Environmental noise evaluation method based on annoyance perception index

Abstract

The invention provides an environmental noise evaluation method based on an annoyance perception index. The method comprises the following steps: determining an annoyance degree evaluation value of a sample according to the obtained environmental noise sample; training an XGBoost model by taking the environmental noise sample and the annoyance degree evaluation value as training samples to obtain a trained model; obtaining a to-be-tested original audio signal, and inputting the original audio signal into the trained model to obtain an annoyance degree evaluation value of the environmental noise; converting the environmental noise annoyance degree evaluation value into a sound source annoyance degree of the environmental noise according to the age of the target crowd; and finally, weighting the annoyance degree generated by a sound source and the annoyance degree generated by a sound level in the environmental noise, and calculating the annoyance degree of the environmental noise by taking decibel as a measurement unit. And the influence generated by the environmental noise can be described more accurately by taking the annoyance degree of people as an index.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental noise evaluation, and in particular to an environmental noise evaluation method based on annoyance perception index. Background Art

[0002] As cities move towards ecological goals, environmental noise includes more and more natural, cultural and other sound sources that people like to hear and hear. The content of environmental noise is no longer the environmental noise of industrialized cities. Although traffic noise, mechanical noise, and noisy urban life noise are still the main environmental noise in ecological cities, natural sounds and human sounds can no longer be ignored. Noise evaluation technology that simply regards all environmental noise sources as pollution sources can no longer reflect the annoying results that environmental noise brings to people. At present, the high level of noise complaints and the declining noise monitoring data year by year reflect that traditional environmental noise evaluation can no longer meet the needs of social development.

[0003] In the 21st century sound environment quality guidance values, the World Health Organization uses "annoyance" as an important indicator to measure the community noise environment. Soundscape research points out that the quality of the sound environment depends on people's perception of environmental noise. Annoyance is a key indicator for evaluating the quality of the sound environment and is the connotation of noise. Therefore, the basic standard for evaluating environmental noise is the irritation perception caused by environmental noise. Unlike traditional noise, which uses sound level as the only evaluation factor, soundscape uses all factors that cause people's annoyance perception as noise evaluation factors. Among them, the source of environmental noise has been proven to play a key role, especially when the environmental noise level is in the range of 50dB-70dB. A large number of studies have shown that there are significant differences in the annoyance perception brought to people by environmental noise with different semantics. When evaluating environmental noise, the traditional method of using equivalent sound level as the main evaluation indicator can no longer meet the needs of modern ecological cities.

[0004] In order to distinguish between the natural sounds and social and cultural sounds that people like and the urban traffic, production and life sounds that they hate, evaluating noise based on people's annoyance perception of environmental noise is the key to measuring the quality of the acoustic environment in an ecological city. In response to this demand, it is particularly important to invent a more accurate environmental noise evaluation method based on ecological and environmental conditions. Summary of the invention

[0005] In order to overcome the deficiencies of the prior art, an object of the present invention is to provide an environmental noise evaluation method based on annoyance perception index, which can more accurately describe the annoyance level of environmental noise.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] An environmental noise evaluation method based on annoyance perception index, comprising:

[0008] Determine an annoyance evaluation value of the sample according to the acquired environmental noise sample;

[0009] The environmental noise samples and the annoyance evaluation values ​​are used as training samples to train the XGBoost model to obtain a trained environmental noise annoyance evaluation value prediction model;

[0010] Acquire an original audio signal to be tested, and input the original audio signal into the environmental noise annoyance evaluation value prediction model to obtain an annoyance evaluation value of the environmental noise;

[0011] Converting the annoyance evaluation value of the environmental noise into the sound source annoyance of the environmental noise according to the age of the target population;

[0012] The sound source annoyance degree and the sound level annoyance degree corresponding to the original audio signal are weighted to obtain a final environmental noise annoyance degree.

[0013] Preferably, determining the annoyance evaluation value of the sample according to the acquired environmental noise sample includes:

[0014] Ambient noise samples are obtained by sampling using a calibrated binaural measurement system;

[0015] The 10 seconds with the most obvious sound source characteristics in the environmental noise sample are intercepted as a research segment, and a subjective evaluation experiment on the annoyance of environmental noise is carried out to obtain the annoyance evaluation value of the sample.

[0016] Preferably, the environmental noise samples and the annoyance evaluation values ​​are used as training samples to train the XGBoost model to obtain a trained environmental noise annoyance evaluation value prediction model, including:

[0017] Performing audio preprocessing on the environmental noise sample to obtain a preprocessed audio signal;

[0018] Extracting a frequency spectrogram reflecting perceptual characteristics from the preprocessed audio signal;

[0019] The frequency spectrum that can reflect the perception characteristics and the annoyance evaluation value are input into the XGBoost model for training to obtain a trained environmental noise annoyance evaluation value prediction model.

[0020] Preferably, extracting a frequency spectrogram reflecting perceptual characteristics from the preprocessed audio signal comprises:

[0021] Extracting the power spectrum of the preprocessed audio signal at different time segments;

[0022] The power spectra of different time segments are processed to obtain a Fourier transform matrix, and the Fourier transform matrix is ​​determined as the frequency spectrum that can reflect the perceptual characteristics.

[0023] Preferably, converting the annoyance evaluation value of the environmental noise into the sound source annoyance of the environmental noise according to the age of the target population includes:

[0024] If the target population is in the adult range, the sound source annoyance is determined using the formula y=-1.190x+4.183; wherein x is the annoyance rating of the ambient noise, and y is the value of the sound source annoyance;

[0025] If the target population is in the elderly range, the sound source annoyance degree is determined using the formula y=-1.000x+4.253;

[0026] If the age of the target population is in the teenage range, the sound source annoyance degree is determined using the formula y=-1.496x+9.622.

[0027] Preferably, the obtained sound source annoyance and the sound level annoyance are weighted to obtain the environmental noise annoyance, wherein the sound level annoyance is measured by an acoustic measurement device.

[0028] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0029] The present invention provides an environmental noise evaluation method based on annoyance perception index, including: determining the annoyance evaluation value of the sample according to the acquired environmental noise sample; training the XGBoost model with the environmental noise sample and the annoyance evaluation value as training samples to obtain a trained environmental noise annoyance evaluation value prediction model; obtaining the original audio signal to be tested, and inputting the original audio signal into the environmental noise annoyance evaluation value prediction model to obtain the annoyance evaluation value of the environmental noise; converting the annoyance evaluation value of the environmental noise into the sound source annoyance of the environmental noise according to the age of the target population; weighting according to the sound source annoyance and the sound level annoyance corresponding to the original audio signal to obtain the final environmental noise level. The present invention adopts the "annoyance" generated by the sound source in the environmental noise and the "annoyance" generated by the sound level to weight as the environmental noise annoyance, which can more accurately describe people's annoyance feelings to environmental noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1A flow chart of a method provided by an embodiment of the present invention;

[0032] Figure 2 A schematic diagram of audio extraction by short-time Fourier transform provided in an embodiment of the present invention;

[0033] Figure 3 A schematic diagram of the linear relationship between the sound source annoyance degree and the annoyance evaluation value provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] The purpose of the present invention is to provide an environmental noise evaluation method based on annoyance perception index, which can more accurately describe the noise level of environmental noise.

[0036] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides an environmental noise evaluation method based on annoyance perception index, comprising:

[0038] Step 100: determining an annoyance evaluation value of the sample according to the acquired environmental noise sample;

[0039] Step 200: Using the environmental noise sample and the annoyance evaluation value as training samples to train the XGBoost model, and obtaining a trained environmental noise annoyance evaluation value prediction model;

[0040] Step 300: obtaining an original audio signal to be tested, and inputting the original audio signal into the environmental noise annoyance evaluation value prediction model to obtain an annoyance evaluation value of the environmental noise;

[0041] Step 400: converting the annoyance evaluation value of the environmental noise into the sound source annoyance of the environmental noise according to the age of the target population;

[0042] Step 500: weighting is performed according to the sound source annoyance degree and the ambient noise level corresponding to the original audio signal to obtain a final ambient noise level.

[0043] Specifically, the process 1 of this embodiment is to first establish a prediction model of the noise perception of source (NPS) of environmental noise. Specifically, the XGBoost "extreme gradient boosting" model is used to obtain the noise source noise perception of source (NPS) based on the subjective evaluation experiment. The following is the complete solution:

[0044] 1. Use a calibrated binaural measurement system (such as an artificial head) to sample environmental noise samples, intercept the 10 seconds with the most obvious sound source characteristics in the sound sample as the research segment, and conduct a subjective evaluation experiment on the noise level of the environmental noise source (Note: This experimental method refers to the method in "Evaluation and Prediction Method of Acoustic Noise Annoyance Degree GBT42473-2023") to obtain the annoyance evaluation value of the subjects. These environmental noise samples and the evaluation values ​​of the subjects will be used as training samples to build the XGBoost model.

[0045] 2. Take the original audio signal as the input variable;

[0046] 3. The trained XGBoost model will predict the annoyance rating of the ambient noise corresponding to the audio signal (i.e. the annoyance perception value of the sound source semantics) based on the input original audio signal.

[0047] In detail, the XGBoost model consists of three parts: audio preprocessing, extracting spectrograms that reflect perceptual features, and machine learning regression:

[0048] (1) Audio preprocessing. Audio preprocessing is divided into three processes: sampling, encoding, and normalization. The original audio signal is a time-continuous voltage signal. The original audio signal needs to be converted into a discrete digital signal through sampling before it can be processed by the computer. This process is calculated using the following formula:

[0049] s d (n) = s c (n / f)

[0050] Where f is the sampling frequency, which is set to 44100 Hz here; n is the serial number of the sampled digital signal; s(n) is the sampled digital signal, which is further encoded into a 16-bit binary number by the computer, with a sampling accuracy of 1 / 2 15 Considering the human auditory perception reaction time, the project sets the sampling accuracy to 10s, so the model preprocessing audio duration is also 10s. After sampling, there are 441,000 frames of signal, and the encoded data is 882kB in size. To be consistent with the subjective evaluation experimental conditions and exclude the influence of sound level, the equivalent sound level of the preprocessed audio of 10s is normalized to 60dB.

[0051] (2) Extracting the spectrum that can reflect the perceptual characteristics as the input of machine learning. The frequency envelope information in the spectrum is an important source of information for people to perceive different sound sources. It basically reflects the amplitude, frequency domain and time domain characteristics of the sound. Therefore, this technology uses the spectrum as a prerequisite for judging the noise level of the environmental noise source. The model will extract the spectrum features from the preprocessed audio signal as the input of the subsequent machine learning regression. The extraction process includes: short-time Fourier transform (STFT), calculation of power spectrum, and calculation of inverse frequency coefficients.

[0052] Figure 2 The power spectrum of audio extracted by short-time Fourier transform at different time segments is shown. The figure uses the time dimension as the basis and cuts out the audio digital signal of different time segments in the form of a sliding window.

[0053] Among them, the window width is 2048 frames, about 46.4ms; the sliding step is 512 frames, about 11.6ms, and the cutting process is shown in the following formula.

[0054] s(n,m)=s d (n+m(N s -1))

[0055] Where m is the serial number of the time window, n is the serial number of the frame in the time window, and N s is the sliding step size, the value is 512. The audio signal on each time segment needs to be processed by the window function to compensate for the spectrum leakage caused by not meeting the signal stationary assumption. w (n,m) performs discrete Fourier transform, as shown below

[0056]

[0057] Where k = 0, 1, 2, ..., N-1 represents the frequency, and N is the number of signal frames in a time segment. The modulus of S(k, m) is the energy of the signal at different frequencies in a time window, that is, the power spectrum. The argument of S(k, m) is the phase of the signal at different frequencies in this time window.

[0058] In order to simulate the difference in sensitivity of the human ear to sounds of different frequencies, the project uses a Mel filter bank to process the audio power spectrum of different time periods to obtain the Fourier transform matrix S at the Mel scale frequency. mel (k,m). The cepstral coefficients are a summary of the frequency envelope of the Mel power spectrum curve over a period of time, as shown in the following formula, and are generated by the discrete cosine transform of the Fourier transform result matrix.

[0059]

[0060] Where n = 0, 1, 2, ... 12, Cn is the cosine coefficient, In other cases, it is 1. After obtaining the Mel-cepstrum coefficient matrix, the matrix is ​​downsampled in the time series direction to obtain the parameters of six dimensions: mean, first-order difference mean, first-order difference variance, second-order difference mean, skewness, and kurtosis, which is a 13×6 Mel-cepstrum coefficient matrix.

[0061] (3) Machine learning regression acquisition. Machine learning regression uses the extracted Mel-cepstrum coefficient matrix as input variables, and uses the XGBoost model to integrate multiple decision trees that have undergone regression learning to establish a logical connection between the Mel-cepstrum coefficient matrix features and the annoyance rating value (NPS) of the environmental noise source. For a single decision tree, its regression is a supervised learning algorithm used to establish a prediction model. The basic principle is to gradually split the data set through a series of binary judgments, divide the data into multiple subsets, and perform regression prediction on each subset. Decision tree regression recursively divides the data set into smaller subsets, fits a linear regression model on each subset, and finally combines the linear regression models of all subsets into an overall model.

[0062] A total of 363 audio samples with NPS labels were used to build the model. The accuracy calculation formula is as follows:

[0063]

[0064] Among them, N correct It indicates the sum of the number of samples whose absolute error is less than σ=1 (σ is the variance of NPS of the subjective evaluation test subjects), that is, the number of samples with accurate prediction. total It represents the total number of samples. After training and testing the samples, the mean absolute error (MAE) of the model is 0.57, the root mean square error (RMSE) is 0.73, and the accuracy rate reaches 84% ​​at ±σ=1.

[0065] Specifically, XGBoost, which stands for eXtreme Gradient Boosting, is an ensemble learning algorithm based on the Gradient Boosting Decision Tree (GBDT), which improves prediction performance by using multiple decision trees that have undergone regression learning. Here is how XGBoost works:

[0066] 1. Initialize the model: First, build an initial decision tree model and fit the model using training data.

[0067] 2. Calculate the residual: For each sample in the training data set, calculate the error between its true value and the model's predicted value to obtain the residual of the sample. The residual refers to the model's prediction deviation for the sample in the current round.

[0068] 3. Build a new decision tree: In each round of iteration, use the residual to build a new decision tree model to fit the residual value. The new decision tree model is built on the basis of the previous round of model, and the prediction ability of the model is further improved by learning the residual.

[0069] 4. Update model prediction: Integrate the newly constructed decision tree model with the previous model to get the updated model. The final prediction value is obtained by adding the prediction results of each model.

[0070] 5. Adjust the learning rate: To improve the accuracy of the model, XGBoost also introduces the concept of learning rate. The learning rate controls the contribution of each new model in the ensemble. By adaptively adjusting the learning rate, the performance of the model can be further improved.

[0071] 6. Iterative training: Repeat the above steps for multiple rounds of iterative training. Each round of iteration will build a new decision tree model to fit the residual of the previous round, and then integrate the new model with the previous model. Through the iterative process, the model is gradually improved and the prediction performance is improved.

[0072] Furthermore, process 2 of this embodiment is to convert the noise perception value of the environmental noise source obtained in process 1 from a psychological description score into an acoustic physical quantity in the same unit as the sound pressure level index commonly used in the current environmental noise evaluation, namely, the noise source annoyance level of environmental noise (NS). This conversion process aims to combine the subjective evaluation results with the objective physical measurement standard (sound pressure level expressed in decibels) to provide a more comprehensive, objective and quantifiable environmental noise evaluation index. Because different people have different annoyance perceptions, the specific structural model is as follows: Figure 3 shown.

[0073] Considering that the main users of a large number of places in the city are adults, the formula is as follows:

[0074] y=-1.190x+4.183

[0075] Among them, x is the environmental noise source annoyance rating value (NPS), and y is the environmental noise source annoyance (NS). When the users of the venue are the elderly, the following formula can be used:

[0076] y=-1.000x+4.253

[0077] When the venue is for teenagers, the following formula can be used:

[0078] y=-1.496x+9.622

[0079] Optionally, process 3 of this embodiment is to weight the "annoyance" generated by the sound source semantics and the "annoyance" generated by the sound level as the environmental noise annoyance (Noise level, referred to as "NL"), providing an accurate method for environmental noise evaluation in an ecological environment. SL is "Source Level" and is the sound level annoyance. The formula is (in dB):

[0080] NL=SL+NS

[0081] In general, the environmental noise level measurement is obtained by measuring the annoyance perception decibel value (i.e., sound source annoyance) and sound level annoyance generated by the sound source semantics in the environment. Among them, the sound source annoyance is obtained according to the structural model provided in step 2, and the sound level annoyance is the equivalent sound level in a certain period of time, which can be provided by the acoustic measurement equipment that has passed the quality inspection on the market.

[0082] The beneficial effects of the present invention are as follows:

[0083] Compared with the existing evaluation index that only uses the "annoyance level" of sound level to measure environmental noise, the present invention uses the "annoyance level" caused by the sound source in the environmental noise and the "annoyance level" caused by the sound level to weight as the environmental noise annoyance level, which can more accurately describe the annoyance effect of environmental noise on people.

[0084] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0085] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for evaluating environmental noise based on annoyance perception index, characterized in that: include: Determine an annoyance evaluation value of the sample according to the acquired environmental noise sample; The environmental noise samples and the annoyance evaluation values ​​are used as training samples to train the XGBoost model to obtain a trained environmental noise annoyance evaluation value prediction model; Acquire an original audio signal of the environmental noise to be measured, and input the original audio signal into the environmental noise annoyance evaluation value prediction model to obtain an annoyance evaluation value of the environmental noise; Converting the annoyance evaluation value of the environmental noise into the sound source annoyance of the environmental noise according to the age of the target population; The sound source annoyance degree and the sound level annoyance degree corresponding to the original audio signal are weighted to obtain a final environmental noise annoyance degree.

2. The environmental noise evaluation method based on annoyance perception index according to claim 1, characterized in that: Determine the annoyance evaluation value of the sample according to the acquired environmental noise sample, including: Ambient noise samples are obtained by sampling using a calibrated binaural measurement system; The 10 seconds with the most obvious sound source characteristics in the environmental noise sample are intercepted as a research segment, and an environmental noise subjective evaluation experiment is carried out to obtain the annoyance evaluation value of the sample.

3. The environmental noise evaluation method based on annoyance perception index according to claim 1, characterized in that: The environmental noise sample and the annoyance evaluation value are used as training samples to train the XGBoost model to obtain a trained environmental noise annoyance evaluation value prediction model, including: Performing audio preprocessing on the environmental noise sample to obtain a preprocessed audio signal; Extracting a frequency spectrogram reflecting perceptual characteristics from the preprocessed audio signal; The spectrum graph environmental noise sample signal spectrum graph that can reflect the perception characteristics and the annoyance evaluation value are input into the XGBoost model for training to obtain a trained environmental noise annoyance evaluation value prediction model.

4. The environmental noise evaluation method based on annoyance perception index according to claim 3 is characterized in that: Extracting a frequency spectrum reflecting perceptual characteristics from the preprocessed audio signal includes: Extracting the power spectrum of the preprocessed audio signal at different time segments using a short-time Fourier transform method; The power spectra of different time segments are processed to obtain a Fourier transform matrix, and the Fourier transform matrix is ​​used to obtain a frequency spectrum that can reflect the perception characteristics.

5. The environmental noise evaluation method based on annoyance perception index according to claim 1, characterized in that: The annoyance evaluation value of the environmental noise is converted into the annoyance of the sound source of the environmental noise according to the age of the target population, including: If the target population is in the adult range, the sound source annoyance is determined using the formula y=-1.190x+4.183; wherein x is the annoyance rating of the ambient noise, and y is the value of the sound source annoyance; If the target population is in the elderly range, the sound source annoyance degree is determined using the formula y=-1.000x+4.253; If the age of the target population is in the teenage range, the sound source annoyance degree is determined using the formula y=-1.496x+9.

622.

6. The environmental noise evaluation method based on annoyance perception index according to claim 1, characterized in that: The obtained sound source annoyance degree and the sound level annoyance degree are weighted to obtain the environmental noise annoyance degree, wherein the sound level annoyance degree is measured by an acoustic measurement device.

Images