Split type air conditioner sound quality evaluation method based on double neural network subjective and objective fusion model

Abstract

The invention discloses a split-type air conditioner sound quality evaluation method based on a double-neural network subjective and objective fusion model, and the method comprises the steps: 1, collecting two types of indoor / outdoor noise sources, and dividing working conditions according to low / medium / high wind speeds; 2, constructing a two-dimensional subjective label set, labeling samples as well as / not as well as binary labels, performing subjective scoring on each group of samples, and obtaining an average score after exception elimination; 3, four types of psychological acoustic parameters of the sample are extracted to form 96-dimensional features, and optimal objective features are obtained through redundancy elimination and principal component dimensionality reduction; and 4, constructing a working condition-based double neural network, respectively training a binary classification model and a score prediction model, and after MAE and RMSE verification are carried out to reach the standard, fusing a binary result and initial score adjustment to output a final score. According to the method, accurate mapping of objective features and subjective perception is established, evaluation is promoted to be changed from experience guiding to data driving, and support is provided for accurate and standardized evaluation of the sound quality of the air conditioner.

Description

Technical Field

[0001] This invention relates to the field of air conditioner noise evaluation technology, and in particular to a method for evaluating the sound quality of split-type air conditioners based on a dual neural network subjective-objective fusion model, which is applicable to the accurate quantification and standardized evaluation of air conditioner noise. Background Technology

[0002] The sound quality of air conditioner operating noise directly affects user experience. Traditional evaluation methods have two limitations: first, they rely on subjective ratings, which are greatly affected by individual differences among evaluators and have poor consistency; second, they only evaluate based on single objective parameters such as sound pressure level, which cannot reflect the complex perception of human hearing. Existing subjective-objective fusion methods mostly use simple weighting or single-model mapping, which makes it difficult to establish a deep correlation between objective features and subjective perception, resulting in limited evaluation accuracy. Therefore, this invention proposes a dual-neural network subjective-objective fusion evaluation model construction method. By using a binary classification and scoring prediction model trained under different operating conditions, combined with a dynamic adjustment strategy, it achieves accurate and coordinated representation of objective parameters and subjective perception, overcoming the limitations of traditional methods. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention provides a method for evaluating the sound quality of split-type air conditioners based on a dual neural network subjective-objective fusion model, aiming to achieve a deep mapping between objective acoustic characteristics and human subjective perception, thereby improving the accuracy and standardization of the evaluation.

[0004] To achieve the above objectives, the present invention adopts the following technical solution. The present invention provides a method for evaluating the sound quality of split-type air conditioners based on a dual neural network subjective-objective fusion model, characterized by the following steps: Step 1: Collect noise signals from a split-type air conditioner under typical operating conditions, and denot any type of noise source as the [number]. Noise sources, when When, it indicates an indoor noise source. When, it indicates an outdoor noise source; Based on the actual fan speed setting of the split-type air conditioner, the first Wind speed settings for noise signals are divided into the first... One gear, when When, it indicates the low wind speed setting. When, it indicates the stroke gear; when When, it indicates the high wind speed setting; Collection of the first The first type of noise source Each wind speed setting Below A set of noise signals is used to form a working condition tag set. ; Step 2: Settings The binary label is Thus obtain binary tag library ;when When, it means This is a substandard noise signal; when When, it means This is a passing noise signal; Step 3: Construct a scoring indicator library ; Step 4: Calculation In the Key psychoacoustic features in the Bark band Thus, the key psychoacoustic feature matrix is ​​obtained. ;in, for In the A specific loudness on a Bark frequency band for In the Sharpness on each Bark frequency band for In the Roughness on each Bark frequency band for In the jitter in each Bark frequency band , Indicates the total number of Bark bands; Step 5: [Regarding...] Redundant feature removal and principal component analysis were performed to reduce the dimensionality of the acoustic feature matrix. ;in, Indicates the first The dimension-reduced acoustic feature vector corresponding to each noise signal; Step 6: Construct a binary classification neural network and perform... Processing is performed to obtain The predicted pass rate, and compared with Construct a loss function to train a binary classification neural network and obtain the optimal binary classification noise model; Step 7: Construct a subjective rating prediction neural network and perform... Processing is performed to obtain The predicted score, and with Construct a loss function to train the subjective rating neural network and obtain the optimal subjective rating model; Step 8: Transfer the new noise signal After processing according to steps 4-5, the data is input into the optimal binary classification noise model and the optimal subjective rating model for further processing, and the corresponding outputs are obtained. Predicted pass rate and initial prediction score ; Step 9: If Then determine Predicted labels Otherwise, determine Predicted labels ; when At that time, if Then As Final score prediction ;in, This indicates the preset passing score; when At that time, if Then, according to equation (1), Adjustments are made to obtain the adjusted predicted score. , and as ; (1) In equation (1), express The upper limit, and has ; when At that time, if Then As Final score prediction ; when At that time, if Then, according to equation (2), Adjustments are made to obtain the adjusted predicted score. , and as ; (2) In equation (2), express The lower limit, and has ; The characteristic of the sound quality evaluation method for split-type air conditioners based on a dual neural network subjective-objective fusion model described in this invention is that step 3 includes: Step 3.1: For conduct The second rating received rating set ;in, express The Second rating; Step 3.2: Calculation average and standard deviation And the Grubbs test was used to remove them. After scoring the anomalies, calculate The mean of the remaining ratings Thus, a scoring index library is obtained. .

[0005] Furthermore, step 3.2 includes: Step 3.2.1: Calculate using equations (3) and (4) respectively. average and standard deviation : (3) (4) Step 3.2.2: Calculate using equation (5) Grubbs statistic : (5) like Then determine Scoring and removing abnormal entries; otherwise, This is represented as a valid score, thus obtaining... The set of scores after elimination ,in, At the significance level, Indicates the number of samples is And the significance level is Grubbs critical value at time; Step 3.2.3: Calculate using equation (6) Average rating ; (6) In equation (4), for The number of valid scores. , for The Middle One valid score.

[0006] Furthermore, step 4 includes: Step 4.1: Calculate using equation (7) In the Specific loudness in a Bark frequency band : (7) In equation (7), express In the Sound excitation on the Bark frequency band Indicates the first Absolute threshold excitation in each Bark frequency band Indicates reference incentive; Step 4.2: Calculate using equation (8) In the Sharpness on the Bark band : (8) In equation (6), Indicates the first The center Bark value of each Bark frequency band; Step 4.3: Calculate using equation (9) In the Roughness in each Bark frequency band : (9) In equation (7), express In the Modulation depth in each Bark band express In the Characteristic modulation frequencies in each Bark band; Step 4.4: Calculate using equation (10) In the jitter in the Bark frequency band : (10) In equation (8), express In the Frequency modulation amplitude on each Bark band Indicates the first The center frequency of the Bark band.

[0007] Furthermore, step 5 includes: Step 5.1: Let The total number of noise signals in the data is Then the first The acoustic feature vector corresponding to each noise signal is denoted as . ,in, Indicates the first The first noise signal corresponding to the Acoustic characteristics, ; Calculate using equation (11) The a-th acoustic feature With the bth characteristic acoustic feature Pearson correlation coefficient between ,like Then remove or Otherwise, keep and Thus, the acoustic feature matrix after removal is obtained. , This represents the number of acoustic features after removal. This is the correlation threshold; (11) In equation (11), express The noise signal at the ... The average value over each feature Indicates the first The noise signal of the first One characteristic, express The noise signal at the ... The average value over each feature.

[0008] Step 5.2: Use equation (12) to... The Middle The noise signal of the first Acoustic feature value Decentralization is performed to obtain the first The noise signal at the ... Standardized eigenvalues ​​of each acoustic feature Thus, the standardized acoustic feature matrix is ​​obtained. ; (12) In equation (12), express The Middle The mean of the acoustic features, express The Middle Standard deviation of acoustic characteristics; Step 5.3: Calculate using equation (13) covariance matrix ; (13) In equation (13), Indicates transpose; Step 5.4: Use equation (14) to... Perform eigenvalue decomposition to obtain eigenvalue sequence and the corresponding feature vectors ; (14) In equation (12), for The 1 eigenvector for The 1 eigenvalue, and ; Step 5.5: Calculate the cumulative variance contribution rate using equation (15). Select the option that satisfies the cumulative variance contribution rate. The smallest integer Thus constructing the former Constructing a projection matrix from acoustic feature vectors ;in, The cumulative variance contribution rate threshold. The number of main components; (15) Step 5.6: [The sentence is incomplete and requires more context to be translated accurately.] After projection onto the principal component space, the dimension-reduced acoustic feature matrix is ​​obtained. ,and .

[0009] The present invention provides an electronic device, including a memory and a processor, characterized in that the memory is used to store a program supporting the processor in performing the method described therein, and the processor is configured to execute the program stored in the memory.

[0010] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program is executed by a processor to perform the steps of the method described thereon.

[0011] The present invention has the following advantages over the prior art: 1. Existing technologies mostly adopt a shallow fusion method of simple weighting of objective parameters and subjective scoring or single-model mapping, which makes it difficult to associate the complex relationship between objective features and subjective perception. This invention constructs a dual-dimensional label system of professional binary judgment and subjective quantitative scoring, combined with a dual neural network trained under different working conditions, and introduces a dynamic adjustment strategy for prediction probability and working condition score statistics to achieve a deep collaborative representation of objective parameters and subjective perception, effectively eliminating the bias of single-dimensional evaluation.

[0012] 2. Existing technologies often ignore the differences in noise characteristics of different noise sources and different fan speed settings of air conditioners, and use a uniform model for evaluation, which leads to large errors under specific operating conditions. This invention independently constructs models for 6 typical operating conditions to accurately match the differences in psychoacoustic characteristics of noise under each operating condition.

[0013] 3. Existing technologies often rely on single objective parameters such as sound pressure level and sound power level, or fail to perform redundant processing on high-dimensional features, resulting in one-sided feature information or complex model calculations. This invention extracts 24 Bark frequency band features of 4 core parameters based on psychoacoustic theory, then removes redundant features through Pearson correlation coefficient, and obtains the optimal low-dimensional features by combining principal component analysis. This ensures the integrity of feature information, reduces the computational complexity of neural networks, and avoids overfitting. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0015] In this embodiment, a method for evaluating the sound quality of a split-type air conditioner based on a dual-neural network subjective-objective fusion model addresses the core pain points of existing air conditioner sound quality evaluation technologies, such as shallow subjective-objective fusion and poor adaptability to operating conditions. It constructs a systematic dual-neural network subjective-objective fusion model, achieving a deep mapping between objective parameters and subjective perception through dual-dimensional labels and dual models for different operating conditions. This solves the problem of single-dimensional evaluation bias and adapts to the differences in operating conditions of different noise sources and fan speed settings, significantly improving evaluation accuracy and robustness. Specifically, as... Figure 1 As shown, the method includes the following steps: Step 1: Collect noise signals from a split-type air conditioner under typical operating conditions, and denot any type of noise source as the [number]. Noise sources, when When, it indicates an indoor noise source. When, it indicates an outdoor noise source; Based on the actual fan speed setting of the split-type air conditioner, the first Wind speed settings for noise signals are divided into the first... One gear, when When, it indicates the low wind speed setting. When, it indicates the stroke gear; when When, it indicates the high wind speed setting; Collection of the first The first type of noise source Each wind speed setting Below A set of noise signals is used to form a working condition tag set. ; In this example, a semi-anechoic chamber was selected as the data collection environment. The background noise level in this environment is ≤25dB(A), which eliminates the interference of external environmental noise on the air conditioner noise sample. For a certain brand of 1.5HP split-type air conditioner, based on the indoor unit... / Outdoor unit Low wind Wind speed 1.2 m / s / moderate stroke Wind speed 2.5 m / s / high wind The wind speed was 3.8 m / s, and the data was collected continuously under six different operating conditions. Group sample.

[0016] Step 2: Settings The binary label is Thus obtain binary tag library ;when When, it means This is a substandard noise signal; when When, it means This is a passing noise signal; The 180 samples from 6 working conditions were played in random order, and the evaluators independently labeled them "1 (pass) / 0 (fail)," with the labeling results recorded in an Excel spreadsheet. If all 5 evaluators agreed, the result was taken as the final label; if one evaluator disagreed, the sample was played again and discussed before a final label was determined. In this embodiment, none of the 180 samples had two or more disagreements, thus forming a binary label library L.

[0017] Step 3: Construct a scoring indicator library ; Step 3.1: For conduct The second rating received rating set ;in, express The Second rating; Step 3.2: Calculation average and standard deviation And the Grubbs test was used to remove them. After scoring the anomalies, calculate The mean of the remaining ratings Thus, a scoring index library is obtained. .

[0018] Step 3.2.1: Calculate using equations (3) and (4) respectively. average and standard deviation : (3) (4) Step 3.2.2: Calculate using equation (5) Grubbs statistic : (5) like Then determine Scoring and removing abnormal entries; otherwise, This is represented as a valid score, thus obtaining... The set of scores after elimination ,in, At the significance level, Indicates the number of samples is The significance level is The critical value of Grubbs.

[0019] Step 3.2.3: Calculate using equation (6) Average rating ; (6) In equation (4), for The number of valid scores. , for The Middle One valid score.

[0020] In the example, the noise signal of group 8 in the indoor unit's fan baffle is used. For example, if the rating is given 15 times, the results of the 15 ratings are: 58, 62, 60, 59, 61, 57, 95, 63, 60, 58, 62, 61, 59, 60, 58. The arithmetic mean can be obtained according to equation (1). The standard deviation can be obtained from equation (2). By consulting the Grubbs test critical value table, we can obtain... According to formula (3), 95 points are obtained as abnormal scores and are removed; the average score of the remaining 14 valid scores is... This serves as the final subjective score for the noise signal.

[0021] Step 4: Calculation In the Key psychoacoustic features in the Bark band Thus obtain exist Key psychoacoustic feature sequences in the Bark frequency band This leads to the acquisition of the key psychoacoustic feature matrix. ;in, for In the A specific loudness on a Bark frequency band for In the Sharpness on each Bark frequency band for In the Roughness on each Bark frequency band for In the jitter in each Bark frequency band , This indicates the total number of Bark bands.

[0022] make The total number of noise signals in the data is Then the first The acoustic feature vector corresponding to each noise signal is denoted as . ,in, Indicates the first The first noise signal corresponding to the Acoustic characteristics, ; Total sample size Each sample contains 4 parameters × 24 Bark bands = 96 features, therefore the overall feature matrix is... This is an array of 180 rows × 96 columns. During construction, all sample files are iterated through, 96 feature values ​​are calculated for each sample, and then filled in. , Corresponding to the The 96-dimensional feature vector of each sample .

[0023] Step 4.1: Calculate using equation (7) In the Specific loudness in a Bark frequency band : (7) In equation (7), express In the Sound excitation on the Bark frequency band Indicates the first Absolute threshold excitation in each Bark frequency band This indicates a reference incentive.

[0024] Step 4.2: Calculate using equation (8) In the Sharpness on the Bark band : (8) In equation (6), Indicates the first The center Bark value of each Bark frequency band.

[0025] Step 4.3: Calculate using equation (9) In the Roughness in each Bark frequency band : (9) In equation (7), express In the Modulation depth in each Bark band express In the Characteristic modulation frequencies on a Bark band.

[0026] Step 4.4: Calculate using equation (10) In the jitter in the Bark frequency band : (10) In equation (8), express In the Frequency modulation amplitude on each Bark band Indicates the first The center frequency of the Bark band.

[0027] Step 5: [Regarding...] Redundant feature removal and principal component analysis were performed to reduce the dimensionality of the acoustic feature matrix. ;in, Indicates the first The dimension-reduced acoustic feature vector corresponding to each noise signal; Step 5.1: Calculate using equation (11) The a-th acoustic feature With the bth characteristic acoustic feature Pearson correlation coefficient between ,like Then remove or Otherwise, keep and Thus, the acoustic feature matrix after removal is obtained. , This represents the number of acoustic features after removal. This is the correlation threshold; (11) In equation (11), express The noise signal at the ... The average value over each feature Indicates the first The noise signal of the first One characteristic, express The noise signal at the ... The average value over each feature.

[0028] Set correlation threshold 31 pairs of features were obtained. For each pair of highly correlated features, the one with the smaller variance is removed, and finally, the remaining features are... These features form a feature matrix Ω'.

[0029] Step 5.2: Use equation (12) to... The Middle The noise signal of the first Acoustic feature value Decentralization is performed to obtain the first The noise signal at the ... Standardized eigenvalues ​​of each acoustic feature Thus, the standardized acoustic feature matrix is ​​obtained. ; (12) In equation (12), express The Middle The mean of the acoustic features, express The Middle Standard deviation of acoustic characteristics; by Taking the 20th feature as an example, this feature's , For the 50th sample The standardized value is calculated according to equation (10): Calculate all features iteratively to obtain... .

[0030] Step 5.3: Calculate using equation (13) covariance matrix ; (13) In equation (13), This indicates transpose.

[0031] Step 5.4: Use equation (14) to... Perform eigenvalue decomposition to obtain eigenvalue sequence and the corresponding feature vectors ; (14) In equation (12), for The 1 eigenvector for The 1 eigenvalue, and .

[0032] Step 5.5: Calculate the cumulative variance contribution rate using equation (15). Select the option that satisfies the cumulative variance contribution rate. The smallest integer Thus constructing the former Constructing a projection matrix from acoustic feature vectors ;in, The cumulative variance contribution rate threshold. The number of main components; (15) Step 5.6: [The sentence is incomplete and requires more context to be translated accurately.] After projection onto the principal component space, the dimension-reduced acoustic feature matrix is ​​obtained. ,and ; Cumulative variance contribution rate threshold The cumulative variance contribution rate is 0.9, calculated according to equation (13). It is 12 o'clock. Therefore ; Construct a projection matrix based on the first 12 eigenvectors. ,according to Calculate the optimal eigenvector , The 12-dimensional reduced features corresponding to the q-th sample.

[0033] Step 6: Construct a binary classification neural network and perform... Processing is performed to obtain The predicted pass rate, and compared with A loss function is constructed to train a binary classification neural network, resulting in the optimal binary classification noise model.

[0034] Step 7: Construct a subjective rating prediction neural network and perform... Process it to obtain the predicted score, and compare it with Construct a loss function to train the subjective scoring neural network and obtain the optimal subjective scoring model; Step 8: After processing the new noise signal according to the processing in Steps 4 - 5, input it into the optimal binary classification noise model and the optimal subjective scoring model for processing respectively, and output the corresponding predicted passing rate and the initial predicted score .

[0035] Calculate the mean absolute error MAE of using Equation (16), and calculate the root mean square error RMSE of using Equation (17). If corresponding to the optimal binary classification noise model and the optimal subjective scoring model and , it is determined that the model accuracy meets the standard. Otherwise, adjust the number of nodes in the hidden layer and the learning rate of the model and retrain until the accuracy meets the standard. is the preset threshold for the mean absolute error, is the preset threshold for the root mean square error; (14) (15) According to the subjective scoring error tolerance, set the MAE preset threshold = 5, allowable error ≤ 5 points, RMSE preset threshold = 8, 95% prediction error ≤ 8 points.

[0036] Take the indoor unit medium wind speed scoring model as an example. , , it is determined that the model is qualified; for the outdoor unit high wind speed scoring model after initial training , , it is determined that the model is unqualified. Increase the number of nodes in hidden layer 1 from 15 to 20, reduce the learning rate from 0.01 to 0.005, and retrain for 38 rounds , RMSE = 6.8 ≤ 8, the model is qualified; Step 9: If , then determine the predicted label of , otherwise, determine the predicted label of ; When , if , then take as the final score prediction value ;in, This indicates the preset passing score; when At that time, if Then, according to equation (1), Adjustments are made to obtain the adjusted predicted score. , and as ; (1) In equation (1), express The upper limit, and has ; when At that time, if Then As Final score prediction ; when At that time, if Then, according to equation (2), Adjustments are made to obtain the adjusted predicted score. , and as ; (2) In equation (2), express The lower limit, and has ; After processing the windshield noise signal of the new outdoor unit according to the operating conditions in steps 4-5, we obtain... And input it into the optimal binary classification noise model, output ,determination Input into the optimal subjective rating model, output ;calculate The average rating is According to equation (1), we get Therefore, .

[0037] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0038] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A method for evaluating the sound quality of split-type air conditioners based on a dual-neural network subjective-objective fusion model, characterized in that, Includes the following steps: Step 1: Collect noise signals from a split-type air conditioner under typical operating conditions, and denot any type of noise source as the [number]. Noise sources, when When, it indicates an indoor noise source. When, it indicates an outdoor noise source; Based on the actual fan speed setting of the split-type air conditioner, the first Wind speed settings for noise signals are divided into the first... One gear, when When, it indicates the low wind speed setting. When, it indicates the stroke gear; when When, it indicates the high wind speed setting; Collection of the first The first type of noise source Each wind speed setting Below A set of noise signals is used to form a working condition tag set. ; Step 2: Settings The binary label is Thus obtain binary tag library ;when When, it means This is a substandard noise signal; when When, it means This is a passing noise signal; Step 3: Construct a scoring indicator library ; Step 4: Calculation In the Key psychoacoustic features in the Bark band Thus, the key psychoacoustic feature matrix is ​​obtained. ;in, for In the A specific loudness on a Bark frequency band for In the Sharpness on each Bark frequency band for In the Roughness in each Bark frequency band for In the jitter in each Bark frequency band , Indicates the total number of Bark bands; Step 5: [Regarding...] Redundant feature removal and principal component analysis were performed to reduce the dimensionality of the acoustic feature matrix. ;in, Indicates the first The dimension-reduced acoustic feature vector corresponding to each noise signal; Step 6: Construct a binary classification neural network and perform... Processing is performed to obtain The predicted pass rate, and compared with Construct a loss function to train a binary classification neural network and obtain the optimal binary classification noise model; Step 7: Construct a subjective rating prediction neural network and perform... Processing is performed to obtain The predicted score, and with Construct a loss function to train the subjective rating neural network and obtain the optimal subjective rating model; Step 8: Transfer the new noise signal After processing according to steps 4-5, the data is input into the optimal binary classification noise model and the optimal subjective rating model for further processing, and the corresponding outputs are obtained. Predicted pass rate and initial prediction score ; Step 9: If Then determine Predicted labels Otherwise, determine Predicted labels ; when At that time, if Then As Final score prediction ;in, This indicates the preset passing score; when At that time, if Then, according to equation (1), Adjustments are made to obtain the adjusted predicted score. , and as ; (1) In equation (1), express The upper limit, and has ; when At that time, if Then As Final score prediction ; when At that time, if Then, according to equation (2), Adjustments are made to obtain the adjusted predicted score. , and as ; (2) In equation (2), express The lower limit, and has .

2. The method for evaluating the sound quality of a split-type air conditioner based on a dual neural network subjective-objective fusion model according to claim 1, characterized in that, Step 3 includes: Step 3.1: For conduct The second rating received rating set ;in, express The Second rating; Step 3.2: Calculation average and standard deviation And the Grubbs test was used to remove them. After scoring the anomalies, calculate The mean of the remaining ratings Thus, a scoring index library is obtained. .

3. The method for evaluating the sound quality of split-type air conditioners based on a dual neural network subjective-objective fusion model according to claim 2, characterized in that, Step 3.2 includes: Step 3.2.1: Calculate using equations (3) and (4) respectively. average and standard deviation : (3) (4) Step 3.2.2: Calculate using equation (5) Grubbs statistic : (5) like Then determine Scoring and removing abnormal entries; otherwise, This is represented as a valid score, thus obtaining... The set of scores after elimination ,in, At the significance level, Indicates the number of samples is And the significance level is Grubbs critical value at time; Step 3.2.3: Calculate using equation (6) Average rating ; (6) In equation (4), for The number of valid scores. , for The Middle One valid score.

4. The method for evaluating the sound quality of a split-type air conditioner based on a dual neural network subjective-objective fusion model according to claim 1, characterized in that, Step 4 includes: Step 4.1: Calculate using equation (7) In the Specific loudness in a Bark frequency band : (7) In equation (7), express In the Sound excitation on the Bark frequency band Indicates the first Absolute threshold excitation in each Bark frequency band Indicates reference incentive; Step 4.2: Calculate using equation (8) In the Sharpness on the Bark band : (8) In equation (6), Indicates the first The center Bark value of each Bark frequency band; Step 4.3: Calculate using equation (9) In the Roughness in each Bark frequency band : (9) In equation (7), express In the Modulation depth in each Bark band express In the Characteristic modulation frequencies in each Bark band; Step 4.4: Calculate using equation (10) In the jitter in the Bark frequency band : (10) In equation (8), express In the Frequency modulation amplitude on each Bark band Indicates the first The center frequency of the Bark band.

5. The method for evaluating the sound quality of a split-type air conditioner based on a dual neural network subjective-objective fusion model according to claim 4, characterized in that, Step 5 includes: Step 5.1: Let The total number of noise signals in the data is Then the first The acoustic feature vector corresponding to each noise signal is denoted as . ,in, Indicates the first The first noise signal corresponding to the Acoustic characteristics, ; Calculate using equation (11) The a-th acoustic feature With the bth characteristic acoustic feature Pearson correlation coefficient between ,like Then remove or Otherwise, keep and Thus, the acoustic feature matrix after removal is obtained. , This represents the number of acoustic features after removal. This is the correlation threshold; (11) In equation (11), express The noise signal at the ... The average value over each feature Indicates the first The noise signal of the first One characteristic, express The noise signal at the ... The average value over each feature; Step 5.2: Use equation (12) to... The Middle The noise signal of the first Acoustic feature value Decentralization is performed to obtain the first The noise signal at the ... Standardized eigenvalues ​​of each acoustic feature Thus, the standardized acoustic feature matrix is ​​obtained. ; (12) In equation (12), express The Middle The mean of the acoustic features, express The Middle Standard deviation of acoustic characteristics; Step 5.3: Calculate using equation (13) covariance matrix ; (13) In equation (13), Indicates transpose; Step 5.4: Use equation (14) to... Perform eigenvalue decomposition to obtain eigenvalue sequence and the corresponding feature vectors ; (14) In equation (12), for The 1 eigenvector for The 1 eigenvalue, and ; Step 5.5: Calculate the cumulative variance contribution rate using equation (15). Select the option that satisfies the cumulative variance contribution rate. The smallest integer Thus constructing the former Constructing a projection matrix from acoustic feature vectors ;in, The cumulative variance contribution rate threshold. The number of main components; (15) Step 5.6: [The sentence is incomplete and requires more context to be translated accurately.] After projection onto the principal component space, the dimension-reduced acoustic feature matrix is ​​obtained. ,and .

6. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports a processor in executing the method of any one of claims 1-5, the processor being configured to execute the program stored in the memory.

7. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the steps of the method according to any one of claims 1-5.

Images