Sound field partition control method, device, vehicle-mounted speaker and storage medium

Abstract

This invention relates to the field of audio noise reduction technology, and provides a sound field zoning control method, device, vehicle speaker, and storage medium. The method includes: acquiring transfer function signals from the speaker to each sound field zone; obtaining filter parameters for the speaker when each sound field zone is considered a bright / dark zone based on the transfer function signals and a neural network model; iterating the parameters of the neural network model based on the differences in sound field energy between the bright and dark zones in each sound field zone, wherein the sound field energy in the bright and dark zones is determined based on the transfer function signals and the filter parameters; and using the neural network model after parameter iteration to determine the sound field zoning filter parameters for the speaker to be controlled. The sound field zoning control method, device, vehicle speaker, and storage medium provided by this invention achieve a targeted improvement in isolation experience in multi-bright-zone environments from the user's listening perspective.

Description

Technical Field

[0001] The present invention relates to the technical field of audio noise reduction, and in particular to a sound field partition control method, device, vehicle-mounted speaker and storage medium. Background Art

[0002] With the rapid development of new energy vehicles in recent years, smart cockpits have become the main development direction of new energy vehicles, and intelligent in-car entertainment has gradually become a standard feature of automakers. In-car sound field control technology has begun to enter the public's field of vision as an important means to improve the in-car sound quality environment and enhance the comfort of drivers and passengers.

[0003] The in-vehicle sound field partitioning technology is designed to provide different audio and video services to drivers and passengers in different areas without interfering with each other. For example, the front-seat driver listens to the navigation while driving, while the passengers in the back want to enjoy the music. In this process, the front-seat driver will not be disturbed by the music in the back and can concentrate on driving, while the passengers in the back will not be disturbed by the navigation sound.

[0004] Existing sound field partitioning technologies use speaker arrays to build the target sound field in the bright area, while playing the reverse sound source in the dark area through speakers to offset the sound leaking from the bright area. However, in scenarios where multiple bright areas are working at the same time, it is often difficult to achieve the ideal effect, affecting the user's listening experience. Summary of the invention

[0005] The present invention provides a sound field partition control method, device, vehicle-mounted loudspeaker and storage medium, which are used to solve the defect in the prior art that the sound field partition effect is not ideal in the scenario where multiple bright areas work simultaneously.

[0006] The present invention provides a sound field partition control method, comprising: Obtain transfer function signals from the loudspeaker to each sound field partition; Based on the transfer function signal and the neural network model, filter parameters corresponding to the loudspeaker when each of the sound field partitions is used as a bright area or a dark area are obtained; Based on the difference between the sound field energies of the light and dark zones under each sound field partition, the neural network model is iterated in parameters. The sound field energies of the light and dark zones are determined based on the transfer function signal and the filter parameters. The neural network model after the parameter iteration is used to determine the sound field partition filter parameters of the speaker to be controlled.

[0007] According to the sound field partition control method provided by the present invention, the parameter iteration of the neural network model based on the difference between the sound field energies of the bright and dark zones in each sound field partition comprises: Determine the dark zone sound field energy of any sound field partition based on the transfer function signal of any sound field partition and the filter parameters corresponding to any sound field partition when the any sound field partition is used as a dark zone; Determine the bright zone sound field energy of any sound field partition based on the transfer function signal of any sound field partition and the filter parameters corresponding to any sound field partition when the any sound field partition is used as a bright zone; Based on the contrast between the sound field energy in the dark area and the sound field energy in the bright area in each sound field partition, parameter iteration is performed on the neural network model.

[0008] According to the sound field partition control method provided by the present invention, the neural network model is iterated based on the contrast between the sound field energy in the dark area and the sound field energy in the bright area in each sound field partition, including: Determine the dark area reconstruction loss based on the dark area sound field energy of each sound field partition and the contrast between the dark area sound field energy and the bright area sound field energy; Based on the dark area reconstruction loss, parameters of the neural network model are iterated.

[0009] According to the sound field partition control method provided by the present invention, the parameter iteration of the neural network model based on the dark area reconstruction loss includes: Determine the bright area reconstruction loss based on the bright area sound field energy of each sound field partition and the difference between the target sound field energies of each sound field partition; Based on the bright area reconstruction loss and the dark area reconstruction loss, parameters of the neural network model are iterated.

[0010] According to the sound field partition control method provided by the present invention, the step of obtaining a transfer function signal from a loudspeaker to each sound field partition includes: Acquire the original transfer function signal from the loudspeaker to each sound field partition, and perform frequency domain transformation on the original transfer function signal to obtain a frequency domain signal; Divide the frequency domain signal into sub-bands from low frequency to high frequency according to the intervals from small to large, to obtain signals of each frequency band; The transfer function signal is determined based on the signals in each frequency band.

[0011] According to the sound field partition control method provided by the present invention, the determining of the transfer function signal based on the signals of each frequency band includes: Based on the frequency band characteristics of the signals in each frequency band, randomly perturbing the signals in each frequency band to obtain a perturbation signal; The disturbance signal and the signals in each frequency band are used as the transfer function signal.

[0012] According to the sound field partition control method provided by the present invention, the parameter iteration of the neural network model based on the difference between the sound field energies of the bright and dark zones in each sound field partition comprises: Determine the frequency band loss of the signal in each frequency band based on the acoustic characteristics of the signal in each frequency band; Based on the frequency band loss and the difference between the sound field energies of the bright and dark areas under each sound field partition, the neural network model is iterated on parameters.

[0013] According to the sound field partition control method provided by the present invention, the filter parameters corresponding to the speaker when each sound field partition is used as a bright or dark zone are obtained based on the transfer function signal and the neural network model, including: The transfer function signal is sequentially input into the encoder, long short-term memory network and decoder of the neural network model to obtain the filter parameters corresponding to the speaker output by the decoder when each sound field partition is used as a light or dark area.

[0014] The present invention also provides a sound field partition control device, comprising: A signal acquisition unit, used to acquire a transfer function signal from the loudspeaker to each sound field partition; A parameter determination unit, configured to obtain, based on the transfer function signal and the neural network model, filter parameters corresponding to the loudspeaker when each of the sound field partitions is used as a bright area or a dark area respectively; A parameter iteration unit is used to iterate the parameters of the neural network model based on the difference between the sound field energies of the light and dark zones under each sound field partition, wherein the sound field energies of the light and dark zones are determined based on the transfer function signal and the filter parameters, and the neural network model after the parameter iteration is completed is used to determine the sound field partition filter parameters of the speaker to be controlled.

[0015] The present invention also provides a vehicle-mounted speaker, comprising a speaker body, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the sound field partition control method when executing the computer program.

[0016] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the above-mentioned sound field partition control methods is implemented.

[0017] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the sound field partition control method as described above is implemented.

[0018] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned sound field partition control methods.

[0019] The sound field partition control method, device, vehicle-mounted speaker and storage medium provided by the present invention obtain the filter parameters corresponding to the speaker when each sound field partition is respectively used as a light and dark zone based on the transfer function signal and the neural network model, and calculate the filter parameters using the nonlinear mapping ability of the neural network model. Compared with the idea of ​​matrix inversion, the overfitting phenomenon is avoided and the robustness is higher; compared with the method of independently calculating the filter parameters of different light and dark zones, the present invention can simultaneously obtain the filter parameters corresponding to each sound field partition when it is respectively used as a light and dark zone, so that the filter parameters corresponding to the light and dark zones are considered as a whole, and then the light and dark zone sound field energy of the same sound field partition is obtained. On this basis, based on the difference between the light and dark zone sound field energies under each sound field partition, the neural network model is iterated in parameters, so as to achieve the targeted improvement of the isolation experience in a multi-light zone environment from the perspective of the user's hearing sense. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 This is one of the flow charts of the sound field partition control method provided by the present invention.

[0022] Figure 2 It is a flowchart diagram of step 130 in the sound field partition control method provided by the present invention.

[0023] Figure 3 It is a flow chart of step 110 in the sound field partition control method provided by the present invention.

[0024] Figure 4 It is a schematic diagram of the training structure of the neural network model provided by the present invention.

[0025] Figure 5 It is a structural schematic diagram of the neural network model provided by the present invention.

[0026] Figure 6 It is a structural schematic diagram of the sound field partition control device provided by the present invention.

[0027] Figure 7 It is a structural schematic diagram of the vehicle-mounted speaker provided by the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0029] The in-car sound field is mainly divided into two parts: one part is the noise caused by various excitation sources in the car. The in-car noise of traditional fuel vehicles includes mechanical noise, internal combustion engine noise, road noise and wind noise. Electric vehicles use motors as power devices. Although they lack internal combustion engine noise, they have high-frequency electromagnetic noise; the other part is the multimedia sound field generated by the in-car audio system, including navigation sound, music sound and various safety prompt sounds. Traditional sound field control generally focuses on active noise reduction control. Compared with the research on car noise, the in-car audio and video entertainment system has received less attention. Therefore, the research on sound field control technology for in-car audio and video entertainment can not only meet consumers' diversified needs for the car acoustic environment, but also further improve the riding comfort, which has a great role in realizing the intelligence of cars and improving brand competitiveness.

[0030] In summary, the in-vehicle sound field zoning technology is to reconstruct the target sound field in a specific area (bright area) while minimizing the impact of the reconstructed sound field in the bright area on other areas (dark areas). With the increasing number of speakers in the current car, it is possible to use speaker arrays for sound field zoning control. On the one hand, the target sound field is reconstructed to ensure the sound quality requirements of the bright area, and on the other hand, leakage to the dark area is reduced to meet the passengers' independent audio enjoyment space.

[0031] The existing sound field partitioning technology uses a speaker array to construct the target sound field in the bright area, while playing the reverse sound source through the speakers in the dark area to offset the sound leaking from the bright area. Each speaker is pre-placed with a filter obtained by the algorithm. The calculation form of the filter parameters can be divided into offline parameters calculated based on the pre-measured transfer function and real-time parameters obtained based on the adaptive filter algorithm and the reference microphone.

[0032] The online adaptive filter algorithm uses partitioning as a method of active noise reduction. It uses reference microphones in the car and generates reverse signals in real time based on the reference signals to cancel out the sound waves. The adaptive algorithm requires additional microphones and headrest speakers, which requires high hardware costs. In actual applications, uncertainties such as difficulty in convergence may occur, which greatly affects the user experience. Therefore, it is usually used in combination with offline solutions.

[0033] The offline parameter calculation method mainly measures the transmission path from the car speaker to the control area in advance, and then inverts the entire system matrix. However, the algorithm's inversion of the system matrix will cause serious overfitting, and the actual experience effect cannot be guaranteed.

[0034] In addition, the current mainstream offline algorithms all calculate each light and dark area parameter as a set of independent parameters. In scenarios where multiple light areas work simultaneously, it is often difficult to achieve the desired effect, affecting the user's listening experience. For example, when the front row is regarded as a light area and the back row as a dark area, the contrast pursued by the existing technology is the ratio of the sound energy in the front row to the sound energy in the back row. When the front and back rows are both regarded as light areas, the user experience of the front row user is actually the ratio of the sound energy in the front row light area to the sound energy leaked from the back row light area to the front row.

[0035] In view of the above problems, an embodiment of the present invention proposes a sound field partition control method, in which, based on the transfer function signal and the neural network model, the filter parameters corresponding to the speaker when each sound field partition is respectively used as a light and dark zone are obtained, and the filter parameters are calculated using the nonlinear mapping ability of the neural network model. Compared with the idea of ​​matrix inversion, the overfitting phenomenon is avoided and the robustness is higher; compared with the method of independently calculating the filter parameters of different light and dark zones, the present invention can simultaneously obtain the filter parameters corresponding to each sound field partition when it is respectively used as a light and dark zone, so that the filter parameters corresponding to the light and dark zones are considered as a whole, and then the light and dark zone sound field energy of the same sound field partition is obtained. On this basis, based on the difference between the light and dark zone sound field energies under each sound field partition, the neural network model is iterated in parameters, so as to achieve the targeted improvement of the isolation experience in a multi-light zone environment from the perspective of the user's hearing sense.

[0036] The embodiments of the present invention can be applied to scenes where sound field zoning control is required, such as scenes where sound field zoning control is required in open spaces such as carriages, theaters, living rooms, and museums. The execution subject of the method can be a terminal device, a computer, a server, a server cluster, or an electronic device such as a specially designed sound field zoning control, or a sound field zoning control device set in the electronic device, and the sound field zoning control device can be implemented by software, hardware, or a combination of the two.

[0037] Combine the following Figure 1-Figure 7 , the sound field partition control method of the present invention is described by taking the sound field partition in a vehicle-mounted scenario as an example. Figure 1 is one of the flow charts of the sound field partition control method provided by the present invention, such as Figure 1 As shown, the method comprises the following steps: Step 110: Acquire a transfer function signal from the loudspeaker to each sound field partition.

[0038] Specifically, each sound field partition can be pre-divided, and the sound field is divided into multiple partitions according to actual needs, each area represents a sound field partition, and the position and range of each partition are determined. Taking a car as an example, it can be divided into two sound field partitions: the front area and the rear area.

[0039] The transfer function signal reflects the frequency response characteristics of the loudspeaker to each sound field partition. Further, each sound field partition may include multiple control points, and the transfer function signals of all control points from the loudspeaker to each sound field partition are obtained.

[0040] The transfer function signal can be measured in advance inside the vehicle. Select an appropriate test signal, such as a swept frequency signal or pink noise. The test signal should cover the frequency range of interest in the frequency domain. For example, use a speaker to play the test signal to ensure that the signal can be clearly perceived in each sound field partition. Place a microphone in each sound field partition to receive the sound signal from the speaker. Compare the signal received by the microphone with the original signal during playback, and calculate the transfer function signal in the frequency domain.

[0041] Step 120, based on the transfer function signal and the neural network model, obtain the filter parameters corresponding to the speaker when each sound field partition is used as a bright area and a dark area respectively.

[0042] Specifically, the sound field partition is a bright area, which means that a specific sound source is expected to be replayed in the sound field partition, while other areas corresponding to the specific sound source are dark areas. For example, if the front driver is expected to receive the navigation sound more clearly and minimize the impact of the navigation sound on the rear passengers, the front area where the driver is located can be regarded as a bright area and the rear area as a dark area. At this time, the corresponding filter parameters for the front bright area and the rear dark area can be expressed as .

[0043] For another example, if we want the rear passengers to enjoy the music more happily while minimizing the impact of the music on the front driver, we can use the rear area where the passengers are located as the bright area and the front area as the dark area. The corresponding filter parameters when the front area is dark and the rear area is bright can be expressed as .

[0044] The corresponding filter parameters when each sound field partition is used as a bright or dark area can be realized through a transfer function signal and a neural network model. The transfer function signal is input into the neural network model, and the neural network model analyzes the transfer function signal to obtain the corresponding filter parameters. The architecture of the neural network here can adopt a recurrent neural network (RNN) and its variants (such as LSTM, GRU).

[0045] Step 130, based on the difference between the sound field energies of the light and dark zones in each sound field partition, the parameters of the neural network model are iterated, the sound field energies of the light and dark zones are determined based on the transfer function signal and the filter parameters, and the neural network model after the parameter iteration is completed is used to determine the sound field partition control parameters of the speaker.

[0046] Specifically, considering the scenario where multiple bright areas work at the same time, for example, when the front and rear partitions are used as bright areas at the same time, for any bright area, the speaker not only needs to reconstruct the sound field of the bright area, but also needs to offset the sound energy leaked from another bright area to the bright area as much as possible. Therefore, in this embodiment, the neural network model is trained based on the difference between the sound field energies of the bright and dark areas under each sound field partition. The sound field energy of the bright and dark areas here is for the same sound field partition, that is, the difference between the sound field energy of the bright area and the sound field energy of the dark area of ​​the same sound field partition needs to be large enough, so that when the two bright areas work at the same time, even if the dark area has a poor cancellation effect in some frequency bands, because the bright area energy in this frequency band is relatively large in this area, the dark area residue can be well covered by the bright area sound, without affecting the listening experience. Here, the sound field energy of the bright and dark areas is determined based on the transfer function signal and the filter parameters.

[0047] During the neural network parameter iteration process, the weights and biases of the neural network need to be continuously adjusted to maximize the difference between the energy of the bright and dark sound fields in each sound field partition. After the parameter iteration, the neural network model learns the nonlinear mapping relationship between the transfer function signal and the filter parameters. In practical applications, the sound field partition filter parameters of the speaker to be controlled can be determined based on the transfer function signal from the speaker to be controlled to each sound field partition.

[0048] The method provided by the embodiment of the present invention obtains the filter parameters corresponding to the speaker when each sound field partition is respectively used as a light and dark zone based on the transfer function signal and the neural network model, and calculates the filter parameters using the nonlinear mapping capability of the neural network model. Compared with the idea of ​​matrix inversion, the overfitting phenomenon is avoided, and the robustness is higher; compared with the method of independently calculating the filter parameters of different light and dark zones, the present invention can simultaneously obtain the filter parameters corresponding to each sound field partition when it is respectively used as a light and dark zone, so that the filter parameters corresponding to the light and dark zones are considered as a whole, and then the light and dark zone sound field energy of the same sound field partition is obtained. On this basis, based on the difference between the light and dark zone sound field energies under each sound field partition, the neural network model is iterated in parameters, so as to achieve the targeted improvement of the isolation experience in a multi-light zone environment from the perspective of the user's hearing sense.

[0049] Based on any of the above embodiments, Figure 2 is a flow chart of step 130 in the sound field partition control method provided by the present invention, such as Figure 2As shown, based on the difference between the sound field energies of the bright and dark areas in each sound field partition, the neural network model is iterated in parameters, that is, step 130 specifically includes: Step 131, determining the dark zone sound field energy of any sound field partition based on the transfer function signal of the sound field partition and the filter parameters corresponding to the sound field partition when the sound field partition is used as a dark zone; Step 132, determining the bright area sound field energy of the sound field partition based on the transfer function signal of the sound field partition and the filter parameters corresponding to the sound field partition as a bright area; Step 133, based on the contrast between the sound field energy in the dark area and the sound field energy in the bright area in each sound field partition, iterate the parameters of the neural network model.

[0050] Specifically, the difference between the sound field energies of the bright and dark areas in each sound field partition can be represented by contrast. For any sound field partition, the sound field energy of the dark area and the sound field energy of the bright area of ​​the sound field partition are first determined respectively.

[0051] The dark zone sound field energy of the sound field partition is determined based on the transfer function signal of the sound field partition and the corresponding filter parameters when the sound field partition is used as a dark zone. The bright zone sound field energy of the sound field partition is determined based on the transfer function signal of the sound field partition and the corresponding filter parameters when the sound field partition is used as a bright zone.

[0052] Taking the front and rear row partitions inside the car as an example, let the path from the speaker to all front row control points be , the transmission path to all rear control points is , that is, the transfer function signal of the front partition can be expressed as , the transfer function signal of the rear partition can be expressed as Let the filter parameters of the front bright area and the back dark area be , the filter parameters for the front dark area and the back bright area are That is, for the front partition, the corresponding filter parameter when the front partition is the bright area is , the corresponding filter parameters when the front partition is used as the dark area are ; For the rear partition, the corresponding filter parameters when the rear partition is used as a bright area are , the corresponding filter parameters when the rear partition is used as the dark area are .

[0053] For the front partition, the dark area sound field energy can be expressed as , the sound field energy in the bright area can be expressed as , the contrast between the dark area sound field energy and the bright area sound field energy in the front partition can be expressed as ; For the rear partition, the dark area sound field energy can be expressed as , the sound field energy in the bright area can be expressed as , the contrast between the energy of the dark area sound field and the energy of the bright area sound field in the rear partition can be expressed as .

[0054] Here, during training, the loss constraints of the neural network model can be determined based on the contrast between the dark area sound field energy and the bright area sound field energy in each sound field partition. By constraining the bright and dark areas of the same sound field partition to have sufficient contrast, when the two bright areas work at the same time, even if the dark area has a poor cancellation effect in some frequency bands, the dark area residue can be well covered by the bright area sound because the bright area energy in this area is relatively large, without affecting the listening experience.

[0055] Based on any of the above embodiments, based on the contrast between the sound field energy in the dark area and the sound field energy in the bright area in each sound field partition, the neural network model is iterated in parameters, that is, step 133 includes: Step 133-1, determining a dark area reconstruction loss based on the dark area sound field energy of each sound field partition and the contrast between the dark area sound field energy and the bright area sound field energy; Step 133 - 2, based on the dark area reconstruction loss, iterate the parameters of the neural network model.

[0056] Specifically, for the neural network model, parameter iteration is performed. In addition to constraining the sound field energy of the bright and dark areas of the same sound field partition to have sufficient contrast, the dark area target also needs to constrain the sound field energy of the dark area to be as small as possible. Therefore, the dark area reconstruction loss can be determined based on the dark area sound field energy of each sound field partition, and the contrast between the dark area sound field energy and the bright area sound field energy.

[0057] Referring to the above embodiment, the dark area sound field energy of the front partition can be expressed as , the dark area sound field energy of the rear partition can be expressed as , then the dark area reconstruction loss can be expressed as: in is a balance parameter.

[0058] Based on any of the above embodiments, the neural network model is iterated based on the dark area reconstruction loss, that is, step 133-2 specifically includes: Step 133-21, determining the bright area reconstruction loss based on the bright area sound field energy of each sound field partition and the difference between the target sound field energies of each sound field partition; Step 133-22, based on the bright area reconstruction loss and the dark area reconstruction loss, the neural network model is iterated on parameters.

[0059] Specifically, the neural network model is iterated in parameters. In addition to constraining the dark area reconstruction loss, the bright area reconstruction loss can be further constrained. The bright area reconstruction loss is determined based on the bright area sound field energy of each sound field partition and the difference between the target sound field energy of each sound field partition. The target sound field energy of each sound field partition can be pre-set.

[0060] Referring to the above embodiment, the loss function of the neural network model can be expressed as follows: The loss function consists of two parts, and is the proportionality coefficient, where Rebuilding losses for the Dark Zone. To reconstruct the loss in the bright area, the main purpose is to constrain and optimize the sound quality of the bright area sound field. The specific formula is as follows: in and are the target sound field energies for the front and rear partitions, is the dark area sound field energy of the front row partition, It is the dark area sound field energy of the rear partition.

[0061] Based on any of the above embodiments, Figure 3 is a flow chart of step 110 in the sound field partition control method provided by the present invention, such as Figure 3 As shown, the transfer function signal from the loudspeaker to each sound field partition is obtained, that is, step 110 specifically includes: Step 111, obtaining an original transfer function signal from the loudspeaker to each sound field partition, and performing frequency domain transformation on the original transfer function signal to obtain a frequency domain signal; Step 112, dividing the frequency domain signal into sub-bands from low frequency to high frequency according to the interval from small to large, to obtain signals of each frequency band; Step 113: determine a transfer function signal based on the signals in each frequency band.

[0062] Specifically, considering that the control points of each sound field partition are limited, resulting in a limited number of input transfer function signals and causing insufficient training data, an embodiment of the present invention preprocesses the original transfer function signal to further improve the training effect of the neural network model.

[0063] First, the original transfer function signal is transformed into the frequency domain to obtain the frequency domain signal. .

[0064] Then the frequency domain signal is divided into sub-bands from low frequency to high frequency according to the intervals from small to large, that is, the intervals of sub-band division gradually increase as the frequency gradually increases. For example, sub-band division is performed every 10Hz at low frequency, every 100Hz at medium frequency, and every 500Hz at high frequency. The signals of each frequency band thus obtained have higher resolution at low frequency with the same number of points.

[0065] Then, each frequency band signal may be directly determined as a transfer function signal, or each frequency band signal may be further processed and the processed frequency band signal may be used as a transfer function signal.

[0066] Based on any of the above embodiments, the transfer function signal is determined based on the signals of each frequency band, that is, step 113 specifically includes: Step 113-1, based on the frequency band characteristics of the signals in each frequency band, randomly perturb the signals in each frequency band to obtain a perturbation signal; Step 113 - 2 , using the disturbance signal and the signals of each frequency band as transfer function signals.

[0067] Specifically, the data can be continuously expanded by random perturbation to solve the problem of insufficient training data. According to the increasingly poor stability of the low-frequency to high-frequency transfer function, random perturbations are added in a manner that gradually increases the perturbation intensity from low frequency to high frequency according to the frequency band characteristics of the signals in each frequency band, that is, random perturbations of different amplitudes are applied to different frequency bands to obtain perturbation signals.

[0068] On this basis, the disturbance signal and the signals of each frequency band are used together as transfer function signals to solve the problem of insufficient training data and improve the robustness of the model.

[0069] Based on any of the above embodiments, based on the difference between the sound field energies of the bright and dark areas in each sound field partition, the neural network model is iterated in parameters, including: Determine the frequency band loss of the signal in each frequency band based on the acoustic characteristics of the signal in each frequency band; Based on the frequency band loss and the difference between the sound field energy of the bright and dark areas in each sound field partition, the parameters of the neural network model are iterated.

[0070] Specifically, although the current filter parameters are calculated independently for each frequency band, the acoustic characteristics of different frequency bands are not considered. The actual experience of the same contrast in different frequency bands is different, so it is necessary to formulate corresponding loss function weight relationships for different frequency bands. First, based on the acoustic characteristics of the signals in each frequency band, the frequency band loss of the signals in each frequency band is determined, such as freely adding different constraints such as speaker power, minimum isolation, and minimum sound quality loss. At the same time, different constraints can be imposed on different frequency bands to obtain frequency band losses.

[0071] On this basis, when the neural network model is iterated on parameters, the loss function is determined based on the frequency band loss and the difference between the sound field energy of the bright and dark areas under each sound field partition. Compared with the matrix inversion method, the embodiment of the present invention utilizes the nonlinear mapping ability of the neural network model and can impose different frequency band constraints on different frequency bands, which has higher flexibility.

[0072] Based on any of the above embodiments, Figure 4 is a schematic diagram of the training structure of the neural network model provided by the present invention, referring to Figure 4 , based on the transfer function signal and the neural network model, the corresponding filter parameters of the loudspeaker when each sound field partition is used as a bright and dark area are obtained, including: The transfer function signal is sequentially input into the encoder, long short-term memory network and decoder of the neural network model to obtain the filter parameters corresponding to the loudspeaker output by the decoder when each sound field partition is used as a bright or dark area.

[0073] Specifically, the structure of the neural network model includes an encoder, a long short-term memory network and a decoder. The transfer function signal is first input to the encoder for encoding, then passes through a two-layer long short-term memory network (Long Short Time Memory, LSTM), and finally output by the decoder. Finally, the sub-band is re-integrated and mapped into a mask, that is, the desired frequency domain filter parameters .in Figure 5 Schematic diagram of the structure of the neural network model provided by the present invention. The encoder and decoder are respectively composed of a number of convolution units and deconvolution units. The internal structures of the convolution unit and the deconvolution unit are as follows: Figure 5 As shown in the figure, since the input signal is complex, complex convolution is used, and then it passes through the batch normalization layer and finally reaches the activation function output. The encoder and decoder units are connected by skip connections to prevent the gradient from disappearing when the number of layers is too deep.

[0074] The neural network model structure provided by the embodiment of the present invention can simultaneously output the filter parameters corresponding to the speaker when each sound field partition is used as a light and dark zone. Compared with calculating each light and dark zone parameter as a set of independent parameters, the present invention can combine the parameters of different light and dark zones to improve the experience effect of multiple light zones working simultaneously.

[0075] Based on any of the above embodiments, a sound field partition control method is provided, including: S1, obtain the original transfer function signal from the speaker to each sound field partition, and perform frequency domain transformation on the original transfer function signal to obtain a frequency domain signal; divide the frequency domain signal into sub-bands from low frequency to high frequency according to the interval from small to large to obtain each frequency band signal; based on the frequency band characteristics of each frequency band signal, perform random perturbation on each frequency band signal to obtain a perturbation signal; use the perturbation signal and each frequency band signal as the transfer function signal.

[0076] S2, the transfer function signal is sequentially input into the encoder, long short-term memory network and decoder of the neural network model to obtain the filter parameters corresponding to the loudspeaker output by the decoder when each sound field partition is used as a light and dark area.

[0077] S3, based on the transfer function signal of any sound field partition and the corresponding filter parameters when the sound field partition is a dark zone, determine the dark zone sound field energy of the sound field partition; based on the transfer function signal of the sound field partition and the corresponding filter parameters when the sound field partition is a bright zone, determine the bright zone sound field energy of the sound field partition; based on the dark zone sound field energy of each sound field partition and the contrast between the dark zone sound field energy and the bright zone sound field energy, determine the dark zone reconstruction loss; based on the bright zone sound field energy of each sound field partition and the difference between the target sound field energies of each sound field partition, determine the bright zone reconstruction loss; based on the bright zone reconstruction loss and the dark zone reconstruction loss, perform parameter iteration on the neural network model. And / or, Based on the acoustic characteristics of the signals in each frequency band, the frequency band loss of the signals in each frequency band is determined; based on the frequency band loss and the difference between the sound field energies of the bright and dark areas in each sound field partition, the parameters of the neural network model are iterated.

[0078] The sound field energy of the bright and dark areas is determined based on the transfer function signal and the filter parameters, and the neural network model after the parameter iteration is completed is used to determine the sound field partition filter parameters of the speaker to be controlled.

[0079] The method provided in the embodiment of the present invention uses the nonlinear mapping ability of the neural network to calculate the filter parameters instead of the idea of ​​matrix inversion. Therefore, although the effect on the training set is not as good as ACC-PM, it has high robustness and often has less loss on the test set. This embodiment divides the frequency domain signal into sub-bands from low frequency to high frequency according to the intervals from small to large, and can impose different constraints on different frequency bands, which conforms to the psychoacoustic characteristics and has a better listening experience. In addition, the ACC-PM method calculates the filter parameters of different light and dark areas independently during the calculation process. The embodiment of the present invention considers it as a whole and puts forward the concept of hearing ACC, that is, the light and dark areas of the same sound field partition need to have sufficient contrast, which has a targeted improvement in the isolation experience in a multi-light area environment.

[0080] The sound field partition control device provided by the present invention is described below. The sound field partition control device described below and the sound field partition control method described above can be referred to each other.

[0081] Based on any of the above embodiments, Figure 6 Schematic diagram of the structure of the sound field partition control device provided by the present invention. Figure 6 As shown, a sound field partition control device is provided, comprising: A signal acquisition unit 610 is used to acquire a transfer function signal from the loudspeaker to each sound field partition; A parameter determination unit 620, configured to obtain, based on the transfer function signal and the neural network model, filter parameters corresponding to the speaker when each of the sound field partitions is used as a bright area or a dark area respectively; The parameter iteration unit 630 is used to perform parameter iteration on the neural network model based on the difference between the sound field energies of the light and dark zones under each sound field partition. The sound field energies of the light and dark zones are determined based on the transfer function signal and the filter parameters. The neural network model after the parameter iteration is completed is used to determine the sound field partition filter parameters of the speaker to be controlled.

[0082] The device provided by the embodiment of the present invention obtains the filter parameters corresponding to the speaker when each sound field partition is respectively used as a light and dark zone based on the transfer function signal and the neural network model, and calculates the filter parameters using the nonlinear mapping capability of the neural network model. Compared with the idea of ​​matrix inversion, it avoids overfitting and has higher robustness. Compared with the method of independently calculating the filter parameters of different light and dark zones, the present invention can simultaneously obtain the filter parameters corresponding to each sound field partition when it is respectively used as a light and dark zone, so as to regard the filter parameters corresponding to the light and dark zones as a whole and consider them as a whole, and then obtain the light and dark zone sound field energy of the same sound field partition. On this basis, based on the difference between the light and dark zone sound field energies under each sound field partition, the neural network model is iterated in parameters, so as to achieve the targeted improvement of the isolation experience in a multi-light zone environment from the perspective of the user's hearing sense.

[0083] Based on the above embodiment, the parameter iteration unit is specifically used for: Determine the dark zone sound field energy of any sound field partition based on the transfer function signal of any sound field partition and the filter parameters corresponding to any sound field partition when the any sound field partition is used as a dark zone; Determine the bright zone sound field energy of any sound field partition based on the transfer function signal of any sound field partition and the filter parameters corresponding to any sound field partition when the any sound field partition is used as a bright zone; Based on the contrast between the sound field energy in the dark area and the sound field energy in the bright area in each sound field partition, parameter iteration is performed on the neural network model.

[0084] Based on the above embodiment, the parameter iteration unit is specifically used for: Determine the dark area reconstruction loss based on the dark area sound field energy of each sound field partition and the contrast between the dark area sound field energy and the bright area sound field energy; Based on the dark area reconstruction loss, parameters of the neural network model are iterated.

[0085] Based on the above embodiment, the parameter iteration unit is specifically used for: Determine the bright area reconstruction loss based on the bright area sound field energy of each sound field partition and the difference between the target sound field energies of each sound field partition; Based on the bright area reconstruction loss and the dark area reconstruction loss, parameters of the neural network model are iterated.

[0086] Based on the above embodiment, the signal acquisition unit is specifically used for: Acquire the original transfer function signal from the loudspeaker to each sound field partition, and perform frequency domain transformation on the original transfer function signal to obtain a frequency domain signal; Divide the frequency domain signal into sub-bands from low frequency to high frequency according to the intervals from small to large, to obtain signals of each frequency band; The transfer function signal is determined based on the signals in each frequency band.

[0087] Based on the above embodiment, the signal acquisition unit is specifically used for: Based on the frequency band characteristics of the signals in each frequency band, randomly perturbing the signals in each frequency band to obtain a perturbation signal; The disturbance signal and the signals in each frequency band are used as the transfer function signal.

[0088] Based on the above embodiment, the parameter iteration unit is specifically used for: Determine the frequency band loss of the signal in each frequency band based on the acoustic characteristics of the signal in each frequency band; Based on the frequency band loss and the difference between the sound field energies of the bright and dark areas under each sound field partition, the neural network model is iterated on parameters.

[0089] Based on the above embodiment, the parameter determination unit is specifically used for: The transfer function signal is sequentially input into the encoder, long short-term memory network and decoder of the neural network model to obtain the filter parameters corresponding to the speaker output by the decoder when each sound field partition is used as different light and dark areas.

[0090] Figure 7 The following is a schematic diagram of the physical structure of a car speaker. Figure 7As shown, the vehicle-mounted speaker may include: a speaker body, a processor 710, a communications interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communications interface 720 and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the sound field partition control method, which includes: obtaining a transfer function signal from the speaker to each sound field partition; based on the transfer function signal and the neural network model, obtaining the filter parameters corresponding to the speaker when each sound field partition is respectively used as a bright and dark zone; based on the difference between the sound field energy of the bright and dark zones under each sound field partition, performing parameter iteration on the neural network model, the sound field energy of the bright and dark zones is determined based on the transfer function signal and the filter parameters, and the neural network model after the parameter iteration is completed is used to determine the sound field partition filter parameters of the speaker to be controlled.

[0091] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0092] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the sound field partition control method provided by the above methods, which includes: obtaining a transfer function signal from a loudspeaker to each sound field partition; based on the transfer function signal and a neural network model, obtaining filter parameters corresponding to the loudspeaker when each sound field partition is used as a light or dark zone respectively; based on the difference between the sound field energies of the light and dark zones under each sound field partition, performing parameter iteration on the neural network model, the sound field energy of the light and dark zones is determined based on the transfer function signal and the filter parameters, and the neural network model after the parameter iteration is completed is used to determine the sound field partition filter parameters of the loudspeaker to be controlled.

[0093] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the sound field partition control method provided by the above-mentioned methods, the method comprising: obtaining a transfer function signal from a loudspeaker to each sound field partition; based on the transfer function signal and a neural network model, obtaining filter parameters corresponding to the loudspeaker when each sound field partition is used as a light or dark zone respectively; based on the difference between the sound field energies of the light and dark zones under each sound field partition, performing parameter iteration on the neural network model, the sound field energies of the light and dark zones being determined based on the transfer function signal and the filter parameters, and the neural network model after the parameter iteration is completed is used to determine the sound field partition filter parameters of the loudspeaker to be controlled.

[0094] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0095] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A sound field partition control method, characterized in that: include: Obtain transfer function signals from the loudspeaker to each sound field partition; Based on the transfer function signal and the neural network model, filter parameters corresponding to the loudspeaker when each of the sound field partitions is used as a bright area and a dark area are obtained; Based on the difference between the sound field energies of the light and dark zones under each sound field partition, the neural network model is iterated in parameters. The sound field energies of the light and dark zones are determined based on the transfer function signal and the filter parameters. The neural network model after the parameter iteration is used to determine the sound field partition filter parameters of the speaker to be controlled.

2. The sound field partition control method according to claim 1, characterized in that: The performing parameter iteration on the neural network model based on the difference between the sound field energies of the bright and dark areas under each sound field partition comprises: Determine the dark zone sound field energy of any sound field partition based on the transfer function signal of any sound field partition and the filter parameters corresponding to any sound field partition when the any sound field partition is used as a dark zone; Determine the bright zone sound field energy of any sound field partition based on the transfer function signal of any sound field partition and the filter parameters corresponding to any sound field partition when the any sound field partition is used as a bright zone; Based on the contrast between the sound field energy in the dark area and the sound field energy in the bright area in each sound field partition, parameter iteration is performed on the neural network model.

3. The sound field partition control method according to claim 2, characterized in that: The step of iterating parameters of the neural network model based on the contrast between the sound field energy in the dark area and the sound field energy in the bright area in each sound field partition includes: Determine the dark area reconstruction loss based on the dark area sound field energy of each sound field partition and the contrast between the dark area sound field energy and the bright area sound field energy; Based on the dark area reconstruction loss, parameters of the neural network model are iterated.

4. The sound field partition control method according to claim 3, characterized in that: The step of iterating parameters of the neural network model based on the dark area reconstruction loss comprises: Determine the bright area reconstruction loss based on the bright area sound field energy of each sound field partition and the difference between the target sound field energies of each sound field partition; Based on the bright area reconstruction loss and the dark area reconstruction loss, parameters of the neural network model are iterated.

5. The sound field partition control method according to any one of claims 1 to 4, characterized in that: The step of obtaining a transfer function signal from the loudspeaker to each sound field partition includes: Acquire the original transfer function signal from the loudspeaker to each sound field partition, and perform frequency domain transformation on the original transfer function signal to obtain a frequency domain signal; Divide the frequency domain signal into sub-bands from low frequency to high frequency according to the intervals from small to large, to obtain signals of each frequency band; The transfer function signal is determined based on the signals in each frequency band.

6. The sound field partition control method according to claim 5, characterized in that: The step of determining the transfer function signal based on the signals of each frequency band comprises: Based on the frequency band characteristics of the signals in each frequency band, randomly perturbing the signals in each frequency band to obtain a perturbation signal; The disturbance signal and the signals in each frequency band are used as the transfer function signal.

7. The sound field partition control method according to claim 5, characterized in that: The performing parameter iteration on the neural network model based on the difference between the sound field energies of the bright and dark areas under each sound field partition comprises: Determine the frequency band loss of the signal in each frequency band based on the acoustic characteristics of the signal in each frequency band; Based on the frequency band loss and the difference between the sound field energies of the bright and dark areas under each sound field partition, the neural network model is iterated on parameters.

8. The sound field partition control method according to any one of claims 1 to 4, characterized in that: The obtaining, based on the transfer function signal and the neural network model, filter parameters corresponding to the speaker when each of the sound field partitions is used as a bright area or a dark area respectively, comprises: The transfer function signal is sequentially input into the encoder, long short-term memory network and decoder of the neural network model to obtain the filter parameters corresponding to the speaker output by the decoder when each sound field partition is used as a light or dark area.

9. A sound field partition control device, characterized in that: include: A signal acquisition unit, used to acquire a transfer function signal from the loudspeaker to each sound field partition; A parameter determination unit, configured to obtain, based on the transfer function signal and the neural network model, filter parameters corresponding to the loudspeaker when each of the sound field partitions is used as a bright area or a dark area respectively; A parameter iteration unit is used to iterate the parameters of the neural network model based on the difference between the sound field energies of the light and dark zones under each sound field partition, wherein the sound field energies of the light and dark zones are determined based on the transfer function signal and the filter parameters, and the neural network model after the parameter iteration is completed is used to determine the sound field partition filter parameters of the speaker to be controlled.

10. A vehicle-mounted speaker, characterized in that: The method comprises a speaker body, a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the sound field zoning control method as claimed in any one of claims 1 to 8 when executing the computer program.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the sound field partition control method according to any one of claims 1 to 8 is implemented.

Images