forming, gear in process of production,
Since process equipment precision is insufficient, assembly manipulation is unreasonable, there is microscopic appearance out-of-flatness and local defect etc. and asks in gear teeth face
Topic, and then gear mesh is caused to occur ear-piercing abnormal sound of uttering long and high-pitched sounds in engagement process, signal is usually expressed as gear pair engagement frequency
Rate pure tone and its frequency multiplication pure tone ingredient are prominent.Accurately identify cause the frequency pure tone feature for the problem of uttering long and high-pitched sounds to instructing gear is microcosmic to be repaired
Shape, reduction abnormal sound of uttering long and high-pitched sounds are of great significance.
Abnormal sound of uttering long and high-pitched sounds is a kind of typical subjective evaluation index, is evaluated by human ear, and human ear is a strongly non-linear system,
Different to the pure tone susceptibility of different frequency, even if identical energy, the pure tone auditory perception of different frequency is different.Directly pass through 1/
The size of 3 octave analysis centre frequency sound pressure levels only considers noise signal energy to judge to utter long and high-pitched sounds, and has ignored psychology, the life of people
The influence of the composite factors such as reason, environment can not provide the quantitative target for abnormal sound of uttering long and high-pitched sounds according to human ear subjective sensation.Subtract variable-speed motor
Multiple pure tone ingredients are generally occurred in noise signal, the TNR index in ECMA standard calculates the signal-to-noise ratio in critical band, can
Expression pure tone is to the percentage contribution uttered long and high-pitched sounds to a certain extent, but has ignored the influence of the factors such as psychology, physiology, environment, can not
Each pure tone component is provided to the percentage contribution for abnormal sound of uttering long and high-pitched sounds.
Summary of the invention
In order to overcome the disadvantages of the above prior art, the invention proposes one kind to be based on orthogonal-pairs of comparative test and SVR
Quantitative evaluation method of uttering long and high-pitched sounds, this method is based on emulation signal and jury's subjective evaluation result, with support vector regression structure
It builds more pure tone audio signal Contribution Models, realizes that prominent frequency contribution in overall noise of uttering long and high-pitched sounds quantifies in audio signal.
In order to achieve the above object, the technical scheme adopted by the invention is as follows:
A kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR, comprising the following steps:
Step 1: design orthogonal test:
1.1) count audio signal characteristic, choose frequency domain in amplitude A, frequency f, 3 features of background noise energy Eb be because
Element；
1.2) horizontal number being determined according to 3 factors and choosing orthogonal design table, symbiosis is at M test sample；
Step 2: according to orthogonal test generate M test sample generation M group analogue audio sample, and make it is orthogonal-at
To comparing marking table；
Step 3: choosing n people jury, n > 15 carry out subjective scoring, methods of marking are as follows: to M to M analogue audio sample
A analogue audio sample is compared two-by-two, it is believed that analogue audio sample A ratio B more troublesome, then comparison result is 1；Think
Analogue audio sample B ratio A more troublesome, then comparison result is -1；Think analogue audio sample A troublesome as B,
Then comparison result is 0；It carries out altogetherSecondary comparison, 10 pairs of analogue audio samples of every comparison are rested 5 minutes；Scoring process repeats to listen
Sound, each pair of analogue audio sample need to carry out the two-way audition of AB and BA；Evaluation result is filled in into orthogonal-pairs of relatively marking
In the upper Delta Region of table；
Step 4: jury carries out second to M analogue audio sample and scores, and scoring process is the same as step 3；Compare twice
Marking result is modified the content in orthogonal-pairs of relatively marking table；
Step 5: orthogonal-pairs of relatively marking table is subjected to antisymmetry conversion, forms antisymmetric matrix, i.e., it is orthogonal-pairs of
When comparing upper triangle line n m column marking respectively 1, -1 and 0 in marking table, lower triangle in orthogonal-pairs of relatively marking table
Accordingly filling in marking is -1,1 and 0, generates complete orthogonal-pairs of relatively marking table；
Step 6: according to n orthogonal-pairs of relatively marking tables, calculating n people and give a mark to M the subjective of sample, form one
A scoring matrix Vn×M；
Step 7: correlation analysis being carried out to the marking of n reviewer, calculates the phase between each reviewer's marking
Relationship number, and the subjective of the marking biggish reviewer of error is rejected according to the size of related coefficient and is given a mark；
Step 8: by the scoring matrix V after correlation analysisw×MIt is averaged by column, w≤n, obtains M sample sound
Subjectivity marking SV1×M；It puts it into orthogonal test table, obtains complete orthogonal test table；
Step 9: range analysis being carried out according to complete orthogonal design table, obtains frequency A, amplitude f, background noise energy Eb
Three factors are to index S V1×MInfluence degree, draw factor-indicatrix, observe factor-indicatrix shape, determine return mould
Type: support vector regression model；Variance analysis is carried out according to complete orthogonal design table, determines the credible of test error and each factor
Degree；
Step 10: support vector regression constructs scoring model of uttering long and high-pitched sounds: with the number in the complete orthogonal test table of gained in step 8
According to for training set, Selecting All Parameters and optimization method, regression modeling is carried out using libsvm support vector machines program bag；
Step 11: the k meshing frequency and its frequency multiplication for subtracting variable-speed motor actual measurement audio sample are calculated, filtering bandwidth is chosen, if
Filter is counted, k meshing frequency is successively filtered out, obtains k sub- audio sample signals；
Step 12: generating audio according to signal obtained in step 11 and evaluate sample, give a mark for jury；
Step 13: design pure tone contribution subjective assessment marking table, n reviewer successively comment audio evaluation sample
Estimate marking: comparing audio evaluates the difference of front and back sound in sample, and provides grade scoring；
Step 14: n reviewer of statistics, as a result, with step 7, calculates n people's to the marking of m audio evaluation sample
Average correlation coefficient will be less than R0The corresponding marking of reviewer reject, calculate being averaged for all remaining reviewer marking
Value Ci(i=1,2,3 ..., m), as the subjective contribution margin of meshing frequency pure tone, score value is smaller, contributes bigger；Score value is bigger,
It contributes smaller；
Step 15: calculate in step 11 in g audio sample the corresponding amplitude of k prominent meshing frequency pure tone, frequency and
M objective parameter sample is obtained in the background noise energy of the actual measurement audio sample；By the corresponding master of m sample in step 14
It sees evaluation of estimate to be merged into m objective parameter sample, forms Sm×4Sample set, the sample set is as test set；
Step 16: by test set Sm×4It is put into more pure tone percentage contribution regression mathematical models that step 10 is trained and carries out
Prediction, obtains the prediction subjectivity contribution margin sv of each pure tonei；
Step 17: subjective contribution margin sv will be predicted by linear compression and normalizationiBe converted to relatively subjective contribution margin；
Step 18: calculating mean square error mse and coefficient of determination R2, model validation is verified,
Wherein, m is total sample number, yiFor sample predictions value,For sample true value；
The step 1- step 10 is orthogonal-pairs of comparative test, for constructing theoretical prediction model.
The step 11- step 14 is that meshing frequency pure tone is used for the subjective scoring method uttered long and high-pitched sounds in measured signal
Quantify different meshing frequencies to the percentage contribution uttered long and high-pitched sounds, verifying of the scoring as theoretical prediction model.
The step 15- step 18 is the inspection of more pure tone percentage contribution regression mathematical models, the life including test set
At, the validation verification of the normalized of prediction result, model.
The definition method of ambient noise in the step 1.1), ambient noise amplitude are to be K to audio signal frequency spectrum
The amplitude of point sliding average, K point moving average filter:
Yy (1)=y (1)
Yy (2)=(y (1)+y (2)+y (3))/3
Yy (3)=(y (1)+y (2)+y (3)+y (4)+y (5))/M
Yy (4)=(y (2)+y (3)+y (4)+y (5)+y (6))/M
……。
Background noise energy Eb calculation formula in the step 1.1) are as follows:
Wherein, N is frequency spectrum points, AiFor ambient noise amplitude.
It is taken in the step 1.1) according to the experience of 3 amplitude A in frequency domain, frequency f, background noise energy Eb factors
It is worth range, each factor equidistantly chooses p value as level from low to high, and chooses LM(pC) orthogonal test table, M is formed altogether
Secondary test generates M analogue audio signal.
The generation method of analogue audio sample in the step 2 are as follows:
Sound=Asin (2 π ft)+D*randn (1, N)
A-amplitude
F-frequency
D-random noise standard deviation
Randn (1, N)-generates the random number for meeting standardized normal distribution of 1 row N column
Wherein the relationship of random signal standard deviation D and background noise energy Eb is obtained by Parseval theorem:
Eb=20log10 (D2*N)。
Orthogonal-pairs of relatively marking table in the step 2, illustrates column, triangle marking comprising essential information column, scoring
Table and points for attention.
Orthogonal-pairs of relatively marking table modification method in the step 4: compare second of marking knot with first time
Fruit, if giving a mark twice one is -1, one is 1, then marking is changed to 0；If giving a mark twice one is 0, one is -1, then will beat
Point result is changed to -1；If marking twice one is 1 for 0 one, marking result is changed to 1.
Subjectivity in the step 6 comments marking calculation method: complete orthogonal-pairs of relatively marking each column of table of statistics
In -1 number, using statistical result as the subjective assessment value of M analogue audio sample, statistical result is smaller, utter long and high-pitched sounds and be more obvious,
Statistical result is bigger, utters long and high-pitched sounds more unobvious.
Subjective scoring data processing method in the step 7: correlation point is carried out to the marking of n reviewer
Analysis, using Pearson correlation coefficient:
xi- evaluation personnel x gives a mark to i-th of the subjective of analogue audio sample
yi- evaluation personnel y gives a mark to i-th of the subjective of analogue audio sample
Average value of-evaluation personnel the x to the subjective marking of all analogue audio samples
Average value of-evaluation personnel the y to the subjective marking of all analogue audio samples.
Obtain Rn×nCorrelation coefficient charts, in every a line non-1 n-1 related coefficient is averaged, average correlation is obtained
Coefficient；N reviewer has n average correlation coefficient, and Trimmed mean related coefficient is less than R0Reviewer marking as a result,
1 > R0> 0.7.
The selection of support vector regression parameter in the step 10: using e-SVR support vector regression model, chooses diameter
To base kernel function, insensitive error takes e=0.01.
The selection of support vector regression parameter optimization method in the step 10: using k-fold cross validation method,
It is minimised as optimization aim with cross validation mean square error CVMSE, using grid-search algorithms to penalty factor and nuclear parameter γ
It optimizes, CVMSE formula is as follows:
K-cross validation broken number
N-cross validation points.
The frequency of the step 11 neutron audio sample signal filters out bandwidth choosing method: what each meshing frequency filtered out
Frequency range is the corresponding critical band of the meshing frequency, and critical frequency bandwidth calculation formula is as follows:
Wherein: f0For meshing frequency；ΔfcFor critical frequency bandwidth；
Critical band bound are as follows:
Wherein: f1For critical band lower bound, f2For the critical band upper bound.
The frequency of the step 11 neutron audio sample signal filters out filter design method: design is with audio sample frequency
Rate fsFor the digital band-reject filter of sampling frequency, performance requirement are as follows: band hinders range from f1To f2, decay not at this two frequency
Greater than 3dB, in f1- 100Hz and f2Decaying is not less than 20dB at+100Hz frequency；With this filter process audio signal, k is obtained
A sub- audio signal.
Step 12 sound intermediate frequency evaluates sample generating method: g actual measurement audio sample and k sub- audio samples are connected
It connects, the clear band of midfeather 1s, difference front and back sound.
The step 13 middle grade methods of marking, comparing audio evaluate the difference of front and back sound in sample, assess consonant
Utter long and high-pitched sounds weakening degree of the frequency sample relative to actual measurement audio sample, point 8 grades: very big -0, greatly -1, big -2, larger -
3, weak -4, weaker -5, very weak -6, without -7.
In the step 13 pure tone contribution subjective assessment give a mark table, by essential information column, grade scoring table, training table,
Formal marking statistical form and 5 part of points for attention composition.
The calculation method of relatively subjective contribution margin in the step 17, by predicted value linear compression evaluation of estimate 0-7 it
Between,
Wherein: ymax=7, ymin=0
More pure tone ingredients in single audio are normalized:
Value has the property that after normalization
Therefore, regard relatively subjective contribution margin as pure tone to the percentage contribution of overall whistler.
The invention has the benefit that
Orthogonal test of the invention has neat comparable, is the characteristic of dispersion, can effectively reduce test number (TN), in pairs
Comparative test can allow the people of not drama professional knowledge easily to give a mark, and error rate is low；The two tests are combined can be effective
It generates and covers comprehensive simulation sample, play a significant role to subsequent unified model of establishing；The building for surveying audio sample uses
The thought for excluding method of comparison, i.e., filter out sound at the beginning of frequency to be assessed from signal, compares and filters out noise rear and before filtering out,
It can effectively judge each pure tone in overall noise to the percentage contribution for abnormal sound of uttering long and high-pitched sounds；E-SVR then has small-sample learning, Gao Fei
The characteristics of Linear Mapping, high robust, is suitble to establish the mathematical model of weak regular structure.The present invention in summary three kinds of methods,
It can effectively judge to subtract variable-speed motor to utter long and high-pitched sounds problem.
Detailed description of the invention
Fig. 1 is the flow chart of present invention method.
Fig. 2 utters long and high-pitched sounds for orthogonal test of the embodiment of the present invention and emulates signal time-domain and frequency-domain.
Fig. 3 is that orthogonal-pairs of relatively marking of the embodiment of the present invention indicates to be intended to.
Fig. 4 is orthogonal test of embodiment of the present invention factor-indicatrix.
Fig. 5 is the composition of sample of embodiment of the present invention schematic diagram.
Fig. 6 is speed changer structure of embodiment of the present invention figure.
Fig. 7 is that audio of the embodiment of the present invention evaluates composition of sample figure.
Fig. 8 is that pure tone contributes subjective assessment to indicate to be intended to.
Fig. 9 is support vector regression model testing figure.
Figure 10 is support vector regression model prediction figure.
Figure 11 is the schematic diagram of grid-search algorithms of the embodiment of the present invention.
Specific embodiment
The present invention is described in more detail with reference to the accompanying drawings and examples.
As shown in Figure 1, a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR, including following step
It is rapid:
Step 1: design orthogonal test:
1.1) count audio signal characteristic, choose amplitude (A) in frequency domain, frequency (f), background noise energy (Eb) 3 because
Element, each factor successively choose p parameter from low to high, and the present embodiment takes 5 parameters, and wherein ambient noise amplitude is to audio
Signal spectrum does the amplitude of K point sliding average, takes K=1% × N, the following is K point moving average filter:
Yy (1)=y (1)
Yy (2)=(y (1)+y (2)+y (3))/3
Yy (3)=(y (1)+y (2)+y (3)+y (4)+y (5))/M
Yy (4)=(y (2)+y (3)+y (4)+y (5)+y (6))/M
……
Background noise energy Eb calculation formula are as follows:
Wherein, N is frequency spectrum points, AiFor ambient noise amplitude；
Selected amplitude (A) value are as follows: section [0.005 0.08] are chosen at equal intervals, wav audio amplitude；
Institute selected frequency (f) value are as follows: section [1,000 6000] are chosen at equal intervals, unit Hz；
Selected background noise energy (Eb) value are as follows: section [35 47] are chosen at equal intervals, unit dB；
1.2) parameter is corresponded to according to 3 factors choose LM(pC), wherein M is test number (TN), and p is number of levels, and c is factor
Number, following with M=25, p=5, c=3 illustrate that orthogonal test table designs orthogonal test, and raw 25 test samples of common property generate
Orthogonal test table, as shown in table 1；
Table 1
Step 2: generating 25 analogue audio signals, analogue audio signal according to 25 test samples that orthogonal test generates
Formula is as follows:
Sound=Asin (2 π ft)+D*randn (1, N)
A-amplitude
F-frequency
D-random noise standard deviation
Randn (1, N)-generates the random number for meeting standardized normal distribution of 1 row N column
Wherein the relationship of random noise standard deviation D and background noise energy Eb can be obtained by Parseval theorem:
Eb=20log10 (D2*N)
It is as shown in Figure 2 to emulate signal time-domain and frequency-domain；
Step 3: choosing 17 people juries, subjective scoring, methods of marking are as follows: to 25 are carried out to 25 analogue audio samples
Analogue audio sample is compared two-by-two, it is believed that analogue audio sample A ratio B more troublesome, then comparison result is 1；Think imitative
True audio sample B ratio A more troublesome, then comparison result is -1；Think analogue audio sample A troublesome as B, then
Comparison result is 0.Altogether relativelySecondary comparison, 10 pairs of analogue audio samples of every comparison are rested 5 minutes；Scoring process can weigh
Diplacusis sound, each pair of analogue audio sample need to carry out the two-way audition of AB and BA.Evaluation result is filled in orthogonal-pairs of comparison to beat
Divide in the upper Delta Region of table；Orthogonal-pairs of relatively marking table illustrates column, triangle marking table and note by essential information column, scoring
Item of anticipating composition, structure are as shown in Figure 3；
Step 4: jury carries out second to 25 analogue audio samples and scores, and scoring process is the same as step 3；Compare second
The secondary marking with first time is as a result, be -1 if giving a mark twice one, one is 1, then marking is changed to 0；If giving a mark one twice
It is 0, one is -1, then marking result is changed to -1；If marking twice one is 1 for 0 one, marking result is changed to 1；
Marking result after correction is filled in orthogonal-pairs of relatively upper Delta Region of marking table；
Step 5: orthogonal-pairs of relatively marking table is subjected to antisymmetry conversion, forms antisymmetric matrix, i.e., it is orthogonal-pairs of
When comparing upper triangle line n m column marking respectively 1, -1 and 0 in marking table, lower triangle in orthogonal-pairs of relatively marking table
Accordingly filling in marking is -1,1 and 0, generates complete orthogonal-pairs of relatively marking table, as shown in table 2；
Table 2
Step 6: -1 number in complete orthogonal-pairs of relatively marking each column of table of statistics, using statistical result as 25
The subjective assessment value of analogue audio sample.17 reviewers can obtain the marking of corresponding 25 analogue audio samples as a result,
Form a V17×25Scoring matrix, as shown in table 3；
Table 3
Step 7: correlation analysis is carried out to the marking of 17 reviewers, using Pearson related coefficient:
xi- evaluation personnel x gives a mark to i-th of the subjective of analogue audio sample
yi- evaluation personnel y gives a mark to i-th of the subjective of analogue audio sample
Average value of-evaluation personnel the x to the subjective marking of all analogue audio samples
Average value of-evaluation personnel the y to the subjective marking of all analogue audio samples
Obtain R17×17Correlation coefficient charts, as shown in table 4；
Table 4
In every a line non-1 16 related coefficients are averaged, obtain average correlation coefficient, as shown in table 5；
Table 5
17 reviewers have 17 average correlation coefficients, and reviewer of the Trimmed mean related coefficient less than 0.7 beats
Divide result；There is 1 people (number N) average correlation coefficient less than 0.7 in the present embodiment, rejects its subjective evaluation result；
Step 8: by the scoring matrix V after correlation analysis16×25It is averaged by column, as shown in table 6, obtains 25 sound
The subjective marking SV of sound sample1×25；
Table 6
It puts it into orthogonal test table, obtains complete orthogonal test table, as shown in table 7；
Table 7
Step 9: range analysis being carried out according to complete orthogonal design table, as shown in table 7, obtains frequency (A), amplitude (f), back
Three factors of scape noise energy (Eb) are to index (SV1×25) influence degree, draw factor-indicatrix, as shown in figure 4, observation
Known to factor-indicatrix shape: factor is in stronger non-linear with index, and because sample size is less, is determined back
Return model: ε-SVR model；Variance analysis is carried out according to complete orthogonal design table, determines the confidence level of test error Yu each factor,
In favor of subsequent regression analysis, as shown in table 8；
Table 8
Step 10: support vector regression constructs scoring model of uttering long and high-pitched sounds.With the number in the complete orthogonal test table of gained in step 8
According to for training set, regression modeling is carried out using libsvm support vector machines program bag.Using e-SVR support vector regression model,
Radial basis kernel function is chosen, using k-fold cross validation method, insensitive error takes e=0.01；Penalty factor and nuclear parameter
γ carries out optimal selection using grid-search algorithms, as shown in figure 11；
CVMSE is cross validation mean square error:
K-cross validation broken number, k=5 in this example
N-cross validation points
Optimized parameter is inputted into e-SVR model, using the data in the complete orthogonal test table of gained in step 8 as training set,
The self-test precision of more pure tone percentage contribution regression mathematical models and computation model is obtained, model self-test precision is such as in the present embodiment
Under, models fitting situation is as shown in Figure 9:
Step 11: (k is a, in the present embodiment for the main meshing frequency and its frequency multiplication that calculating subtracts variable-speed motor actual measurement audio sample
K=6, hereafter referred to collectively as meshing frequency), k meshing frequency is successively filtered out, k sub- audio sample signals are obtained.Each engagement frequency
The frequency range that rate filters out is the corresponding critical band of the meshing frequency, and critical frequency bandwidth calculation formula is as follows:
Wherein: f0For meshing frequency；ΔfcFor critical frequency bandwidth；
Critical band bound are as follows:
Wherein: f1For critical band lower bound, f2For the critical band upper bound；
Design is with audio sampling frequency fsFor the digital band-reject filter of sampling frequency, performance requirement are as follows: band hinders range from f1
To f2, decaying is not more than 3dB at this two frequency, in f1- 100Hz and f2Decaying is not less than 20dB at+100Hz frequency.With this
Filter process audio signal obtains k sub- audio signals；
Design sample structural representation is as shown in figure 5, the present embodiment carries out verifying explanation, speed change by taking certain domestic speed changer as an example
Device structure as shown in fig. 6, the audio signal that 3,4 gears of selection are uttered long and high-pitched sounds under most serious revolving speed is handled, choose by 3 gears
Audio signal when 2182rpm-2382rpm, 4 gears choose audio signal when 2192rpm-2372rpm, every 20rpm interception
One section of sample, symbiosis is at 21 samples；
Step 12: 21 actual measurement audio samples being connect, the clear band of midfeather 1s with 6 sub- audio samples, distinguished
Front and back sound forms m=21 × 6=126 audio evaluation sample and gives a mark for subsequent jury, and audio evaluates the composition of sample such as
Shown in Fig. 7:
13:17 reviewer of step successively carries out assessment marking to audio evaluation sample: comparing audio is evaluated in sample
The difference of front and back sound, utter long and high-pitched sounds weakening degree of the assessment sub-audio sample relative to actual measurement audio sample, point 8 grades are as follows
Table:
Pure tone contribute subjective assessment marking table by essential information column, grade scoring table, training table, formal marking statistical form and
5 part of points for attention composition, as shown in Figure 8；
Step 14: 17 reviewers of statistics, as a result, with step 7, calculate 17 to the marking of 126 audio evaluation samples
The average correlation coefficient of people, as shown in table 9,
Table 9
There is the average correlation coefficient of 2 people (number J, number L) less than 0.7 in the present embodiment, reject its subjective assessment value, counts
Calculate the average value C of all remaining reviewer markingi(i=1,2,3 ..., 126), it is as shown in table 10, pure as meshing frequency
The subjective contribution margin of sound, score value is smaller, and contribution is bigger, and score value is bigger, contributes smaller；
Table 10
Step 15: calculate in step 11 6 corresponding amplitudes of prominent meshing frequency pure tone in 21 audio samples, frequency and
126 objective parameter samples are obtained in the background noise energy of the actual measurement audio sample.126 samples in step 13 are corresponding
Subjective assessment value be merged into 126 groups of objective parameter samples, formed S126×4Sample set, the sample set as test set,
As shown in table 11；
Table 11
Step 16: by test set S126×4Be put into more pure tone percentage contribution regression mathematical models that step 10 is trained into
Row prediction, obtains the prediction subjectivity contribution margin sv of each pure tonei；
Step 17: will predict subjective contribution margin linear compression between evaluation of estimate 0-7,
Wherein: ymax=7, ymin=0
More pure tone ingredients in single audio are normalized:
Value has the property that after normalization
Therefore, relatively subjective contribution margin can be regarded as pure tone to the percentage contribution of overall whistler；
Step 18: calculating mean square error mse and coefficient of determination R2, model validation is verified,
Wherein, m is total sample number, yiFor sample predictions value,For sample true value；
Fitting result is as shown in Figure 10, and predicted value is generally identical with the trend of true value as can be seen from Figure 10,
Square error only has 3.3 × 10-4, the order of magnitude (1 × 10 relative to percentage contribution-2) very small, there is preferable practicability.
The above is only the preferred embodiment of the present invention, it is noted that for the ordinary person of the art,
Without departing from the principle of the present invention, several improvement can also be made, these improvement also should be regarded as protection model of the invention
It encloses.
Claims (20)
1. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR, which is characterized in that including following step
It is rapid:
Step 1: design orthogonal test:
1.1) audio signal characteristic is counted, choosing Eb3 amplitude A in frequency domain, frequency f, background noise energy features is factor；
1.2) horizontal number being determined according to 3 factors and choosing orthogonal design table, symbiosis is at M test sample；
Step 2: according to orthogonal test generate M test sample generation M group analogue audio sample, and make it is orthogonal-in contrast with
Compared with marking table；
Step 3: choosing n people jury, n > 15 carry out subjective scoring, methods of marking to M analogue audio sample are as follows: imitative to M
True audio sample is compared two-by-two, it is believed that analogue audio sample A ratio B more troublesome, then comparison result is 1；Think to emulate
Audio sample B ratio A more troublesome, then comparison result is -1；Think analogue audio sample A troublesome as B, then compares
It as a result is 0；It carries out altogetherSecondary comparison, 10 pairs of analogue audio samples of every comparison are rested 5 minutes；Scoring process repeats audition, each pair of
Analogue audio sample needs to carry out the two-way audition of AB and BA；Evaluation result is filled in upper the three of orthogonal-pairs of relatively marking table
In angular zone；
Step 4: jury carries out second to M analogue audio sample and scores, and scoring process is the same as step 3；It gives a mark more twice
As a result the content in orthogonal-pairs of relatively marking table is modified；
Step 5: orthogonal-pairs of relatively marking table being subjected to antisymmetry conversion, forms antisymmetric matrix, i.e., orthogonal-pairs of comparison
When upper triangle line n m column marking in table of giving a mark is respectively 1, -1 and 0, lower triangle is accordingly filled out in orthogonal-pairs of relatively marking table
Writing marking is -1,1 and 0, generates complete orthogonal-pairs of relatively marking table；
Step 6: according to n orthogonal-pairs of relatively marking tables, calculating n people and give a mark to M the subjective of sample, form one and beat
Sub-matrix Vn×M；
Step 7: correlation analysis being carried out to the marking of n reviewer, calculates the phase relation between each reviewer's marking
Number, and the subjective of the marking biggish reviewer of error is rejected according to the size of related coefficient and is given a mark；
Step 8: by the scoring matrix V after correlation analysisw×MIt is averaged by column, w≤n, obtains the subjectivity of M sample sound
Give a mark SV1×M；It puts it into orthogonal test table, obtains complete orthogonal test table；
Step 9: range analysis is carried out according to complete orthogonal design table, obtain frequency A, amplitude f, background noise energy Eb tri- because
Element is to index S V1×MInfluence degree, draw factor-indicatrix, observe factor-indicatrix shape, determine regression model: branch
Hold vector regression model；Variance analysis is carried out according to complete orthogonal design table, determines the confidence level of test error Yu each factor；
Step 10: support vector regression constructs scoring model of uttering long and high-pitched sounds: being with the data in the complete orthogonal test table of gained in step 8
Training set, Selecting All Parameters and optimization method carry out regression modeling using libsvm support vector machines program bag；
Step 11: calculating the k meshing frequency and its frequency multiplication for subtracting variable-speed motor actual measurement audio sample, choose filtering bandwidth, design filter
Wave device successively filters out k meshing frequency, obtains k sub- audio sample signals；
Step 12: generating audio according to signal obtained in step 11 and evaluate sample, give a mark for jury；
Step 13: design pure tone contribution subjective assessment marking table, n reviewer successively carry out assessment to audio evaluation sample and beat
Point: comparing audio evaluates the difference of front and back sound in sample, and provides grade scoring；
Step 14: n reviewer of statistics, as a result, with step 7, calculates being averaged for n people to the marking of m audio evaluation sample
Related coefficient will be less than R0The corresponding marking of reviewer reject, calculate the average value C of all remaining reviewer markingi(i
=1,2,3 ..., m), as the subjective contribution margin of meshing frequency pure tone, score value is smaller, contributes bigger；Score value is bigger, and contribution is got over
It is small；
Step 15: calculating in step 11 in g audio sample the corresponding amplitude of k prominent meshing frequency pure tone, frequency and the reality
The background noise energy for surveying audio sample, is obtained m objective parameter sample；The corresponding subjectivity of m sample in step 14 is commented
Value is merged into m objective parameter sample, forms Sm×4Sample set, the sample set is as test set；
Step 16: by test set Sm×4It is put into more pure tone percentage contribution regression mathematical models that step 10 is trained and is predicted,
Obtain the prediction subjectivity contribution margin sv of each pure tonei；
Step 17: subjective contribution margin sv will be predicted by linear compression and normalizationiBe converted to relatively subjective contribution margin；
Step 18: calculating mean square error mse and coefficient of determination R2, model validation is verified,
Wherein, m is total sample number, yiFor sample predictions value,For sample true value；
2. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the step 1- step 10 is orthogonal-pairs of comparative test, for constructing theoretical prediction model.
3. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the step 11- step 14 is that meshing frequency pure tone is used for the subjective scoring method uttered long and high-pitched sounds in measured signal
Quantify different meshing frequencies to the percentage contribution uttered long and high-pitched sounds, verifying of the scoring as theoretical prediction model.
4. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the step 15- step 18 is the inspection of more pure tone percentage contribution regression mathematical models, the life including test set
At, the validation verification of the normalized of prediction result, model.
5. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the definition method of the ambient noise in the step 1.1), ambient noise amplitude are to be K to audio signal frequency spectrum
The amplitude of point sliding average, K point moving average filter:
Yy (1)=y (1)
Yy (2)=(y (1)+y (2)+y (3))/3
Yy (3)=(y (1)+y (2)+y (3)+y (4)+y (5))/M
Yy (4)=(y (2)+y (3)+y (4)+y (5)+y (6))/M
……。
6. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 5,
It is characterized in that: background noise energy Eb calculation formula in the step 1.1) are as follows:
Wherein, N is frequency spectrum points, AiFor ambient noise amplitude.
7. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
It is characterized in that: being taken in the step 1.1) according to the experience of Eb3 amplitude A in frequency domain, frequency f, background noise energy factor
It is worth range, each factor equidistantly chooses p value as level from low to high, and chooses LM(pC) orthogonal test table, M is formed altogether
Secondary test generates M analogue audio signal.
8. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
It is characterized in that: the generation method of the analogue audio sample in the step 2 are as follows:
Sound=Asin (2 π ft)+D*randn (1, N)
A-amplitude
F-frequency
D-random noise standard deviation
Randn (1, N)-generates the random number for meeting standardized normal distribution of 1 row N column
Wherein the relationship of random signal standard deviation D and background noise energy Eb is obtained by Parseval theorem:
Eb=20log10 (D2*N)。
9. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: orthogonal-pairs of relatively marking table in the step 2 illustrates that column, triangle are beaten comprising essential information column, scoring
Divide table and points for attention.
10. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
It is characterized in that: orthogonal-pairs of relatively marking table modification method in the step 4: comparing second of the marking with first time
As a result, being -1 if giving a mark twice one, one is 1, then marking is changed to 0；If giving a mark twice one is 0, one is -1, then will
Marking result is changed to -1；If marking twice one is 1 for 0 one, marking result is changed to 1.
11. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the subjectivity in the step 6 comments marking calculation method: complete orthogonal-pairs of relatively marking each column of table of statistics
In -1 number, using statistical result as the subjective assessment value of M analogue audio sample, statistical result is smaller, utter long and high-pitched sounds and be more obvious,
Statistical result is bigger, utters long and high-pitched sounds more unobvious.
12. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
It is characterized in that: correlation point the subjective scoring data processing method in the step 7: being carried out to the marking of n reviewer
Analysis, using Pearson correlation coefficient:
xi- evaluation personnel x gives a mark to i-th of the subjective of analogue audio sample
yi- evaluation personnel y gives a mark to i-th of the subjective of analogue audio sample
Average value of-evaluation personnel the x to the subjective marking of all analogue audio samples
Average value of-evaluation personnel the y to the subjective marking of all analogue audio samples.
Obtain Rn×nCorrelation coefficient charts, in every a line non-1 n-1 related coefficient is averaged, average phase relation is obtained
Number；N reviewer has n average correlation coefficient, and Trimmed mean related coefficient is less than R0Reviewer marking as a result, 1
> R0> 0.7.
13. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
It is characterized in that: the selection of support vector regression parameter in the step 10: using e-SVR support vector regression model, choosing
Radial basis kernel function, insensitive error take e=0.01.
14. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the selection of support vector regression parameter optimization method in the step 10: using k-fold cross validation method,
It is minimised as optimization aim with cross validation mean square error CVMSE, using grid-search algorithms to penalty factor and nuclear parameter γ
It optimizes, CVMSE formula is as follows:
K-cross validation broken number
N-cross validation points.
15. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the frequency of the step 11 neutron audio sample signal filters out bandwidth choosing method: each meshing frequency filters out
Frequency range be the corresponding critical band of the meshing frequency, critical frequency bandwidth calculation formula is as follows:
Wherein: f0For meshing frequency；ΔfcFor critical frequency bandwidth；
Critical band bound are as follows:
Wherein: f1For critical band lower bound, f2For the critical band upper bound.
16. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the frequency of the step 11 neutron audio sample signal filters out filter design method: design is with audio sample
Frequency fsFor the digital band-reject filter of sampling frequency, performance requirement are as follows: band hinders range from f1To f2, decay at this two frequency
No more than 3dB, in f1- 100Hz and f2Decaying is not less than 20dB at+100Hz frequency；With this filter process audio signal, obtain
K sub- audio signals.
17. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
It is characterized in that: the step 12 sound intermediate frequency evaluation sample generating method: by g actual measurement audio sample and k sub- audio samples
Connection, the clear band of midfeather 1s, difference front and back sound.
18. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the step 13 middle grade methods of marking, comparing audio evaluate the difference of front and back sound in sample, assess consonant
Utter long and high-pitched sounds weakening degree of the frequency sample relative to actual measurement audio sample, point 8 grades: very big -0, greatly -1, big -2, larger -3,
Weak -4, weaker -5, very weak -6, without -7.
19. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: in the step 13 pure tone contribution subjective assessment marking table, by essential information column, grade scoring table, training table,
Formal marking statistical form and 5 part of points for attention composition.
20. a kind of quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR according to claim 1,
Be characterized in that: the calculation method of relatively subjective contribution margin in the step 17, by predicted value linear compression evaluation of estimate 0-7 it
Between,
Wherein: ymax=7, ymin=0
More pure tone ingredients in single audio are normalized:
Value has the property that after normalization
Therefore, regard relatively subjective contribution margin as pure tone to the percentage contribution of overall whistler.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910513521.XA CN110261101B (en) | 2019-06-14 | 2019-06-14 | Howling quantitative evaluation method based on orthogonal-pair comparison test and SVR |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910513521.XA CN110261101B (en) | 2019-06-14 | 2019-06-14 | Howling quantitative evaluation method based on orthogonal-pair comparison test and SVR |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110261101A true CN110261101A (en) | 2019-09-20 |
| CN110261101B CN110261101B (en) | 2021-07-27 |
Family
ID=67918262
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910513521.XA Active CN110261101B (en) | 2019-06-14 | 2019-06-14 | Howling quantitative evaluation method based on orthogonal-pair comparison test and SVR |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110261101B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110782910A (en) * | 2019-11-06 | 2020-02-11 | 大连理工大学 | High-detection-rate squeaking audio detection system |
| CN111309810A (en) * | 2020-02-20 | 2020-06-19 | 苏宁云计算有限公司 | Library and table dividing method and system based on small sample KEY value |
| CN112528500A (en) * | 2020-12-11 | 2021-03-19 | 深兰科技（上海）有限公司 | Evaluation method and evaluation equipment for scene graph construction model |
| CN113053350A (en) * | 2021-03-14 | 2021-06-29 | 西北工业大学 | Active control error filter design method based on noise subjective evaluation suppression |
| CN114034490A (en) * | 2021-10-13 | 2022-02-11 | 江铃汽车股份有限公司 | Method for rapidly identifying drive axle squeaking |
| CN115795899A (en) * | 2022-12-12 | 2023-03-14 | 博格华纳汽车零部件（武汉）有限公司 | New energy electric vehicle squeaking noise evaluation method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103323101A (en) * | 2013-06-04 | 2013-09-25 | 格特拉克（江西）传动系统有限公司 | Method of evaluating transmission gear noise quantificationally and objectively by means of pitch |
| KR101314043B1 (en) * | 2012-07-30 | 2013-10-01 | 인하대학교 산학협력단 | Method for sound quality analysis of axle gear whine sound and apparatus thereof |
| CN104697786A (en) * | 2015-03-17 | 2015-06-10 | 重庆长安汽车股份有限公司 | Objective evaluation method of transmission whine noise |
| CN106959159A (en) * | 2017-05-31 | 2017-07-18 | 重庆长安汽车股份有限公司 | Method based on order hump amount quantitative assessment AT transmission gear whistlers |
| CN107664579A (en) * | 2017-11-20 | 2018-02-06 | 哈尔滨东安汽车发动机制造有限公司 | Transmission gear, which is uttered long and high-pitched sounds, engages order extraction and decision method |
| CN109241838A (en) * | 2018-07-27 | 2019-01-18 | 重庆理工大学 | Speed changer quality evaluating method based on psychologic acoustics objective parameter |
| CN109297718A (en) * | 2018-09-29 | 2019-02-01 | 重庆长安汽车股份有限公司 | A kind of evaluation method of order whistler |
-
2019
- 2019-06-14 CN CN201910513521.XA patent/CN110261101B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101314043B1 (en) * | 2012-07-30 | 2013-10-01 | 인하대학교 산학협력단 | Method for sound quality analysis of axle gear whine sound and apparatus thereof |
| CN103323101A (en) * | 2013-06-04 | 2013-09-25 | 格特拉克（江西）传动系统有限公司 | Method of evaluating transmission gear noise quantificationally and objectively by means of pitch |
| CN104697786A (en) * | 2015-03-17 | 2015-06-10 | 重庆长安汽车股份有限公司 | Objective evaluation method of transmission whine noise |
| CN106959159A (en) * | 2017-05-31 | 2017-07-18 | 重庆长安汽车股份有限公司 | Method based on order hump amount quantitative assessment AT transmission gear whistlers |
| CN107664579A (en) * | 2017-11-20 | 2018-02-06 | 哈尔滨东安汽车发动机制造有限公司 | Transmission gear, which is uttered long and high-pitched sounds, engages order extraction and decision method |
| CN109241838A (en) * | 2018-07-27 | 2019-01-18 | 重庆理工大学 | Speed changer quality evaluating method based on psychologic acoustics objective parameter |
| CN109297718A (en) * | 2018-09-29 | 2019-02-01 | 重庆长安汽车股份有限公司 | A kind of evaluation method of order whistler |
Non-Patent Citations (2)
| Title |
|---|
| 周益等: "齿轮精度与变速器啸叫声控制的定量分析", 《噪声与振动控制》 * |
| 臧孟炎等: "基于正交试验设计的变速器啸叫特性优化", 《汽车工程》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110782910A (en) * | 2019-11-06 | 2020-02-11 | 大连理工大学 | High-detection-rate squeaking audio detection system |
| CN110782910B (en) * | 2019-11-06 | 2022-05-20 | 大连理工大学 | High-detection-rate squeal audio detection system |
| CN111309810A (en) * | 2020-02-20 | 2020-06-19 | 苏宁云计算有限公司 | Library and table dividing method and system based on small sample KEY value |
| CN111309810B (en) * | 2020-02-20 | 2022-11-18 | 苏宁云计算有限公司 | Library and table dividing method and system based on small sample KEY value |
| CN112528500A (en) * | 2020-12-11 | 2021-03-19 | 深兰科技（上海）有限公司 | Evaluation method and evaluation equipment for scene graph construction model |
| CN112528500B (en) * | 2020-12-11 | 2023-08-29 | 深兰人工智能应用研究院(山东)有限公司 | Evaluation method and evaluation equipment for scene graph construction model |
| CN113053350A (en) * | 2021-03-14 | 2021-06-29 | 西北工业大学 | Active control error filter design method based on noise subjective evaluation suppression |
| CN113053350B (en) * | 2021-03-14 | 2023-11-17 | 西北工业大学 | Active control error filter design method based on noise subjective evaluation suppression |
| CN114034490A (en) * | 2021-10-13 | 2022-02-11 | 江铃汽车股份有限公司 | Method for rapidly identifying drive axle squeaking |
| CN114034490B (en) * | 2021-10-13 | 2023-10-20 | 江铃汽车股份有限公司 | Method for rapidly identifying drive axle howling |
| CN115795899A (en) * | 2022-12-12 | 2023-03-14 | 博格华纳汽车零部件（武汉）有限公司 | New energy electric vehicle squeaking noise evaluation method |
| CN115795899B (en) * | 2022-12-12 | 2023-09-26 | 博格华纳汽车零部件（武汉）有限公司 | New energy electric automobile howling noise evaluation method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110261101B (en) | 2021-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110261101A (en) | Quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR | |
| Houtgast et al. | Predicting speech intelligibility in rooms from the modulation transfer function. I. General room acoustics | |
| CN102324229B (en) | Method and system for detecting abnormal use of voice input equipment | |
| Yang et al. | Psychoacoustical evaluation of natural and urban sounds in soundscapes | |
| CN109920439A (en) | The variable-speed motor that subtracts based on tone energy and human ear frequency selectivity is uttered long and high-pitched sounds evaluation method | |
| CN105163262B (en) | A kind of loudspeaker sound detection method and detecting system | |
| CN107358966A (en) | Based on deep learning speech enhan-cement without reference voice quality objective evaluation method | |
| CN109668626A (en) | A kind of sound quality evaluation method based on human-computer interaction interface | |
| CN103413547B (en) | A kind of method that room reverberation is eliminated | |
| CN109524020A (en) | A kind of speech enhan-cement processing method | |
| CN105989853A (en) | Audio quality evaluation method and system | |
| CN101896965A (en) | Method and system for speech intelligibility measurement of an audio transmission system | |
| CN106525435A (en) | Engine noise source identification method | |
| CN104658543A (en) | Method for eliminating indoor reverberation | |
| Liu et al. | Strategy and implementing techniques for the sound quality target of car interior noise during acceleration | |
| CN106663450A (en) | Method of and apparatus for evaluating quality of a degraded speech signal | |
| Shrivastav et al. | A model for the prediction of breathiness in vowels | |
| CN109241838A (en) | Speed changer quality evaluating method based on psychologic acoustics objective parameter | |
| Godin et al. | Glottal waveform analysis of physical task stress speech | |
| CN108735192A (en) | A kind of piano performance assessment of acoustics system and method for combination style of song | |
| KR101314043B1 (en) | Method for sound quality analysis of axle gear whine sound and apparatus thereof | |
| CN106725289B (en) | A kind of vehicle sound quality evaluation method based on pupil diameter | |
| Lindh et al. | Humans versus Machine: Forensic Voice Comparison on a Small Database of Swedish Voice Recordings. | |
| CN116933620A (en) | Noise quality evaluation and model building method | |
| Mitchell et al. | The effects of open throat technique on long term average spectra (LTAS) of female classical voices |
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 |