Active noise reduction method, system, vehicle, and storage medium

Abstract

The application relates to the field of active noise reduction technology and discloses an active noise reduction method, an active noise reduction system, a vehicle and a storage medium, which are used for improving the active noise reduction capability. The method part comprises the following steps: acquiring a reference signal, wherein the reference signal comprises M first reference signals and N second reference signals, the first reference signals are environment noise collected by microphones, and the second reference signals are environment noise collected by at least one of an acceleration sensor and a vibration sensor; selecting M1 first reference signals and N1 second reference signals through coherence analysis of the reference signal and noise signals in a noise reduction area, wherein the noise signals in the noise reduction area are collected by microphones at a first preset distance from human ears; and generating an output signal according to the M1 first reference signals and the N1 second reference signals, so that a loudspeaker system plays the output signal to perform active noise reduction on the noise reduction area, and the sum of M1 and N1 is less than the sum of M and N.

Description

Technical Field

[0001] This application relates to the field of active noise cancellation technology, and more particularly to an active noise cancellation method, system, vehicle, and storage medium. Background Technology

[0002] Methods to improve in-vehicle noise mainly include passive noise reduction and active noise reduction. Passive noise reduction generally achieves its purpose by increasing the vehicle's damping / sound-absorbing materials or shock-absorbing springs. Active noise reduction mainly utilizes the vehicle's built-in speaker system to generate anti-phase sound waves to cancel out noise in the noise reduction area, thus achieving the purpose of active noise reduction.

[0003] In active noise cancellation, the primary process involves using active noise cancellation algorithms to process the noise reference signal to generate the output signal of the speaker system. This output signal is then used by the speaker system to cancel out the noise signal in the noise reduction area, thus achieving active noise cancellation. However, to achieve effective in-vehicle active noise control, a sufficient amount of noise reference signal that effectively characterizes the in-vehicle noise is required. Since there are many noise excitation sources, a large number of reference signals are needed. Furthermore, the noise sources differ in different scenarios, and some noise reference signals overlap. Current technologies that directly provide the noise reference signal to the active noise cancellation system without processing not only require enormous computing power but also increase the processing difficulty for the control algorithm, failing to achieve the desired noise control effect. Summary of the Invention

[0004] This application provides an active noise cancellation method, system, vehicle, and storage medium to solve the technical problem of poor active noise cancellation performance.

[0005] Based on the above technical issues, the following solutions are provided: In a first aspect, an active noise reduction method is provided, the method comprising: acquiring reference signals, the reference signals including M first reference signals and N second reference signals, wherein the first reference signals are ambient noise collected by a microphone, and the second reference signals are ambient noise collected by at least one of an accelerometer and a vibration sensor; selecting M1 first reference signals and N1 second reference signals by coherence analysis of the reference signals and noise signals in a noise reduction area, wherein the noise signals in the noise reduction area are collected by a microphone at a first preset distance from the human ear; generating an output signal based on the M1 first reference signals and the N1 second reference signals, such that a speaker system plays the output signal to actively reduce noise in the noise reduction area, wherein the sum of M1 and N1 is less than the sum of M and N.

[0006] Furthermore, the M first reference signals are environmental noise collected by the microphone that is not affected by wind noise; obtaining the M first reference signals includes: performing wind noise identification on the Q environmental noise collected by the microphone, and obtaining M first reference signals from the Q environmental noise collected by the microphone based on the wind noise identification result, where Q is greater than M; the wind noise of the M first reference signals is less than the wind noise of (QM) first reference signals.

[0007] Furthermore, acquiring the N second reference signals includes: acquiring the N second reference signals based on vehicle status information; Obtaining the N second reference signals based on vehicle status information includes: when the vehicle is in motion, acquiring environmental noise collected by an acceleration sensor and / or vibration sensor corresponding to the type of road surface the vehicle is traveling on; when the vehicle is stationary, if the vehicle generator is in operation, acquiring environmental noise collected by an acceleration sensor and / or vibration sensor at a second preset distance from the vehicle generator.

[0008] Further, determining the acceleration sensor and / or vibration sensor corresponding to the road surface type on which the vehicle is traveling includes: From the pre-stored correspondence, the acceleration sensor and / or vibration sensor corresponding to the vehicle driving road surface type are retrieved. The correspondence includes the correspondence between different vehicle driving road surface types and their corresponding acceleration sensors and / or vibration sensors.

[0009] Furthermore, the vehicle driving surface types include a first surface, a third surface, and a fourth surface. The surface roughness of the first surface is less than that of the second surface, and the surface roughness of the second surface is less than that of the third surface. The acceleration sensor and / or vibration sensor corresponding to the first surface includes acceleration sensors installed at the tire suspension, vehicle chassis, and connecting axle, and / or vibration sensors installed at the tire suspension, vehicle chassis, and connecting axle. The acceleration sensor and / or vibration sensor corresponding to the second surface type includes acceleration sensors installed at the tire suspension, support arm, vehicle chassis, and vehicle tower, and / or vibration sensors installed at the tire suspension, support arm, vehicle chassis, and vehicle tower. The acceleration sensor and / or vibration sensor corresponding to the third surface includes acceleration sensors installed at the tire suspension, vehicle chassis, vehicle chassis crossbeam, and vehicle tower, and / or vibration sensors installed at the tire suspension, vehicle chassis, vehicle chassis crossbeam, and vehicle tower.

[0010] Furthermore, the M first reference signals include ambient noise collected by one or more microphones closest to the noise reduction area; the N second reference signals include at least one of the following: ambient noise collected by one or more accelerometers closest to the noise reduction area; ambient noise collected by one or more vibration sensors closest to the noise reduction area.

[0011] Further, the step of selecting M1 first reference signals and N1 second reference signals through coherence analysis of the reference signals and the noise signals in the noise reduction region includes: determining the coherence of each reference signal in the reference signals with the noise signals in the noise reduction region in a preset frequency band; the coherence of the M1 first reference signals with the noise signals in the noise reduction region in the preset frequency band is greater than the coherence of (M-M1) first reference signals with the noise signals in the noise reduction region in the preset frequency band; the coherence of the N1 second reference signals with the noise signals in the noise reduction region in the preset frequency band is greater than the coherence of (N-N1) second reference signals with the noise signals in the noise reduction region in the preset frequency band.

[0012] Furthermore, before acquiring the reference signal, the method further includes: acquiring alternative reference signals; The acquisition of reference signals includes: selecting the candidate reference signal with the highest coherence as the selected reference signal based on the coherence between the candidate reference signal and the noise signal in the noise reduction area, until the number of first reference signals among the selected reference signals reaches M and the number of second reference signals reaches N.

[0013] Furthermore, it also includes: determining the noise reduction area; determining the noise reduction area includes: automatically determining it based on vehicle seating information, or determining it based on user-specified information; the noise reduction area includes one or more of the area where the head is placed on the seat and the middle area of ​​the rear row where the camping space is usually located.

[0014] Furthermore, the microphone at a first preset distance from the ear is one or more microphones closest to the ear; the noise signal in the noise reduction area is obtained through the microphone at the first preset distance from the ear, including: the noise signal collected by the microphone at the headrest position is the noise signal in the noise reduction area; when no microphone signal is set at the headrest position, the signal collected by the one or more microphones closest to the ear is processed by a path transfer function to obtain the noise signal in the noise reduction area, and the path transfer function characterizes the signal attenuation of the signal collected by the one or more microphones closest to the ear as it is transmitted to the noise reduction area.

[0015] Further, the step of generating an output signal based on the M1 first reference signals and the N1 second reference signals includes: inputting the M1 first reference signals and the N1 second reference signals, or the M1 first reference signals, the N1 second reference signals, and reference signal flags, into the trained target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and the N1 second reference signals, where P is less than or equal to the sum of M1 and N1, and the reference signal flags are bound to the M1 first reference signals and the N1 second reference signals; and generating an output signal based on the P target reference signals.

[0016] Further, generating the output signal based on the P target reference signals includes: filtering the P target reference signals according to a fixed filtering method to generate the output signal, wherein the filtering parameters used in the fixed filtering method are the filtering parameters corresponding to the reference signal flags.

[0017] Furthermore, the loudspeaker system includes a first loudspeaker group and a second loudspeaker group. The step of generating an output signal based on the P target reference signals includes: filtering the P target reference signals according to a fixed filtering method to generate an output signal for the first loudspeaker group, wherein the filtering parameters used in the fixed filtering method are the filtering parameters corresponding to the reference signal flags; and filtering the P target reference signals according to an adaptive filtering method to generate an output signal for the second loudspeaker group.

[0018] Secondly, an active noise cancellation system is provided, the active noise cancellation system comprising: A signal selection module is used to acquire reference signals, which include M first reference signals and N second reference signals. The first reference signals are ambient noise collected by a microphone, and the second reference signals are ambient noise collected by at least one of an accelerometer and a vibration sensor. Through coherence analysis of the reference signals and the noise signals in the noise reduction area, M1 first reference signals and N1 second reference signals are selected. The noise signals in the noise reduction area are acquired by a microphone at a first preset distance from the human ear. The output signal generation module is used to generate an output signal based on the M1 first reference signals and the N1 second reference signals, so that the speaker system plays the output signal to actively reduce noise in the noise reduction area, where M1+N1 is less than M+N.

[0019] Furthermore, the output signal generation module includes: The fitting mapping module is used to input the M1 first reference signals and the N1 second reference signals into the trained target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and the N1 second reference signals, where P is less than or equal to M1+N1, and the reference signal flags are bound to the M1 first reference signals and the N1 second reference signals. The filtering module is used to filter the P target reference signals to generate an output signal.

[0020] Furthermore, the output signal generation module includes: The fitting mapping module is used to input the M1 first reference signals, the N1 second reference signals, and reference signal flags into the target neural network, so that the target neural network model fits P target reference signals from the M1 first reference signals and the N1 second reference signals; The filtering module is used to filter the P target reference signals to generate an output signal.

[0021] Thirdly, a vehicle is provided, the vehicle including an active noise cancellation system for implementing the active noise cancellation method as described in any of the preceding claims; or is an active noise cancellation system as described in any of the preceding claims.

[0022] Fourthly, a readable storage medium is provided, the readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method as described in any of the preceding claims, or implements the function of the active noise cancellation system as described in any of the preceding claims.

[0023] One of the solutions provided in this application offers an active noise reduction method, which involves a noise reference signal selection method. This method selects a reference signal related to the noise source in the noise reduction area from a large number of reference signals as the output signal of the loudspeaker system. This reduces the signal input processing load of the active noise reduction system, effectively reduces the computing power requirement for output signal generation, avoids directly using a large number of collected noise reference signals to participate in the calculation of the output signal, and simplifies the calculation process. This reduces computational complexity and resource waste, reduces processing difficulty, and also helps to improve the real-time noise reduction performance of the active noise reduction system. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the system framework of an active noise reduction system according to one embodiment of this application; Figure 2 This is a flowchart illustrating an active noise reduction method according to an embodiment of this application; Figure 3 This is another flowchart illustrating an active noise reduction method in one embodiment of this application; Figure 4 This is a schematic diagram of a training process for a target neural network model according to an embodiment of this application; Figure 5 This is a schematic diagram of a training process for a target neural network model according to an embodiment of this application; Figure 6 This is a schematic diagram of another training process of a target neural network model in one embodiment of this application; Figure 7 This is a schematic diagram of another training process of a target neural network model in one embodiment of this application; Figure 8 This is a schematic diagram of another training process of a target neural network model in one embodiment of this application; Figure 9 This is a schematic diagram of a process for generating an output signal in one embodiment of this application; Figure 10 This is another schematic diagram of the process for generating the output signal in one embodiment of this application; Figure 11 This is another schematic diagram of the process for generating the output signal in one embodiment of this application. Figure 12 This is a schematic diagram of a specified result of the noise reduction region in one embodiment of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] To facilitate understanding of the embodiments of this application, please refer to the following: Figure 1 The following section introduces the relevant concepts involved in the embodiments of this application.

[0028] Reference Signal Sensor Set: The reference signal sensor set is mainly used to acquire a first reference signal and a second reference signal. The reference signal sensors in the set include an accelerometer, a vibration sensor, and a microphone. The layout of the reference signal sensors is designed to acquire as many reference signals as possible for active noise cancellation. For example, the second reference signal might be ambient noise acquired by the accelerometer, or ambient noise acquired by the vibration sensor, or ambient noise acquired by both the accelerometer and vibration sensors, while the first reference signal is ambient noise acquired by the microphone. It should be understood that both vibration sensors and accelerometers can be used to acquire ambient noise characterizing vibration in automotive applications. The difference in automotive applications lies in the measurement location and application scenario of the accelerometer and vibration sensors. Vibration sensors are mainly used to measure the vibration or tremor of an object, acquiring vibration-related signals such as frequency, amplitude, and phase. Accelerometers are mainly used to measure changes in the acceleration of an object, acquiring primarily acceleration information. As an example, an acceleration sensor can be positioned on the vehicle's suspension, connecting rods, chassis, frame, roof, and pillars to pick up noise caused by road / tire excitation and wind noise during vehicle operation. In other words, the acceleration sensor is used to collect noise signals including, but not limited to, road noise, tire noise, and wind noise. A vibration sensor can be positioned at a preset distance from the engine / generator to pick up corresponding environmental noise, including, but not limited to, noise signals from the generator / engine. A microphone can be positioned to facilitate the collection of external noise signals. For example, microphones can be primarily positioned outside the vehicle, such as in gaps around windows and doors, and can also be positioned around gaps inside the vehicle to pick up external environmental noise. It should be noted that the above examples are merely illustrative and do not limit the type of reference signal sensor in the specific reference signal sensor set. For example, an acceleration sensor can be replaced by a vibration sensor; no specific limitations are made, and no detailed explanation is provided.

[0029] Error sensor: This error sensor is used to collect noise signals within the noise reduction area. The noise signals within the noise reduction area are collected by a microphone located at a first preset distance from the listener's ear. For example, the error sensor can be positioned near the armrest or different seat headrests. It should be noted that different error sensors can directly or indirectly reflect the noise conditions of different noise reduction areas. For example, different error sensors can directly or indirectly reflect the noise at different seating positions. Therefore, depending on the noise reduction area requiring noise reduction processing, the signal collected by the error sensor within that noise reduction area can be selected to represent the noise signal of that noise reduction area. For example, the noise reduction area includes, but is not limited to, the area where the headrest is placed in the front seats or the area where the headrest is placed in the rear seats. Taking the area where the headrest is placed in the front seats as an example, the signal collected by the microphone in the area where the headrest is placed in the front seats represents the noise signal in that front area. The microphone in the front area is the signal collected by the microphone located at a first preset distance from the listener's ear within that front area. The vehicle status information providing module is used to provide vehicle status information.

[0030] Below, as Figure 2 As shown, combined with Figure 1 The active noise reduction processing framework of this application is described in detail, along with an active noise reduction method provided in the embodiments of this application.

[0031] like Figure 2 As shown, in one embodiment, an active noise reduction method is provided, the method comprising: S10A, Obtain the reference signal; In this step, as an example, the set of reference signal sensors deployed in the vehicle can be referred to the foregoing explanation, and will not be repeated here. The active noise cancellation system uses a signal selection module to select the reference signals collected by the required reference signal sensors to obtain the desired reference signals. The reference signals include M first reference signals and N second reference signals. The first reference signals are the ambient noise collected by the microphone, and the second reference signals are the ambient noise collected by the accelerometer, or the ambient noise collected by the vibration sensor, or the ambient noise collected by both the accelerometer and vibration sensor. It can also be understood that the reference signals obtained in step S10A include M first reference signals and N second reference signals characterizing the noise source in the noise reduction area.

[0032] S20A: Through coherence analysis of the reference signal and the noise signal in the noise reduction area, select M1 first reference signals and N1 second reference signals. The noise signal in the noise reduction area is collected by a microphone at a first preset distance from the human ear. After selecting and acquiring the reference signal in step S10A, the signal selection module will perform coherence analysis on the acquired reference signal and the noise signal in the noise reduction area. That is, it will perform coherence analysis on the acquired M first reference signals and N second reference signals and the noise signal in the noise reduction area. Based on the coherence analysis results, it will select M1 first reference signals and N1 second reference signals from the M first reference signals and N second reference signals respectively. The microphone in the noise reduction area is a microphone at a first preset distance from the human ear. The sum of M1 and N1 is less than the sum of M and N.

[0033] In one embodiment, the M first reference signals include ambient noise collected by one or more microphones closest to the noise reduction area; the N second reference signals include at least one of the following: ambient noise collected by one or more accelerometers closest to the noise reduction area; ambient noise collected by one or more vibration sensors closest to the noise reduction area.

[0034] It should be noted that the noise signal within the noise reduction area can be obtained by a microphone located at a first preset distance from the ear within the noise reduction area. This microphone at the first preset distance from the ear refers to one or more microphones closest to the ear. In one embodiment, the noise signal within the noise reduction area includes signals collected by microphones located within the noise reduction area, or signals collected by microphones located not far from the noise reduction area. The specific method is not limited. For example, if the noise reduction area includes the front row area, signals collected by microphones in the front row area can be selected. Exemplarily, obtaining the noise signal within the noise reduction area through microphones at a first preset distance from the ear includes: noise signals collected by microphones at the headrest location; when no microphone signal is provided at the headrest location, signals collected by one or more microphones closest to the ear are processed by a path transfer function to obtain the noise signal within the noise reduction area. The path transfer function characterizes the signal attenuation of the signals collected by the one or more microphones closest to the ear as they are transmitted to the noise reduction area.

[0035] A noise reduction zone refers to any area inside the vehicle that requires active noise cancellation. This zone includes, but is not limited to, the front or rear passenger areas. It should be understood that in actual vehicles, microphones are often not available or cannot be placed in the noise reduction zone; microphones located at a distance from the noise reduction zone are the only option. For example, taking a seat area as the noise reduction region, the signals collected by microphones within this region can include signals from microphones located at or around the headrest, or signals from microphones around the door handle next to the seat. The specific microphones used for this noise reduction region can be pre-designed offline and represented by coefficients, which are then written into the active noise cancellation system. This allows the system to determine which microphones should be used for the noise reduction region based on these coefficients. For instance, if a microphone is located at the headrest, the noise signal collected by that microphone is the noise signal within the noise reduction region. If no microphone is located at the headrest, the signals from the one or more microphones closest to the listener's ear are processed by a path transfer function to obtain the noise signal within the noise reduction region. The path transfer function represents the signal attenuation of the one or more microphones closest to the listener's ear as they travel to the noise reduction region, thus ensuring the accuracy of the noise signal acquisition within the noise reduction region. For example, continuing as... Figure 1 As shown, assuming the acquired reference signals include M first reference signals and N second reference signals (e.g., A1 accelerometer signals, B1 vibration sensor signals, and C1 microphone signals, N=A1+B1, M=C1), the signal selection module will perform multicoherence analysis on the M first reference signals and N second reference signals with the noise signals in the noise reduction area to select M1 first reference signals and N1 second reference signals from the M first reference signals and N second reference signals respectively (e.g., A2 accelerometer signals, B2 vibration sensor signals, and C2 microphone signals, N1=A2+B2, M1=C2, A2≤A1, B2≤B1, C2≤C1).

[0036] S30A generates an output signal based on M1 first reference signals and N1 second reference signals, so that the speaker system plays the output signal to actively reduce noise in the noise reduction area.

[0037] In this step, after the signal selection module selects M1 first reference signals and N1 second reference signals, the active noise cancellation system allows the output signal generation module to generate the output signal of the vehicle's speaker system (which can also be understood as the vehicle's horn system) based on the M1 first reference signals and N1 second reference signals. This output signal is the inverse noise signal ultimately used to cancel the noise signal in the noise cancellation area, thereby achieving active noise cancellation processing in the noise cancellation area. Figure 1As shown, after selecting M1 first reference signals and N1 second reference signals, the M1 first reference signals and N1 second reference signals are given to the output signal generation module, so that the output signal generation module generates an output signal for the speaker system based on the M1 first reference signals and N1 second reference signals, so that the speaker system plays the output signal to actively reduce noise in the noise reduction area.

[0038] As can be seen, in this embodiment, the reference signal related to the noise signal source in the noise reduction area is selected from a large number of reference signals as the output signal of the loudspeaker system. This reduces the signal input processing load of the active noise cancellation system, effectively reduces the computing power requirement for output signal generation, avoids directly using a large number of collected noise reference signals to participate in the calculation of the output signal, and the calculation process is also relatively simple, reducing computational complexity and resource waste, reducing processing difficulty, and also helping to improve the real-time noise reduction performance of the active noise cancellation system.

[0039] It should be noted that, as mentioned in the foregoing embodiments, in order to further improve the real-time filtering efficiency and signal effectiveness of the reference signals, when selecting N second reference signals, vehicle status information may be considered for selection, noise reduction area may be considered for selection, or both vehicle status information and noise reduction area may be considered for selection. The following three cases are used to select and obtain N second reference signals, which will be described below.

[0040] The first method involves acquiring N second reference signals based on vehicle state information. In this case, the selected N second reference signals include those chosen based on the vehicle state information. The second reference signals are ambient noise collected by at least one of the corresponding acceleration sensor and vibration sensor under the given vehicle state information. For example, the vehicle state information includes state information characterizing the vehicle's driving state or its stationary state.

[0041] In one embodiment, acquiring N second reference signals based on vehicle state information includes: when the vehicle is in motion, acquiring environmental noise collected by an acceleration sensor and / or vibration sensor corresponding to the type of road surface the vehicle is traveling on; when the vehicle is stationary, if the vehicle's generator is operating, acquiring environmental noise collected by an acceleration sensor and / or vibration sensor at a second preset distance from the vehicle's generator. In some scenarios, for example, when the vehicle is stationary and the engine and generator are not operating, N equals 0, meaning that in this scenario, all acquired reference signals are first reference signals, and there are no second reference signals.

[0042] In this embodiment, when selecting N second reference signals, vehicle state information is taken into account. Specifically, if the vehicle is in motion, the road surface type is identified, and the corresponding second reference signal is obtained; if the vehicle is stationary, the corresponding second reference signal for that stationary state is obtained. As an example, noise spectrum analysis or noise energy analysis can be used to classify and identify the current road surface type, and then the corresponding second reference signal can be obtained. Alternatively, a neural network (NN) approach can be used to classify and identify the road surface type, and then the corresponding second reference signal can be obtained. The specific method is not limited.

[0043] It should be noted that different road surface types for different vehicles will use corresponding reference signal sensors to collect second reference signals, thereby obtaining a matching or suitable second reference signal for the current vehicle status, which helps to improve the targeting of active noise cancellation.

[0044] In one embodiment, the vehicle driving surface types include a first surface, a second surface, and a third surface. The surface roughness of the first surface is less than that of the second surface, and the surface roughness of the second surface is less than that of the third surface. A second reference signal is acquired from a corresponding reference signal sensor based on the different vehicle driving surface types. The reference signal sensors corresponding to the first surface include those located at the tire suspension, vehicle chassis, and connecting bridge. The reference signal sensors corresponding to the second surface include those located at the tire suspension, support arms, vehicle chassis, and vehicle tower top. The reference signal sensors also include those located at the tire suspension, vehicle chassis, vehicle chassis crossbeams, and vehicle tower top. Each reference signal sensor includes at least one of an acceleration sensor and a vibration sensor. It should be noted that the first surface may include a smooth asphalt surface, the second surface may include a rough cobblestone surface, and the third surface may include a rough brick surface. The above driving surface types are merely illustrative and are not specifically limited.

[0045] In one embodiment, the acceleration sensor and / or vibration sensor corresponding to the first road surface includes acceleration sensors disposed at the tire suspension, vehicle chassis, and connecting bridge positions, and / or vibration sensors disposed at the tire suspension, vehicle chassis, and connecting bridge positions; the acceleration sensor and / or vibration sensor corresponding to the second road surface includes acceleration sensors disposed at the tire suspension, support arm, vehicle chassis, and vehicle tower top positions, and / or vibration sensors disposed at the tire suspension, support arm, vehicle chassis, and vehicle tower top positions; the acceleration sensor and / or vibration sensor corresponding to the third road surface type includes acceleration sensors disposed at the tire suspension, vehicle chassis, vehicle chassis crossbeam, and vehicle tower top positions, and / or vibration sensors disposed at the tire suspension, vehicle chassis, vehicle chassis crossbeam, and vehicle tower top positions. It should be noted that the above-mentioned vehicle driving road surface types and corresponding reference signal sensors are only illustrative examples. In practical applications, various different vehicle driving road surface type classifications and reference signal sensor groups can be pre-designed to select the desired reference signal during application, thereby improving the practicality and applicability of the scenario. For example, vehicle driving road surface types can also include other different levels of road surface roughness, such as the fourth road surface, ... the Nth road surface, etc. The second reference signal collected by the corresponding reference signal sensor can be obtained under each different road surface. Examples are not provided here.

[0046] As can be seen, in this embodiment, if the vehicle is in motion, the type of road surface it is driving on will be identified, and then the second reference signal corresponding to the type of road surface will be obtained. In this way, the best second reference signal for the current road surface will be found, which will effectively ensure that the most effective second reference signal representing the noise source of the noise reduction area under the current road surface is selected, thereby ensuring the noise reduction performance of the active noise reduction system.

[0047] In one embodiment, determining the acceleration sensor and / or vibration sensor corresponding to the vehicle's road surface type includes: querying the acceleration sensor and / or vibration sensor corresponding to the vehicle's road surface type from a pre-stored correspondence relationship, wherein the correspondence relationship includes the correspondence between different vehicle road surface types and their corresponding acceleration sensors and / or vibration sensors.

[0048] In this embodiment, the correspondence between different vehicle road surface types and their optimal second reference signals can be pre-analyzed, trained, and saved. That is, all second reference signals and noise signals within the noise reduction area for each vehicle road surface type are pre-collected. Through offline analysis, the optimal second reference signal for each vehicle road surface is determined, thus obtaining and saving the aforementioned correspondence. During subsequent actual operation, the corresponding second reference signal for the current road surface type can be retrieved directly by looking up this correspondence. As an example, this correspondence can be a lookup table built into the active noise cancellation system or the vehicle's infotainment system. Then, when the vehicle is running online, the optimal second reference signal for each vehicle road surface type can be found by looking up the table.

[0049] In this embodiment, different vehicle road surface types and their corresponding optimal second reference signals can be pre-trained. The optimal second reference signal can then be found through a query, significantly reducing the signal selection processing workload during actual real-time operation and improving the real-time efficiency of active noise reduction. It should also be noted that in other embodiments, the selected second reference signal for the chosen vehicle road surface type can be determined through real-time computation, rather than a pre-configured correspondence; this is not specifically limited. Furthermore, in other embodiments, in addition to the vehicle road surface type, vehicle state information can also reference other dimensions of state information to select the corresponding second reference signal under that vehicle state information; this is not specifically limited.

[0050] The second type involves N second reference signals, including ambient noise collected by an accelerometer within the noise reduction region and / or ambient noise collected by a vibration sensor within the noise reduction region.

[0051] In other words, the N second reference signals include ambient noise collected by the accelerometer within the noise reduction area, or the N second reference signals include ambient noise collected by the vibration sensor within the noise reduction area, or the N second reference signals include ambient noise collected by both the accelerometer and vibration sensor within the noise reduction area. For example, if only noise reduction is needed in the front row area, then more noise signals collected by the accelerometer and / or vibration sensor in the front row area of ​​the vehicle are selected; if noise reduction is needed in both the front and rear row areas, then the second reference signals collected by the accelerometer and / or vibration sensor in both the front and rear row areas need to be considered.

[0052] As can be seen, in this embodiment, in addition to selecting the second reference signal based on vehicle state information, the above-mentioned N second reference signals can also be selected based on the noise reduction area. Since the noise source of the noise reduction area is also related to the noise reduction area, selecting the above-mentioned N second reference signals based on the noise reduction area can ensure that the obtained N second reference signals are strongly correlated with the noise source of the noise reduction area, effectively characterizing the noise situation of the noise reduction area, thereby reducing the signal volume of coherence analysis, ensuring the subsequent active noise reduction capability, and also making the implementation of the solution more diverse.

[0053] It's also worth noting that the noise reduction area inside the vehicle may change depending on the scenario. For example, when the vehicle is moving, noise reduction might only target the head area of ​​the driver's seat depending on the occupancy rate, or when the vehicle is stationary, noise reduction might only target the head area of ​​the folded-down passenger seat. Furthermore, the sources of noise in different areas of the vehicle will vary. For instance, road noise in the driver's seat area is more likely to originate from vibrations transmitted from the front left of the vehicle, while road noise in the rear right seat area is more likely to originate from vibrations transmitted from the rear. Therefore, this embodiment can effectively filter out the second reference signal that better characterizes the noise source of the noise reduction area. As an example, the correspondence between the noise reduction area and the optimal N second reference signals can also be obtained and saved in advance through offline pre-training, without any specific limitations. During actual operation, the corresponding N second reference signals for the noise reduction area can be retrieved directly by looking up this correspondence. As an example, this correspondence can also be a lookup table built into the active noise cancellation system or the vehicle's infotainment system. Then, during online operation, the optimal N second reference signals for the noise reduction area can be found by looking up the table.

[0054] The third type involves N second reference signals, which are determined jointly based on both vehicle state information and the noise reduction area. That is, the N second reference signals can be selected based on both vehicle state information and the noise reduction area. For example, vehicle state information includes either the vehicle's driving state or its stationary state. For instance, the optimal N second reference signals for each type of road surface and noise reduction area can be pre-trained offline, allowing the corresponding N second reference signals to be found directly online according to the noise reduction area and the type of road surface.

[0055] In this embodiment, in addition to obtaining N second reference signals based on vehicle status information, N second reference signals can be further selected by combining the noise reduction area, effectively filtering out the second reference signal that best represents the noise source of the noise reduction area, and further ensuring the subsequent active noise reduction capability.

[0056] In one embodiment, in conjunction with any of the above embodiments, the M first reference signals are environmental noise collected by the microphone that is not affected by wind noise. Obtaining the M first reference signals includes: performing wind noise identification on the Q environmental noise collected by the microphone, and obtaining the M first reference signals from the Q environmental noise collected by the microphone based on the wind noise identification result, where Q is greater than M, and the wind noise of the M first reference signals is less than the wind noise of (QM) first reference signals.

[0057] In this embodiment, the M first reference signals are the effective ambient noise collected by the microphone that is not affected by wind noise. During the selection process of input reference signals for active noise reduction, wind noise identification can also be performed on the ambient noise collected by the microphone to further filter out the effective ambient noise that is not affected by wind noise and select the M first reference signals. This avoids the need for more processing of wind noise, improves signal processing efficiency, and enhances the timeliness of active noise reduction.

[0058] In one embodiment, step S20A, namely, selecting M1 first reference signals and N1 second reference signals through coherence analysis of the reference signals and the noise signals in the noise reduction region, includes: determining the coherence of each reference signal in the reference signals in the preset frequency band and the noise signals in the noise reduction region; the coherence of the M1 first reference signals in the preset frequency band and the noise signals in the noise reduction region is greater than the coherence of the (M-M1) first reference signals in the preset frequency band and the noise signals in the noise reduction region; the coherence of the N1 second reference signals in the preset frequency band and the noise signals in the noise reduction region is greater than the coherence of the (N-N1) second reference signals in the preset frequency band and the noise signals in the noise reduction region.

[0059] In this embodiment, M first reference signals and N second reference signals are subjected to coherence analysis, such as multiple coherence analysis, with the noise signal within the denoising region to select M1 first reference signals and N1 second reference signals. Furthermore, the preset frequency band can be set empirically, for example, based on the frequency band of the noise signal within the denoising region; no specific limitation is imposed. The calculated coherence is usually represented numerically; therefore, comparing the coherence values ​​reveals the correlation level. It should be noted that in one embodiment, the signals can be sorted from high to low coherence, and then M1 first reference signals and N1 second reference signals are selected from the reference signals. The number of selected M1 and N1 signals can be pre-configured or selected based on the coherence sorting; no specific limitation is imposed.

[0060] In one embodiment, in conjunction with any of the above embodiments, obtaining a reference signal includes: selecting candidate reference signals one by one and calculating the coherence between the candidate reference signals and the noise signals in the noise reduction area; selecting the candidate reference signal with the highest coherence as the selected reference signal, until the number of first reference signals among the selected reference signals reaches M and the number of second reference signals reaches N. During the process of obtaining M first reference signals and N second reference signals, candidate reference signals that were not selected as first or second reference signals can be selected one by one; combined with the already selected first and second reference signals, the coherence between the candidate reference signals and the noise signals in the noise reduction area can be calculated; the candidate reference signal with the highest coherence can be selected as the selected reference signal; this process is repeated until the number of first reference signals among the selected reference signals reaches M and the number of second reference signals reaches N.

[0061] In this embodiment, during the screening of reference signals, coherence analysis can be performed on each of the noise signals in the noise reduction area to achieve initial screening, thereby improving the effectiveness of the acquired reference signals. Furthermore, when calculating the coherence between the selected M first reference signals and N second reference signals and the noise signals in the noise reduction area, the correlation calculation process of some previously calculated reference signals can be omitted, which is beneficial to improving the overall signal processing timeliness.

[0062] like Figure 3 As shown, in one embodiment, an active noise reduction method is provided, the method comprising: S10B, Obtain the reference signal; S20B: Through coherence analysis of the reference signal and the noise signal in the noise reduction area, select M1 first reference signals and N1 second reference signals, and the microphone in the noise reduction area is a microphone at a first preset distance from the human ear. S30B: Input M1 first reference signals and N1 second reference signals into the trained target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and N1 second reference signals, where P is less than or equal to M1+N1, and the reference signal labels are bound to the M1 first reference signals and N1 second reference signals. S40B: Generates an output signal based on P target reference signals.

[0063] The processing procedures of steps S10B and S20B can be referred to the description of steps S10A-S20A in the foregoing embodiments. Details and effects of other embodiments of the above two steps can also be referred to the description of the foregoing embodiments, and will not be repeated here.

[0064] In this embodiment, for steps S30B-S40B, to further reduce the dimensionality of the reference signals in subsequent processing and improve the coherence between the selected reference signals and the noise signals in the denoising region, after selecting M1 first reference signals and N1 second reference signals, the active denoising system also uses a fitting mapping module to input the M1 first reference signals and N1 second reference signals into the trained target neural network model. This allows the target neural network model to fit P target reference signals from the M1 first reference signals and N1 second reference signals, where P is less than or equal to M1+N1. In other words, the active denoising system can also use the fitting mapping module to achieve signal dimensionality reduction and nonlinear fitting processing. The output signal generation module of the active denoising system then generates the output signal of the speaker system based on the P target reference signals fitted by the fitting mapping module. The number of target reference signals is less than or equal to the sum of the number of first and second reference signals, that is, P is less than or equal to M1+N1. When P is less than M1+N1, signal dimensionality reduction is achieved, thereby reducing the computational load of subsequent processing of the target reference signals. In this embodiment, the reference signal flag may not be used as training data input into the model for training and use, which can reduce training complexity and improve model applicability.

[0065] In one embodiment, an active noise reduction method is also provided, the method comprising: S10D, acquire reference signal; S20D: Through coherence analysis of the reference signal and the noise signal in the noise reduction area, select M1 first reference signals and N1 second reference signals. The microphone in the noise reduction area is a microphone at a first preset distance from the human ear. S30D: Input M1 first reference signals, N1 second reference signals, and reference signal flags into the trained target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and N1 second reference signals, where P is less than or equal to M1+N1, and the reference signal flags are bound to the M1 first reference signals and N1 second reference signals. S40D: Generates an output signal based on P target reference signals.

[0066] The processing steps S10D and S20D can be referred to the description of steps S10A-S20A in the aforementioned embodiments. Details and effects of other embodiments of the above two steps can also be referred to the descriptions of the aforementioned embodiments, and will not be repeated here. It should be noted that in this embodiment, the reference signal flag is bound to M1 first reference signals and N1 second reference signals. The reference signal flag is used to characterize the source of the reference signal; different reference signal flags represent different reference signal sources. The form of the reference signal flag is not limited; for example, it can be represented by numbers such as 1, 2, 3, or other bases. Each reference signal is bound to one reference signal flag. For example, the reference signal flag for the reference signal collected by the microphone at the front seat headrest is 1, the reference signal flag for the reference signal collected by the microphone at the roof is 2, and the reference signal flag for the reference signal collected by the vibration sensor near the engine is 3.

[0067] In one embodiment, an active noise reduction method is provided, the method comprising: S10C: Acquire M1 first reference signals, N1 second reference signals, and reference signal flags, wherein the M1 first reference signals and N1 second reference signals are reference signals selected from the M first reference signals and N second reference signals; S20C: Input M1 first reference signals, N1 second reference signals, and reference signal flags into the trained target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and N1 second reference signals, where P is less than or equal to the sum of M1 and N1, and the reference signal flags are bound to the M1 first reference signals and N1 second reference signals. The training objective of the target neural network model is that the coherence between the P target reference signals output by the model and the noise signal in the noise reduction region meets the preset training requirements. S30C: Generates an output signal based on P target reference signals.

[0068] In this embodiment, the processing steps S20C-S30C are similar to those in the previous embodiments S30B-S40B. However, it is worth emphasizing that the M1 first reference signals and N1 second reference signals in this embodiment can be reference signals selected or generated from the M first reference signals and N second reference signals in the previous embodiments, or they can be unfiltered reference signals. When the M1 first reference signals and N1 second reference signals in this embodiment are unfiltered reference signals, the number of reference signals is large, resulting in a large computational load. When the M1 first reference signals and N1 second reference signals are reference signals selected or generated from the M first reference signals and N second reference signals according to the previous embodiments, on the one hand, these selected reference signals contain more noise in the noise reduction region, which makes the noise suppression effect of the final output signal better after being played by the speaker; on the other hand, the computational load is reduced due to the reduced number of reference signals. It should also be noted that in this embodiment, the reference signal flag can be used to train and select the corresponding neural network model parameters according to different reference signals, making the model processing more targeted, the model fitting results more accurate, and the noise reduction effect effectively improved.

[0069] In one embodiment, an active noise reduction method is provided, the method comprising: S10E: Acquire M1 first reference signals and N1 second reference signals, wherein the M1 first reference signals and N1 second reference signals are reference signals selected from the M first reference signals and N second reference signals; S20E: M1 first reference signals and N1 second reference signals are input into the trained target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and N1 second reference signals, where P is less than or equal to the sum of M1 and N1; The training objective of the target neural network model is that the coherence between the P target reference signals output by the model and the noise signal in the noise reduction region meets the preset training requirements. S30E: Generates an output signal based on P target reference signals.

[0070] In this embodiment, the processing steps S20E-S3E are similar to those in the previous embodiments (S30D-S40D). However, it's worth emphasizing that the M1 first reference signals and N1 second reference signals in this embodiment can be reference signals selected or generated from the M first reference signals and N second reference signals in the previous embodiments, or they can be unfiltered reference signals. When the M1 first reference signals and N1 second reference signals in this embodiment are unfiltered, the number of reference signals is large, resulting in a high computational load. When the M1 first reference signals and N1 second reference signals are selected or generated from the M first reference signals and N second reference signals according to the previous embodiments, on the one hand, these selected reference signals contain more noise in the noise reduction region, resulting in better noise suppression of the final output signal after playback by the speaker; on the other hand, the computational load is reduced due to the reduced number of reference signals. It should also be noted that in this embodiment, the reference signal flags may not be used as model training data input to the model, which can reduce training complexity and improve model applicability.

[0071] In one embodiment, the selection method of the M first reference signals and N second reference signals includes, but is not limited to, the signal selection method described in the preceding embodiments S10A-S20A. That is, by analyzing the coherence between the reference signals and the noise signals within the noise reduction area, M1 first reference signals and N1 second reference signals are selected. The microphone within the noise reduction area is a microphone located at a first preset distance from the human ear. For details, please refer to the descriptions in the above embodiments, which will not be repeated here. Other signal selection methods can also be used, such as randomly selecting M1 first reference signals and N1 second reference signals from the M first reference signals and N second reference signals, respectively. The specific method is not limited.

[0072] In this embodiment, after selecting M1 first reference signals and N1 second reference signals, the active noise reduction system also uses a fitting mapping module to input the M1 first reference signals, N1 second reference signals, and their reference signal flags into the trained target neural network model. This allows the target neural network model to fit P target reference signals from the M1 first reference signals and N1 second reference signals, achieving not only signal dimensionality reduction but also nonlinear signal fitting. Finally, the output signal generation module of the active noise reduction system generates an output signal based on the P target reference signals fitted by the fitting mapping module. This output signal is then used by the speaker system to cancel out noise in the noise reduction area, effectively ensuring the active noise reduction capability.

[0073] In one embodiment, a neural network training method is also provided, characterized in that the method includes: S101. Obtain the first reference signal of M1 samples, the second reference signal of N1 samples, and the reference signal flag.

[0074] S102. Using M1 sample first reference signals, N1 sample second reference signals and reference signal flags as model inputs, and P target output reference signals as model outputs, train the neural network until the trained target neural network model is obtained. As an example, the training objective includes ensuring that the multicoherence of the P target reference signals output by the neural network meets the preset training requirements with the noise signal within the denoising region. As another example, M1 sample first reference signals and N1 sample second reference signals are selected from the M sample first reference signals and N sample second reference signals, respectively. It should be noted that to obtain the P target reference signals in this embodiment, the target neural network model needs to be pre-trained. The neural network input (M1 sample first reference signals, N1 sample second reference signals, and reference signal flags) and the neural network output (i.e., the P target output reference signals) are used as constraints. The training objective is the coherence between the P target output reference signals and the noise signal within the denoising region, and / or the nonlinear fitting result between the P target output reference signals and the noise signal within the denoising region. The neural network is trained until the trained neural network is obtained, which is referred to as the target neural network model in this embodiment. It should be understood that during training, a large amount of training data can be used to improve training convergence efficiency. It should be understood that a certain amount of training data needs to be collected to train the neural network so that the neural network can better meet the training objectives. Generally, the more data collected, the more helpful it is for network training and convergence. It needs to include a large number of M1 sample first reference signals and N1 sample second reference signals.

[0075] In one embodiment, for any of the aforementioned model training embodiments, partial training data can be generated based on the linear or nonlinear relationship between the known M1 sample first reference signals and N1 sample second reference signals and the noise signals in the denoising region, so as to provide more useful training data to help model training and accelerate the model training process.

[0076] In one embodiment, during training in any of the above embodiments, the training objective may further include the nonlinear fitting result as an optimization objective, without specific limitations. By optimizing the nonlinear fitting result, the nonlinear fitting effect of the output P target reference signals can also achieve the desired objective, thereby improving the nonlinear fitting effect of the signal.

[0077] like Figure 4As shown, as one training example, assume the target neural network model transforms M1 first reference signals and N1 second reference signals into P target reference signals, and the number of reference signals output by the model changes from M1+N1 to P. During model training, the neural network input (multiple sets of M1 sample first reference signals, N1 sample second reference signals, and reference signal labels) and the neural network output (P target output reference signals) are used as constraints. The training objective is to select the P target output reference signals as the set of signals with the highest multicoherence with the noise signals in the denoising region. This process continues until the target neural network model is obtained. When using the model, the obtained M1 first reference signals, N1 second reference signals, and their reference signal labels are input into the target neural network model. This yields the P target reference signals output by the target neural network model, fitted from the M1 first reference signals and N1 second reference signals. Alternatively, for example, the training objective could be: the P target output reference signals are the set of signals with the highest coherence with the noise signals in the denoising region. Of course, in specific training, it can also be a constraint condition for other forms of training objectives, which are not specifically limited here.

[0078] As can be seen, in this embodiment, by flexibly configuring the training objectives and constraints, a target neural network model that meets the requirements for signal screening can be trained. This allows the target neural network model to further process the input M1 first reference signals and N1 second reference signals to achieve, but is not limited to, signal dimensionality reduction and nonlinear fitting. In other words, the fitting mapping module utilizes the trained target neural network model primarily to extract signals (including nonlinear and linear) from the M1 first reference signals and N1 second reference signals that are coherent with the noise signals in the noise reduction region. Therefore, while completing dimensionality reduction ((M1+N1)→P) and cost reduction, it can also complete nonlinear fitting. Furthermore, the processing of the fitting mapping module is also independent of secondary paths (such as the response from the speaker to the noise reduction region). Therefore, it can better learn the relationship between the first and second reference signals and the noise signals in the noise reduction region without being affected by secondary paths, thus completing dimensionality reduction, linear and nonlinear fitting. In this way, the P target reference signals output are processed more easily to obtain better noise reduction results.

[0079] In one embodiment, the model input training data includes M1 sample first reference signals and N1 sample second reference signals corresponding to various different reference signal flags. The neural network is trained using the M1 sample first reference signals, N1 sample second reference signals, and reference signal flags as model inputs and P target output reference signals as model outputs until the trained target neural network model is obtained. This includes: training the neural network using each reference signal flag and its corresponding M1 sample first reference signals and N1 sample second reference signals as model inputs and P target output reference signals as model outputs until the model parameters corresponding to each reference signal flag are trained to obtain the target neural network model.

[0080] As one example of training, such as Figure 5 As shown in the training process, assuming the target neural network model transforms M1 first reference signals and N1 second reference signals into P target reference signals, during model training, the neural network input (multiple sets of M1 sample first reference signals, N1 sample second reference signals, and reference signal labels) and the neural network output (P target output reference signals) are used as constraints. The training objective is to train the neural network with the P target output reference signals that have the highest multicoherence with the noise signal in the denoising region. Specifically, for each type of reference signal label, the corresponding model parameters are trained using the M1 sample first reference signals and N1 sample second reference signals, until different target neural network models are obtained. That is, for each different reference signal label, the corresponding model parameters are trained using the M1 sample first reference signals and N1 sample second reference signals. Alternatively, as an example, the training objective could be the multicoherence effect between the model's P target output reference signals and the noise signal in the denoising region within a preset time period to ensure the robustness of the model output results. Figure 5 As shown in the model usage process, when using the model, the model parameters of the target neural network model can be selected based on the reference signal flags. The different model parameters trained above can be pre-stored in the storage unit of the active noise reduction system. After inputting M1 first reference signals, N1 second reference signals, and reference signal flags into the trained target neural network model, the target neural network model can select the corresponding model parameters through the reference signal flags, enabling the target neural network model to complete signal processing and output P target reference signals.

[0081] As can be seen, this embodiment provides a model training and usage example. During model training, the model parameters corresponding to each or several reference signal flags can be trained separately, so that when using the model, the corresponding model parameters can be selected according to the reference signal flags. This makes the signal processing more granular, more targeted and effective, and further ensures that the selected target reference signals are more conducive to subsequent active noise reduction, thus ensuring the performance of active noise reduction.

[0082] In one embodiment, during training, M1 sample first reference signals and N1 sample second reference signals are input into the neural network in a sample-point granularity signal input format, and the neural network outputs P target output reference signals in a sample-point granularity signal output format; or, M1 sample first reference signals and N1 sample second reference signals are input into the neural network in a frame-point granularity signal input format, and the neural network outputs P target output reference signals in a frame-point granularity signal output format. Correspondingly, during use, M1 first reference signals and N1 second reference signals are input into the target neural network model in a sample-point granularity signal input format, and the target neural network model outputs P target reference signals in a sample-point granularity signal output format; or, M1 first reference signals and N1 second reference signals are input into the target neural network model in a frame-point granularity signal input format, and the target neural network model outputs P target reference signals in a frame-point granularity signal output format.

[0083] As a training example, such as Figure 6 As shown in the training process, assuming the target neural network model aims to transform M1 first reference signals and N1 second reference signals into P target reference signals, during model training, the neural network inputs (multiple sets of M1 sample first reference signals and N1 sample second reference signals at the sample level / frame level, and reference signal flags for each set of sample reference signals input in a fixed signal form) and the neural network outputs (P target output reference signals at the sample level / frame level) can be used as constraints. The training objective is to select the set of signals with the highest multicoherence between the P target output reference signals and the noise signal within the denoising region. This process continues until the target neural network model is obtained. When using the model, as follows... Figure 6As shown in the usage process, its input and output processing is similar to the training phase. After inputting M1 first reference signals and N1 second reference signals at the sample-level / frame-level, along with reference signal labels in a fixed signal format, into the target neural network model, P target reference signals can be obtained from the sample-level / frame-level fitted output of the target neural network model. During model training, the model parameters corresponding to each or several reference signal labels can be trained separately. This allows for the selection of appropriate model parameters based on the reference signal labels during use, making signal processing more targeted and improving the model's output performance. It should be noted that sample-level input processing refers to processing the reference signals at the point level, while frame-level input processing refers to processing the reference signals at the frame level.

[0084] In this embodiment, after the selected M1 first reference signals and N1 second reference signals are processed by the model, the output P target reference signals are then processed by the subsequent output signal generation module to generate inverted output signals, which are then sent to the speaker system. Since the inverted output signals have very high latency requirements, during pre-training and using the target neural network model, the model's input and output data can maintain a sample-level or frame-level data processing format to ensure low latency characteristics, thereby improving the overall effectiveness of active noise reduction. It should also be noted that the sample-level or frame-level input and output data processing in this embodiment, as described above... Figure 4 and Figure 5 and in subsequent embodiments Figure 7 and Figure 8 During model training / use, data can be maintained in either sample-level or frame-level format, which will not be explained in detail here.

[0085] In one embodiment, M1 sample first reference signals and N1 sample second reference signals are input into a neural network at a preset low sampling rate, and the neural network outputs P target output reference signals at the preset low sampling rate. Accordingly, in use, M1 first reference signals and N1 second reference signals are input into a target neural network model at a preset low sampling rate, and the target neural network model outputs P target reference signals at the preset low sampling rate.

[0086] In one embodiment, M1 sample first reference signals and N1 sample second reference signals are processed at a preset low sampling rate and then input into a neural network in the form of sample points or frames as signal input. The neural network outputs P target output reference signals in the form of sample points or frames as signal input with a preset low sampling rate. In use, correspondingly, M1 sample first reference signals and N1 sample second reference signals are processed at a preset low sampling rate and then input into a target neural network model in the form of sample points or frames as signal input. The target neural network model outputs P target reference signals in the form of sample points or frames as signal input with a preset low sampling rate.

[0087] As a training example, such as Figure 7 As shown, to further reduce system overhead, the training and use of the target neural network can also be performed after downsampling the reference signal, such as... Figure 7 The training process shown involves processing M1 samples of the first reference signal and N1 samples of the second reference signal at a preset low sampling rate (e.g., from 48kHz to 8kHz) at the sample-level / frame-level input. The training objective is to perform coherence statistical analysis on the low-sampling-rate model output, thereby training the neural network until the target neural network model is obtained. The target neural network model outputs P target output reference signals using sample-level / frame-level low-sampling-rate data (e.g., 8kHz). Figure 7 When using the model, the data processing is similar to the training process. In this way, the output signal generation module can directly generate the speaker system's output signal based on the P target reference signals output by the model. For example, it can directly use a filtering algorithm to process the signal at the same low sampling rate (e.g., 8kHz). In this case, there is no need to downsample the target reference signals before filtering. Of course, the filtering algorithm in the output signal generation module can also process the signal at an even lower sampling rate (e.g., 4kHz). In this case, the target reference signals need to be downsampled, and then the sampling rate of the output signal after filtering is increased back to the required sampling rate, such as 8kHz→48kHz or 4kHz→48kHz. It should be noted that the above-mentioned preset low sampling rate can be configured empirically; the above is only an example and is not specifically limited.

[0088] In this embodiment, in addition to sample-level / frame-level input processing during training, the input data is also processed at a preset low sampling rate, and the same preset low sampling rate is maintained for sample-level / frame-level output processing during output, further reducing system overhead. Furthermore, it allows subsequent filtering algorithms to process data at the same low sampling rate. Figure 7 The downsampling module in the output signal generation module is not needed, thus reducing the processing burden on the output signal generation module.

[0089] In one embodiment, the neural network is trained using M1 sample first reference signals, N1 sample second reference signals, and reference signal flags as model inputs, and the target output reference signal as model output, until a trained target neural network model is obtained. This includes: training the neural network using M1 sample first reference signals, N1 sample second reference signals, and reference signal flags as model inputs, and the target output reference signal and its magnitude P as model output, until a target neural network model is obtained. Further, the magnitude P is a fixed value, or the magnitude P is the training target during the training process.

[0090] It should be noted that, in the embodiments of the present application, the number of target output reference signals (i.e., the P value) output by the target neural network model is configurable, which is not specifically limited. For example, P satisfies P<M1+N1, so as to achieve the purpose of signal dimension reduction; in addition, as an example, the P value may be directly specified by a user, or specified according to available overhead requirements. In the processes of neural network training, use, filtering and the like, the P value is a fixed value determined in any of the foregoing manners. As another example, the P value may also be used as an optimization parameter in the model training process to maximize the balance between performance and overhead. It is easy to understand that the closer the P value is to M1+N1, the easier it is to obtain the output of the target neural network model with higher target coherence, but the larger the P value is, the higher the overhead required for the filtering process of the subsequent output signal generation module will be. In addition, an increase or decrease of 1 in the P value may have a great or small impact on the coherence processing of the model. Therefore, the use of an adjustable P value can achieve a better balance between implementation performance and overhead.

[0091] In an embodiment, the training objective includes: MIN(weight*(1-preset frequency band multi-coherence)+(1-weight)*P / calibration constant), where weight denotes the adjustment weight, the preset frequency band multi-coherence characterizes the coherence between the target output reference signal and the noise signal in the noise reduction region in the preset frequency band, MIN denotes the minimum value, and the training objective is to minimize (weight*(1-preset frequency band multi-coherence)+(1-weight)*P / calibration constant).

[0092] As Figure 8 shown, when training the neural network, the P value is also taken as one of the training optimization objectives to find the balance between the P value and the optimal coherence, and such balance is also taken as a model training objective until the target neural network model is obtained through training.

[0093] In an embodiment, the target neural network model further outputs the quantity value P of the target reference signals; generating an output signal according to the P target reference signals includes: generating the output signal according to the P target reference signals and the quantity value P.

[0094] In this embodiment, when the model is in use, the embodiment of the present application may also directly transmit the correct P value to the subsequent output signal generation module, so that the output signal generation module can quickly process the target reference signals based on the P value to generate the output signal, thereby improving the processing efficiency and accuracy.

[0095] As can be seen, in this embodiment, after the signal selection module effectively filters M first reference signals and N second reference signals to obtain M1 first reference signals and N1 second reference signals, the active noise reduction system further allows the fitting and mapping module to fit P target reference signals from the M1 first reference signals and N1 second reference signals using a trained target neural network model. These P target reference signals are then used in the subsequent output signal generation. On the one hand, this achieves signal dimensionality reduction to reduce computational power, and the training of the target neural network ensures that the multicoherence between the output target reference signals and the noise signals in the noise reduction region is sufficiently high, further guaranteeing the coherence between the fitted noise reference signals and the noise signals in the noise reduction region. On the other hand, it can also compensate for the nonlinearity between the output target reference signals and the noise signals in the noise reduction region, thereby completing the linear and nonlinear fitting between the finally selected reference signals and the noise signals in the noise reduction region. In other words, this embodiment can achieve the purpose of linear dimensionality reduction and nonlinear fitting of the input reference signals, thereby effectively improving the accuracy and efficiency of active noise reduction, ensuring that the finally selected target reference signals can fully guarantee the subsequent active noise reduction capabilities.

[0096] In this embodiment, the main purpose of the fitting mapping module is to extract signals (including nonlinear and linear) that are coherent with the noise signal in the noise reduction region from M1 first reference signals and N1 second reference signals based on the noise signal in the noise reduction region. Therefore, while reducing the overhead of dimensionality reduction (M1+M2→P), nonlinear fitting is also completed. In addition, the processing of the fitting mapping module is also independent of secondary paths (such as the response from the speaker to the noise reduction region). Therefore, it can better learn the relationship between M1 first reference signals, N1 second reference signals, and the noise signal in the noise reduction region without being affected by secondary paths, and complete dimensionality reduction, linear and nonlinear fitting. This makes it easier to obtain better noise reduction effect from the output target reference signal.

[0097] It should be understood that the embodiments of this application do not restrict the network type of the above-mentioned neural network. It can be any neural network that can achieve the above-mentioned training output effect, including but not limited to recurrent neural networks (RNN), gated recurrent units, long short-term memory networks (LSTM), and convolutional neural networks, etc., without specific limitations.

[0098] It should be noted that, in one embodiment, that is, in the process of fitting P target reference signals from M1 first reference signals and N1 second reference signals, other linear dimensionality reduction methods can also be used to achieve signal dimensionality reduction and reduce the input of the final noise reference signal. For example, according to the active noise reduction requirements, P reference signals can be selected from M1 first reference signals and N1 second reference signals as the final target reference signals. Alternatively, principal component analysis can be performed on M1 first reference signals and N1 second reference signals, and based on the principal component analysis results, the number of M1 first reference signals and N1 second reference signals can be reduced to P using the noise transfer matrix to obtain P target reference signals. The specific method is not limited.

[0099] It should be noted that, as Figure 1 As shown in this embodiment, P target reference signals are obtained and combined with the output signal generation module to generate the output signal of the speaker system, so that the vehicle speaker system generates an output signal of anti-phase sound waves to cancel the noise present in the noise reduction area. The active noise reduction algorithm includes, but is not limited to, various time-domain filtering algorithms or other algorithms that can process the selected noise reference signals to obtain the output signal; no specific limitation is made. Since the aforementioned signal selection process and / or fitting mapping process have already preprocessed the input reference signals, using active noise reduction algorithms, such as time-domain filtering algorithms, can more easily achieve good noise reduction results and has significant advantages in real-time processing.

[0100] In one embodiment, a neural network training method is also provided, characterized in that the method includes: S201. Obtain the first reference signal of M1 samples and the second reference signal of N1 samples.

[0101] S202. Using M1 sample first reference signals and N1 sample second reference signals as model inputs and P target output reference signals as model outputs, train the neural network until the trained target neural network model is obtained. As an example, the training objective includes ensuring that the multicoherence of the P target reference signals output by the neural network and the noise signal within the denoising region meets preset training requirements. As another example, the M1 sample first reference signals and N1 sample second reference signals can be selected from the M sample first reference signals and N sample second reference signals, respectively. It should be noted that for more details regarding this embodiment, please refer to the explanations and example embodiments of steps S101-S102 in the foregoing embodiments, such as... Figures 4-8The difference from the previous description is that in this embodiment, the reference signal flag may not be used as training data input to the model for training, nor will it be used as model input in subsequent model use, which can reduce training complexity and improve model applicability.

[0102] The following example primarily uses a selected target reference signal to illustrate how to process the target reference signal and a filtering algorithm to obtain an output signal. In one embodiment, specifically generating an output signal based on the target reference signal, the process includes: The output signal is generated by filtering P target reference signals according to a fixed filtering method, where the filtering parameters used in the fixed filtering method are the filtering parameters corresponding to the reference signal flags; or, the output signal is generated by filtering P target reference signals according to an adaptive filtering method.

[0103] On the one hand, such as Figure 10 As shown, a fixed filtering method based on fixed filtering parameters can be selected. The output signal generation module filters the P target reference signals selected in the final step according to the fixed filtering method to generate the output signal of the speaker system. Figure 1 As shown, in a specific implementation, for example, the output signal generation module downsamples the selected P target reference signals, then filters them using a fixed filtering method, and finally upsamples them to obtain the final output signal. Of course, if using... Figure 7 The target neural network model shown may also have an output signal generation module that does not require downsampling processing. Please refer to the explanations in the foregoing embodiments for details, which will not be elaborated here. It is worth noting that in this application embodiment, when selecting a fixed filtering method, the fixed filtering parameters used are filtering parameters corresponding to the reference signal flags of the M1 first reference signals and N1 second reference signals. That is, the fixed filtering parameters are set based on the previous filtering results of the M1 first reference signals and N1 second reference signals, and are stored in the system memory. During actual operation, the fixed filtering parameters for this scenario can be directly selected according to the corresponding reference signal flags to complete the filtering function. Since only fixed filtering parameters are used during actual operation, the computational cost can be greatly optimized.

[0104] Alternatively, an adaptive filtering method can be selected. This means the output signal generation module can filter the target reference signal using an adaptive filtering method to generate the output signal. For example, the adaptive parameter updates in the adaptive filtering method include, but are not limited to, FxLMS (Filtered-X Least Mean Square), FxNLMS (Filtered-X Normalized Least Mean Square), or self-adaptive update algorithms; no specific limitation is made. In this embodiment, using an adaptive filtering method allows the active noise cancellation system to better cope with unknown changing noise, further improving the active noise cancellation performance.

[0105] In one embodiment, the loudspeaker system includes a first loudspeaker group and a second loudspeaker group, which are different loudspeakers in a vehicle. The first loudspeaker group is a loudspeaker using a fixed filtering method, and the second loudspeaker group is a loudspeaker using an adaptive filtering method. Generating the output signal of the loudspeaker system based on target reference signals includes: filtering P target reference signals according to the fixed filtering method to generate the output signal of the first loudspeaker group, wherein the filtering parameters used in the fixed filtering method are the filtering parameters corresponding to the reference signal flags; and filtering P target reference signals according to the adaptive filtering method to generate the output signal of the second loudspeaker group.

[0106] In this embodiment, such as Figure 10 As shown, based on the filtering method with fixed filter parameters, it is possible to further implement adaptive filtering parameters for some vehicle speakers and fixed filtering parameters for others. Therefore, P target reference signals can be filtered using the fixed filtering method to generate the output signal of the first speaker group. The filtering parameters used in the fixed filtering method are the filtering parameters corresponding to the reference signal flags of M1 first reference signals and N1 second reference signals. Then, the P target reference signals can be filtered using the adaptive filtering method to generate the output signal of the second speaker group. This achieves adaptive filtering parameters for some vehicle speakers and fixed filtering parameters for others.

[0107] It should be noted that the details of the adaptive processing and fixed filtering methods in this embodiment can be found in the corresponding explanations of the foregoing embodiments. For example, the downsampling and upsampling processes will not be explained again here. It is worth mentioning that using fixed filtering parameters for some vehicle speakers and adaptive filtering parameters for others can achieve a better balance between increased overhead and improved noise reduction performance.

[0108] In this embodiment, by simultaneously employing both fixed and adaptive filtering methods, not only is the noise reduction effect better, but it is also applicable to a wider range of scenarios, with even superior noise reduction performance when combined with specific scenarios. For example, in a vehicle without headrest speakers, adaptive filtering can be used for the subwoofer, while fixed filtering parameters can be used for the door speakers. This allows the subwoofer to better track changes in low-frequency noise, improving low-frequency noise reduction. Conversely, in a vehicle with headrest speakers, fixed filtering parameters can be used for the subwoofer and door speakers, while adaptive filtering parameters are used for the headrest speakers. Because the headrest speakers are closer to the ears, adaptive processing achieves better mid-to-high frequency noise reduction, thereby increasing the overall noise reduction bandwidth and better handling of unknown noise variations, thus improving noise reduction performance.

[0109] In one embodiment, the adaptive parameter update of the adaptive filtering method mentioned in the above embodiments includes, but is not limited to, the FxNLMS algorithm or a self-adaptive update algorithm. Taking the FxNLMS algorithm as an example, as a case study, the FxNLMS algorithm includes the FxNLMS algorithm based on virtual mapping, without specific limitations. Taking this as an example, the processing procedure is as follows: Figure 11 As shown, its working process is not fully explained here. This virtual mapping-based FxNLMS algorithm uses the human ear as the optimization point, which can better achieve adaptive noise reduction of the noise reduction region location. Among them, SV (Speaker response function to virtual ear position), SP (Speaker response function to the corresponding error sensor of the noise reduction region), and PV (response function of the corresponding error sensor of the noise reduction region to the virtual ear position) involved in the algorithm can all be predicted and trained offline and saved for use in active noise reduction. The specific offline training process is not described in detail here.

[0110] As can be seen, due to the real-time processing requirements of the active noise cancellation system, the selected target reference signal needs to be downsampled, then subjected to time-domain filtering, and then upsampled to obtain the final output signal, which is then sent to the speaker system. Since the aforementioned signal selection and / or fitting mapping have already preprocessed the reference input signal, it is easier to achieve a good noise reduction effect for the subsequent filtering process. In addition, after obtaining the selected M1 first reference signals and N1 second reference signals, the M1 first reference signals and N1 second reference signals can also be directly combined with the output signal generation module to generate the output signal of the speaker system. For example, a filtering algorithm can be used to directly combine the M1 first reference signals and N1 second reference signals to generate the output signal. The filtering process is the same as the filtering process of the target reference signal in the above embodiment, and will not be described in detail here.

[0111] It should be noted that the foregoing embodiments all involve the noise reduction region. The embodiments of this application provide at least two methods to determine the noise reduction region, which can make the solution more flexible and intelligent.

[0112] In one embodiment, the method further includes: determining the noise reduction area before performing coherence analysis between the reference signal and the noise signal within the noise reduction area. Specifically, the noise reduction area can be determined after the user boards the vehicle, without limitation. Determining the noise reduction area includes: automatically determining it based on vehicle seating information, or determining it based on user-specified information. For example, vehicle seating information may include occupancy rate, seat information, and vehicle mode, without limitation. Specifically, as an optional method for determining the noise reduction area, the noise reduction area can be specified through user-specified information input by the user. For example, the user-specified information can be input by the user through the vehicle's infotainment system (UI) or a mobile terminal UI, allowing the user to specify the noise reduction area. Using this method can greatly improve the user's comfort and intelligence in using the vehicle. As another optional method for determining the noise reduction area, the noise reduction area can also be automatically determined based on vehicle seating information, which is generally used when the vehicle is in motion. The method of automatically determining the noise reduction area is more intelligent.

[0113] In one embodiment, the noise reduction area includes one or more of the areas where the head is placed on the seat and the middle area of ​​the rear row, which is typically used for camping; the specific area is not limited. For example, such as Figure 12 As shown in Example 1, when the vehicle is in motion, the user can specify or automatically determine the area in front of the headrests of the front seats. Similarly, when the vehicle is stationary, certain noise reduction areas can be specified for noise reduction. Figure 12 As shown in Example 2, when a seat is in a reclined (rest mode) state, the area above the headrest can be specified or automatically determined as the noise reduction zone, for example... Figure 12As shown in Example 3, when the rear seats are in normal operating mode (camping / rest mode, with people lying down), the middle area of ​​the rear seats can be designated or automatically determined as the noise reduction zone. It should be noted that... Figure 12 The examples in this application embodiment are merely illustrative and are not intended to limit the specific application.

[0114] By employing the aforementioned method for determining the noise reduction area, users can specify the noise reduction area or have it automatically determined based on vehicle seating information. This enables flexible matching of varying noise reduction areas, meeting different needs in different scenarios and improving adaptability and user experience. It should be noted that, in practical applications, as an example, this embodiment may also determine the conditions or state information that trigger the active noise cancellation system to start / stop, such as whether the car door is closed, etc., without further limitation.

[0115] It should be understood that the above embodiments have described the various embodiments of this application in detail. The following uses two practical application scenarios as examples to describe the working process of some embodiments of the corresponding application scenarios, which are described below.

[0116] In one application scenario embodiment, the method includes the following steps: 101. With the vehicle in motion, noise reduction is applied to the front area of ​​the headrests of the front seats based on the vehicle seating information. 102. It can be determined through the vehicle's CAN (Controller Area Network) information that the vehicle's windows and doors are closed, which is suitable for the activation of the active noise cancellation system and triggers active noise cancellation processing; 103. The active noise reduction system analyzes the road surface type of the vehicle and selects appropriate acceleration sensors and vibration sensors to obtain N second reference signals. If an engine / generator is present, the active noise reduction system will select the environmental noise collected by the acceleration sensor / vibration sensor that can characterize its noise and the environmental noise collected by the acceleration sensor / vibration sensor that can characterize the vibration noise caused by wind noise when the vehicle is driving. 104. The active noise cancellation system analyzes the reference microphone signal and selects the first reference signal that is unaffected by wind noise and can characterize M environmental noises. 105. The active noise cancellation system performs multiple coherence analysis on the selected M first reference signals and N second reference signals, as well as the noise signal in the noise cancellation area collected by the microphone in the front area of ​​the headrest, and selects the M1 first reference signals and N1 second reference signals with the best coherence. 106. The above M1 first reference signals and N1 second reference signals, as well as the reference signal flags, are fed to the pre-trained target neural network model for processing to achieve dimensionality reduction and nonlinear fitting of the reference signals, and output P target reference signals. 107. After downsampling and filtering of P target reference signals, upsampling is performed to generate the final output signal, which is then sent to the speaker system. The speaker system plays the output signal used for active noise reduction to cancel out the noise in the front area of ​​the headrest, thereby achieving the purpose of active noise reduction in the front area of ​​the headrest.

[0117] As can be seen, in this embodiment, the optimal reference signal can be selected for noise generated by different road surfaces to achieve better road noise / tire noise reduction; environmental noise reduction; different reference signals can achieve noise reduction of various types of noise, as well as dimensionality reduction and nonlinear fitting, which can improve the signal processing efficiency of the active noise reduction system, while ensuring the noise reduction effect and improving the nonlinear noise reduction capability.

[0118] In one application scenario embodiment, the method includes the following steps: 201. When the vehicle is stationary, such as when the user sets the vehicle to enter nap mode / camping mode, or specifies active noise cancellation above the headrest of the driver's seat through the UI, the vehicle is stationary and therefore there is no need to reduce road noise or vibration noise caused by wind during driving. 202. The vehicle's windows and doors are closed as determined by the vehicle's CAN information, triggering the active noise cancellation system to perform active noise cancellation processing. 203. If the vehicle generator is working, the active noise cancellation system will collect N second reference signals from the second preset distance between the generator and the acceleration sensor / vibration sensor / microphone. 204. Select M first reference signals that are unaffected by wind noise; 205. Analyze the N second reference signals and M first reference signals selected above, and the multicoherence of the noise signals in the noise reduction area collected by the microphone in the area above the headrest, and select the best M1 first reference signals and N1 second reference signals. 206. Input M1 first reference signals, N1 second reference signals, and reference signal flags to the pre-trained target neural network model for processing, to achieve dimensionality reduction and nonlinear fitting of the reference signals, so as to output P target reference signals; 207. After downsampling and filtering of P target reference signals, upsampling is performed to generate the final output signal, which is then sent to the speaker system. The speaker system plays the output signal used for active noise reduction to cancel the noise in the area above the headrest, thereby achieving the purpose of active noise reduction in the area above the headrest.

[0119] In this embodiment, active noise reduction (such as in-vehicle generator noise and external environmental noise) is implemented for vehicles in a stopped state to improve vehicle comfort in rest mode; signal dimensionality reduction and nonlinear fitting are achieved to improve the signal processing efficiency of the active noise reduction system while ensuring the noise reduction effect and enhancing nonlinear noise reduction capability.

[0120] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0121] The above mainly describes the method provided in the embodiments of this application. The following describes the related system and medium embodiments provided in the embodiments of this application.

[0122] In one embodiment, such as Figure 1 As shown, an active noise cancellation system is provided, the active noise cancellation system includes: The signal selection module is used to acquire reference signals, which include M first reference signals and N second reference signals. Through coherence analysis between the reference signals and the noise signals in the noise reduction area, M1 first reference signals and N1 second reference signals are selected. The noise signals in the noise reduction area are acquired by a microphone at a first preset distance from the human ear. The output signal generation module is used to generate an output signal based on M1 first reference signals and N1 second reference signals, so that the speaker system plays the output signal to actively reduce noise in the noise reduction area, and M1+N1 is less than M+N.

[0123] In one embodiment, the output signal generation module includes: The fitting mapping module is used to input M1 first reference signals and N1 second reference signals, or M1 first reference signals, N1 second reference signals and reference signal flags, into the trained target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and N1 second reference signals, where P is less than or equal to M1+N1, and the reference signal flags are bound to the M1 first reference signals and N1 second reference signals. The filtering module is used to filter P target reference signals to generate an output signal.

[0124] As can be seen, this embodiment provides an active noise cancellation system that considers the noise signal within the noise cancellation area. Therefore, for the noise in the noise cancellation area, a reference signal related to the noise source in the noise cancellation area is selected from a large number of reference signals as the output signal of the speaker system. This reduces the signal input processing load of the active noise cancellation system, effectively reduces the computational power requirement for output signal generation, avoids directly using a large number of collected noise reference signals in the calculation of the output signal, and simplifies the calculation process, reducing computational complexity and resource waste, reducing processing difficulty, and also helps improve the real-time noise cancellation performance of the active noise cancellation system.

[0125] In one embodiment, an active noise cancellation system is provided, the active noise cancellation system including an output signal generation module, the output signal generation module including: The fitting and mapping module is used to acquire M1 first reference signals and N1 second reference signals. The first reference signals are environmental noise collected by the microphone, and the second reference signals are environmental noise collected by at least one of the accelerometer and vibration sensor. The M1 first reference signals and N1 second reference signals are input into the trained target neural network model, so that the target neural network model outputs P target reference signals based on the M1 first reference signals and N1 second reference signals, where P is less than or equal to the sum of M1 and N1. The filtering module is used to filter P target reference signals to generate an output signal, which is then played by the speaker system to actively reduce noise in the noise reduction area.

[0126] In one embodiment, in conjunction with the above embodiments, the active noise cancellation system further includes: The signal selection module is used to perform multiple coherence analysis on the noise reference signals of M first reference signals and N second reference signals with the noise signals of the target noise reduction area, so as to select noise reference signals with coherence requirements of preset frequency bands from the M first reference signals and N second reference signals, and obtain M1 first reference signals and N1 second reference signals.

[0127] It should be noted that further details regarding the active noise cancellation system, including implementation details and technical effects, can be found in the description of the aforementioned method embodiments, and will not be repeated here.

[0128] In one embodiment, a vehicle is provided, the vehicle including an active noise cancellation system, the active noise cancellation system being used to implement the active noise cancellation method as described in any of the foregoing embodiments; or is an active noise cancellation system as described in any of the foregoing embodiments.

[0129] In one embodiment, a readable storage medium is provided, which stores a computer program, characterized in that the computer program, when executed by a processor, implements the active noise reduction method as described in any of the foregoing embodiments; or performs the function of the active noise reduction system as described in any of the foregoing embodiments.

[0130] In one embodiment, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the method of any of the foregoing embodiments.

[0131] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the active noise reduction method as described in any of the foregoing embodiments; or the function of the active noise reduction system as described in any of the foregoing embodiments.

[0132] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory.

[0133] Those skilled in the art will clearly understand that the above-described division of functional units and modules is merely an example. In practical applications, the functions described above can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. Those skilled in the art should understand that modifications to the technical solutions described in the foregoing embodiments, or equivalent substitutions for some of the technical features, should be included within the scope of protection of this application.

Claims

1. An active noise reduction method, characterized in that, The method includes: Acquire reference signals, which include M first reference signals and N second reference signals. The first reference signals are ambient noise collected by the microphone, and the second reference signals are ambient noise collected by at least one of the accelerometer and the vibration sensor. The sum of M and N is greater than or equal to 2. Through coherence analysis of the reference signal and the noise signal in the noise reduction area, M1 first reference signals and N1 second reference signals are selected. The noise signal in the noise reduction area is collected by a microphone at a first preset distance from the human ear. An output signal is generated based on the M1 first reference signals and the N1 second reference signals, so that the speaker system plays the output signal to actively reduce noise in the noise reduction area. The sum of M1 and N1 is greater than or equal to 1, and the sum of M1 and N1 is less than the sum of M and N.

2. The active noise reduction method according to claim 1, characterized in that, The M first reference signals are environmental noise collected by the microphone that is unaffected by wind noise; Obtaining M of the first reference signals includes: Wind noise is identified from the Q ambient noises collected by the microphone. Based on the wind noise identification results, M first reference signals are obtained from the Q ambient noises collected by the microphone, where Q is greater than M. The wind noise of the M first reference signals is less than the wind noise of the (QM) first reference signals.

3. The active noise reduction method according to claim 1, characterized in that, Obtaining the N second reference signals includes: obtaining the N second reference signals based on vehicle status information; Obtaining the N second reference signals based on vehicle status information includes: When the vehicle is in motion, acquire the ambient noise collected by the acceleration sensor and / or vibration sensor corresponding to the type of road surface the vehicle is traveling on; When the vehicle is stationary, if the vehicle generator is in operation, the environmental noise collected by the acceleration sensor and / or vibration sensor at a second preset distance from the vehicle generator is acquired. When the vehicle is stationary, if the vehicle generator and the vehicle engine are not in operation, N equals 0.

4. The active noise reduction method according to claim 3, characterized in that, Determining the acceleration sensor and / or vibration sensor corresponding to the road surface type on which the vehicle is traveling includes: From the pre-stored correspondence, the acceleration sensor and / or vibration sensor corresponding to the vehicle driving road surface type are retrieved. The correspondence includes the correspondence between different vehicle driving road surface types and their corresponding acceleration sensors and / or vibration sensors.

5. The active noise reduction method according to claim 4, characterized in that, The vehicle driving road surface types include a first road surface, a third road surface, and a third road surface. The road surface roughness of the first road surface is less than that of the second road surface, and the road surface roughness of the second road surface is less than that of the third road surface. The acceleration sensor and / or vibration sensor corresponding to the first road surface includes acceleration sensors installed at the tire suspension, vehicle chassis and connecting bridge positions, and / or vibration sensors installed at the tire suspension, vehicle chassis and connecting bridge positions. The acceleration sensor and / or vibration sensor corresponding to the second road surface type includes an acceleration sensor installed on the tire suspension, support arm, vehicle chassis and vehicle tower top, and / or a vibration sensor installed on the tire suspension, support arm, vehicle chassis and vehicle tower top. The acceleration sensor and / or vibration sensor corresponding to the third road surface includes an acceleration sensor installed on the tire suspension, vehicle chassis, vehicle chassis crossbeam and vehicle tower top, and / or a vibration sensor installed on the tire suspension, vehicle chassis, vehicle chassis crossbeam and vehicle tower top.

6. The active noise reduction method according to claim 1, characterized in that, The M first reference signals include ambient noise collected by one or more microphones closest to the noise reduction area; The N second reference signals include at least one of the following: The ambient noise collected by one or more accelerometers closest to the noise reduction area; The ambient noise collected by one or more vibration sensors closest to the noise reduction area.

7. The active noise reduction method according to claim 1, characterized in that, The step of selecting M1 first reference signals and N1 second reference signals through coherence analysis of the reference signals and noise signals within the noise reduction region includes: Determine the coherence between each reference signal in the reference signal and the noise signal in the noise reduction region within a preset frequency band; The coherence of the M1 first reference signals in the preset frequency band and the noise signal in the noise reduction area is greater than the coherence of the (M-M1) first reference signals in the preset frequency band and the noise reduction area. The coherence of the N1 second reference signals in the preset frequency band and the noise signal in the noise reduction region is greater than the coherence of the (N-N1) second reference signals in the preset frequency band and the noise reduction region.

8. The active noise reduction method according to claim 1, characterized in that, The acquisition of the reference signal includes: One by one, candidate reference signals are selected and the coherence between the candidate reference signals and the noise signals in the noise reduction area is calculated. The candidate reference signal with the highest coherence is selected as the selected reference signal, until the number of first reference signals among the selected reference signals reaches M and the number of second reference signals reaches N.

9. The active noise reduction method according to claim 1, characterized in that, Also includes: Determine the noise reduction area; Determining the noise reduction area includes: automatically determining it based on vehicle seating information, or determining it based on user-specified information; The noise reduction zone includes one or more of the areas where the head is placed on the seats and the middle area of ​​the rear seats in the usual camping space.

10. The active noise reduction method according to claim 1, characterized in that, The microphone located at a first preset distance from the human ear is one or more microphones that are closest to the human ear; The noise signal within the noise reduction area is obtained through a microphone located at a first preset distance from the human ear, including: The noise signal collected by the microphone at the headrest of the seat is the noise signal within the noise reduction area; or, When no microphone signal is set at the headrest of the seat, the signal collected by one or more microphones closest to the ear is processed by the path transfer function to obtain the noise signal in the noise reduction area. The path transfer function represents the signal attenuation of the signal collected by the one or more microphones closest to the ear when it is transmitted to the noise reduction area.

11. The active noise reduction method according to any one of claims 1-10, characterized in that, The step of generating an output signal based on the M1 first reference signals and the N1 second reference signals includes: The M1 first reference signals and the N1 second reference signals, or the M1 first reference signals, the N1 second reference signals, and reference signal flags, are input into the trained target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and the N1 second reference signals, where P is less than or equal to the sum of M1 and N1, and the reference signal flags are bound to the M1 first reference signals and the N1 second reference signals. The output signal is generated based on the P target reference signals.

12. The active noise reduction method according to claim 11, characterized in that, The step of generating the output signal based on the P target reference signals includes: The P target reference signals are filtered according to a fixed filtering method to generate an output signal. The filtering parameters used in the fixed filtering method are the filtering parameters corresponding to the reference signal flags.

13. The active noise reduction method according to claim 11, characterized in that, The loudspeaker system includes a first loudspeaker group and a second loudspeaker group. Generating the output signal based on the P target reference signals includes: The P target reference signals are filtered according to a fixed filtering method to generate the output signal of the first loudspeaker group. The filtering parameters used in the fixed filtering method are the filtering parameters corresponding to the reference signal flags. The P target reference signals are filtered using an adaptive filtering method to generate the output signal of the second speaker group.

14. An active noise reduction system, characterized in that, The active noise cancellation system includes: A signal selection module is used to acquire reference signals, which include M first reference signals and N second reference signals. The first reference signals are ambient noise collected by a microphone, and the second reference signals are ambient noise collected by at least one of an accelerometer and a vibration sensor. Through coherence analysis of the reference signals and the noise signals in the noise reduction area, M1 first reference signals and N1 second reference signals are selected. The noise signals in the noise reduction area are acquired by a microphone at a first preset distance from the human ear. The output signal generation module is used to generate an output signal based on the M1 first reference signals and the N1 second reference signals, so that the speaker system plays the output signal to actively reduce noise in the noise reduction area, where M1+N1 is less than M+N.

15. The active noise cancellation system according to claim 14, characterized in that, The output signal generation module includes: The fitting mapping module is used to input the M1 first reference signals and the N1 second reference signals into the target neural network model, so that the target neural network model fits P target reference signals from the M1 first reference signals and the N1 second reference signals, where P is less than or equal to M1+N1; The filtering module is used to perform filtering processing on the P target reference signals to generate an output signal.

16. The active noise cancellation system according to claim 15, characterized in that, The The fitting and mapping module also inputs the M1 first reference signals, the N1 second reference signals, and the reference signal flags into the target neural network, so that the target neural network model fits P target reference signals from the M1 first reference signals and the N1 second reference signals; P is less than or equal to M1+N1, and the reference signal flags are bound to the M1 first reference signals and the N1 second reference signals.

17. A vehicle, characterized in that, The vehicle includes an active noise cancellation system, which is used to implement the active noise cancellation method as described in any one of claims 1-13; or is an active noise cancellation system as described in claims 14-16.

18. A readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 13, or implements the function of the active noise cancellation system as described in claims 14 to 16.

Images