A vehicle-mounted multi-zone sound system based on DNN noise reduction technology

Abstract

The present invention provides a vehicle-mounted multi-sound zone system based on DNN noise reduction technology, and relates to the technical field of vehicle-mounted multi-sound zone systems. The new system comprises: obtaining initial audio in a shielding area of ​​a target vehicle, and using a preset DNN noise reduction model to perform noise reduction on the initial audio to obtain an initial user voice; extracting each keyword in the initial user voice, and then determining whether the initial user voice is a target user voice according to each keyword in the initial user voice and the seat in the vehicle where the user corresponding to the initial user voice sits, and an image of the area corresponding to the seat, and in combination with the number of users in the rear area; playing the target user voice through an audio device arranged in the main driving area; the present invention can, without turning off the shielding function, specifically play the voice spoken by the user to the driver through the audio device to the driver, thereby simplifying the process of user voice communication and improving the efficiency of voice communication between the driver and passengers in the vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle-mounted multi-zone sound systems, and in particular to a vehicle-mounted multi-zone sound system based on DNN noise reduction technology. Background Art

[0002] With the rapid development of automobile technology, more and more cars are optimized for the sound zones within the car to enhance the driver and passengers' sense of audio experience within the car. A common method is to shield different areas within the car through reverse sound waves so that the sounds of different areas do not interfere with each other. For example, the rear passenger area or the co-pilot area is shielded through reverse sound waves so that the sounds in the rear passenger area and the co-pilot area cannot be transmitted to the driving area, thereby preventing the driver from being disturbed by the sounds made by people in the rear seat or co-pilot area. However, in some cases, people in the shielded area need to communicate with the driver. When the sound shielding function is turned on, if the user in the shielded area sends a voice to communicate with the driver, the driver needs to turn off the shielding function first and then send a voice to the driver if he wants to hear the voice. This process is relatively complicated, resulting in low efficiency in voice communication between the driver and passengers in the car. Summary of the invention

[0003] The invention provides a vehicle-mounted multi-zone system based on DNN noise reduction technology to solve the technical problem in the prior art that when a user in the shielded area of ​​the vehicle sends a voice to communicate with the driver, the driver needs to turn off the shielding function first and then send the voice to the driver if he wants to hear the voice. This process is relatively complicated, resulting in low efficiency of voice communication between the driver and passengers in the car.

[0004] According to the window intercom system based on DNN noise reduction technology provided by the present application, the system includes: a processor and a storage medium; the storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the following steps:

[0005] S100, obtaining initial audio QR in a shielded area of ​​a target vehicle; wherein the shielded area includes a passenger area and a rear area; and QR includes a voice uttered by a user.

[0006] S200, inputting QR into a preset DNN noise reduction model to obtain the noise-reduced initial user speech QT corresponding to QR.

[0007] S300, obtaining each keyword corresponding to QT and performing deduplication processing to obtain a keyword list A corresponding to QT = (A 1 , A 2 , …, A i , …, A n ), i=1, 2,...,n; among them, A iis the i-th keyword corresponding to QT, and n is the number of keywords corresponding to QT.

[0008] S400: If the user corresponding to QT is in the passenger seat area, determine whether QT is the target user voice according to the passenger seat area image and A.

[0009] S500: If the user corresponding to QT is located in the back row area and the number of back row users is 1, determine whether QT is the target user voice based on the back row area image and A.

[0010] S600: If the user corresponding to QT is in the back row area and the number of back row users is greater than 1, the audio to be compared TE with a preset time length is set after QR.

[0011] S700, obtaining the similarity η between QR and TE.

[0012] S800: Determine whether QT is the target user's voice based on η, A and the rear area image.

[0013] S900: If QT is the target user voice, QT is played through an audio device set in the main driving area.

[0014] The present invention has at least the following beneficial effects:

[0015] The vehicle-mounted multi-zone system based on DNN noise reduction technology of the present invention obtains the initial audio in the shielding area of ​​the target vehicle, and uses a preset DNN noise reduction model to reduce the noise of the initial audio to obtain the initial user voice; extracts each keyword in the initial user voice, and then determines whether the initial user voice is the target user voice based on each keyword in the initial user voice and the seat in the car where the user corresponding to the initial user voice sits, and the image of the area corresponding to the seat, and combined with the number of users in the rear area; if it is the target user voice, the target user voice is played through an audio device set in the main driving area, so that without turning off the shielding function, the voice spoken by the user to the driver is played to the driver through the audio device in a targeted manner, thereby simplifying the user voice communication process and improving the efficiency of voice communication between drivers and passengers in the car. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1A flowchart of the steps executed by a processor in a vehicle-mounted multi-zone audio system based on DNN noise reduction technology provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by technicians in the relevant technical field without creative work are within the scope of protection of the present application.

[0019] It is noteworthy that in the following description, specific details such as specific system structures, technologies, etc. are proposed for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from hindering the description of the present application.

[0020] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or their collections. It should also be understood that when the term "and / or" is used in the present specification and the appended claims, it refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0021] In the description of the present specification and the appended claims, if the term "if" is used, it can be interpreted as "in the case of" or "when" or "uponce" or "in response to determining" or "in response to detecting" according to the context. Similarly, the phrase "if it is determined" or "if [the described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [the described condition or event] is detected" or "in response to detecting [the described condition or event]" according to the context.

[0022] In the description of this application specification and the appended claims, the terms "first", "second", "third", etc., if used, are only used to distinguish the description and are not to be understood as indicating or implying relative importance, nor are they used to describe a specific order or sequence.

[0023] The following will refer to Figure 1 The flowchart of the steps executed by the processor in the vehicle-mounted multi-zone sound system based on DNN noise reduction technology is shown, and an vehicle-mounted multi-zone sound system based on DNN noise reduction technology is introduced.

[0024] The vehicle-mounted multi-zone audio system based on DNN noise reduction technology includes: a processor and a storage medium; the storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the following steps:

[0025] S100, obtaining initial audio QR in a shielded area of ​​a target vehicle; wherein the shielded area includes a passenger area and a rear area; and QR includes a voice uttered by a user.

[0026] The scenario in which the system in this embodiment is applied is a commercial vehicle scenario, for example, a taxi scenario. Under normal circumstances, if the users in the shielded area have no problems, they will not have too much conversation with the driver. The target vehicle can be any vehicle with a regional sound shielding function and the regional sound shielding function is turned on. It should be noted that a radio device is installed in each shielded area of ​​the vehicle, which can obtain the sound in the corresponding shielded area in real time. Users in the shielded area will not obtain QR when they are not speaking.

[0027] Specifically, QR can be obtained by the following steps:

[0028] S110, obtaining the volume of each preset time point in the preset sliding time window TA, so as to obtain a collection point volume list B corresponding to TA = (B 1 , B 2 , …, B j , …, B m ), j = 1, 2, …, m; where B j is the volume at the jth preset time point in TA, m is the number of preset time points in TA; the time interval between two adjacent preset time points is the same.

[0029] In this embodiment, the duration of the preset sliding time window can be set according to actual needs, for example: the duration of TA is 1S; a preset time point can be set at a certain interval within TA, for example: a preset time point is set every 50ms; the step size of the sliding time window TA can also be set according to actual needs, for example: the step size of TA is 50ms.

[0030] S120, obtain the average volume corresponding to B YB = (1 / m) × ∑ m j=1 B j And the average volume YB' corresponding to the volume at each preset time point in the previous adjacent sliding time window of TA.

[0031] In this embodiment, the volume in the corresponding shielded area can be collected at each preset time point in B; it can be understood that if the user in the shielded area does not speak, the volume corresponding to each preset time point in B is in a relatively balanced state.

[0032] S130, if YB-YB'≥ΔY, then sequentially obtain the average volume corresponding to each sliding time window in the q sliding time windows after TA to obtain an average volume list C=(C 1 , C 2 , …, C p , …, C q ), p = 1, 2, …, q; where C p is the average volume corresponding to the pth sliding time window among the q sliding time windows after TA; ΔY is the preset average volume difference threshold.

[0033] In this embodiment, if YB-YB'≥ΔY, it means that the average volume corresponding to the current sliding time window has changed significantly, and it is possible that the user in the shielded area is talking; however, this situation may also be caused by sudden noise; therefore, continue to obtain the average volume corresponding to each sliding time window in the q sliding time windows after TA in turn to make further judgments; ΔY is an empirical value, which can be obtained based on a large number of experiments.

[0034] S140, obtaining the difference between each average volume in C and YB', so as to obtain a list of average volume differences corresponding to C ΔC = (ΔC 1 , ΔC 2 ,…,ΔC p ,…,ΔC q ); where ΔC p C p Difference from YB'; ΔC p =C p -YB'.

[0035] It can be understood that a number of volumes are collected in each sliding time window in C, and the average volume corresponding to each sliding time window in C can be obtained.

[0036] S150, traverse ΔC, if ΔC1 to ΔCp-1 are all greater than or equal to ΔY, and ΔC p <ΔY, then the starting time point of TA and C p The audio between the end time points is determined as QR.

[0037] In this embodiment, if ΔC1 to ΔCp-1 are all greater than or equal to ΔY, it means that C 1 to C p-1 The user keeps talking in the sliding time window of ΔC p<ΔY, indicating that the user has finished speaking; therefore, the starting time point of TA and C p The audio between the start time point of TA and the end time point of C is determined as QR; it can be understood that, since the judgment is based on the average volume in the sliding time window, the start time point of TA and the end time point of C are p The audio between the end time points is longer than the actual length of the user's voice, thereby ensuring that the user's complete voice is obtained and ensuring the integrity of the QR.

[0038] Furthermore, we can also determine the starting time of TA and C p If the time between the start time of TA and the end time of C is longer than the preset maximum speaking time, p The audio between the end time points is determined as noise.

[0039] It is understandable that the user speaks one sentence at a time, and there will be a pause between two adjacent sentences; therefore, if the recognized QR is too long, it means that the QR is continuous external noise.

[0040] S200, inputting QR into a preset DNN noise reduction model to obtain the noise-reduced initial user speech QT corresponding to QR.

[0041] In this embodiment, it can be understood that QR is obtained by the sound receiving device and contains noise; QR can be input into a preset DNN noise reduction model to remove the noise; the preset DNN noise reduction model is a trained DNN noise reduction model, which can remove the noise in QR and only retain the human voice part; it should be noted that those skilled in the art can use the existing DNN noise reduction model training method to obtain the preset DNN noise reduction model according to actual needs, which will not be elaborated here.

[0042] S300, obtaining each keyword corresponding to QT and performing deduplication processing to obtain a keyword list A corresponding to QT = (A 1 , A 2 , …, A i , …, A n ), i=1, 2,...,n; among them, A i is the i-th keyword corresponding to QT, and n is the number of keywords corresponding to QT.

[0043] In this embodiment, QT can be converted into a text form, and then the keywords in the text corresponding to QT can be extracted; keywords are words other than some auxiliary words; it should be noted that those skilled in the art can use the existing speech-to-text and keyword extraction methods according to actual needs, obtain each keyword corresponding to QT and perform deduplication processing to obtain a keyword list A corresponding to QT, which will not be elaborated here.

[0044] S400: If the user corresponding to QT is in the passenger seat area, determine whether QT is the target user voice according to the passenger seat area image and A.

[0045] In this embodiment, if the user corresponding to QT is in the passenger seat area, and the passenger seat area can only seat one person, the user in the passenger seat area may be talking to the driver or to an electronic device, such as a mobile phone or headphones. Then, it is necessary to determine whether QT is the target user voice based on the image of the passenger seat area and A. The target user voice is the voice of the user speaking to the driver.

[0046] Further, step S400 may include the following steps:

[0047] S410, obtaining a first preset value W=1 and a second preset value NUM=0.

[0048] S420, if W<n, then determine A W Whether it is in the preset target keyword library; enter S430.

[0049] In this embodiment, the preset target keyword library contains a number of keywords corresponding to the scene of the passenger and the driver talking, such as: hello, master, how long, how far, etc.; A can be determined by traversing W Whether it is in the preset target keyword library.

[0050] The preset target keyword library can be obtained in the following way: obtaining texts of several conversations between passengers and drivers, then extracting keywords from the texts, and performing deduplication processing to obtain the preset target keyword library.

[0051] S430, if A W If the target keyword library is in the preset target keyword library, NUM is updated to NUM+1; otherwise, NUM is not updated; W is updated to W+1; and the process goes to S420.

[0052] S440: If W=n, then according to NUM and n, determine QT as the weight γ=NUM / n of the target user's voice.

[0053] Through the above steps, the proportion of keywords in A that exist in the preset target keyword library to the total keywords in A can be determined, that is, γ. The larger γ is, the more likely QT is the user talking to the driver, and the smaller γ is, the less likely QT is the user talking to the driver.

[0054] S450: Determine whether QT is the target user voice according to γ ​​and the passenger area image.

[0055] Further, step S450 may include the following steps:

[0056] S451, if γ≥γ', obtain s images of the passenger area taken in the time period corresponding to QR to obtain a list of images of the passenger area corresponding to QR = (D 1 , D 2 , …, D r , …, D s ), r = 1, 2, …, s; where D r is the rth passenger area image taken in the time period corresponding to QR; otherwise, it is determined that QT is not the target user voice; γ' is the preset weight threshold.

[0057] In this embodiment, if γ≥γ', it means that QT is very likely to be the voice corresponding to the user speaking to the driver; however, the weight of the voice spoken by the user when making a phone call corresponding to the target user's voice may also be relatively large; therefore, in order to more accurately determine the target user's voice, s images of the passenger area taken during the time period corresponding to QR are obtained, and combined with the behavior of the passenger user during the time period, it is used to assist in determining whether QT is the target user's voice. γ' can be set according to actual needs, or it can be an empirical value, for example: γ'=0.7.

[0058] S452, obtaining the direction angle corresponding to the direction of the front passenger user's face in each front passenger area image in D, so as to obtain a first direction angle list α corresponding to D = (α 1 , α 2 , …, α r , …, α s ); where α r D r The corresponding direction angle.

[0059] In this embodiment, the image of the co-pilot area is captured by a binocular camera. The image of the co-pilot area contains three-dimensional information, and can obtain the smallest cube corresponding to the head of the co-pilot user. Then, auxiliary lines are set perpendicular to the surface corresponding to the user's face. The direction angle corresponding to the auxiliary line is the direction angle corresponding to the direction of the co-pilot user's face. The direction angle of the front of the car can be set to 0°, the direction angle offset to the left is positive, and the direction angle offset to the right is negative.

[0060] S453, obtaining the number NA of direction angles in α that are within a first preset direction angle range.

[0061] In this embodiment, if the user in the passenger seat area is talking to the driver, the passenger seat user's face will inevitably be directed toward the driver's position multiple times, that is, the corresponding direction angle is within a certain angle range, for example: the first preset direction angle range is 60° to 90°.

[0062] S454, if NA / s>β, determine QT as the target user voice; otherwise, determine QT is not the target user voice; wherein β is a preset direction angle quantity weight threshold.

[0063] In this embodiment, if NA / s>β, it means that the user is very likely to talk with the driver, and γ≥γ' has been determined in the above steps, so QT can be determined as the target user voice; and if NA / s≤β, it means that the co-pilot user is talking to the electronic device, and the co-pilot area image can be used to further determine whether the co-pilot user is holding or wearing the electronic device to further improve the accuracy of the judgment; β can be set according to actual needs, for example: β=0.6.

[0064] S500: If the user corresponding to QT is located in the back row area and the number of back row users is 1, determine whether QT is the target user voice based on the back row area image and A.

[0065] In this embodiment, it can be understood that the rear area can accommodate at least one passenger, and may be two; if the user corresponding to QT is located in the rear area, and the number of rear users is 1, then based on the rear area image and A, it is determined whether QT is the target user voice.

[0066] Further, step S500 may include the following steps:

[0067] S510, obtaining QT as the weight γ of the target user's speech.

[0068] In this embodiment, in the back row area scenario, the method for determining QT as the weight γ of the target user's voice is the same as the method in steps S410-S440 in the above embodiment, and is not repeated here.

[0069] S520: If γ≥γ', then obtain y rear area images taken in the time period corresponding to QR to obtain a rear area image list E corresponding to QR = (E 1 , E 2 , …, E x , …, E y ), x=1, 2, …, y; where E x is the xth rear area image taken in the time period corresponding to QR; otherwise, it is determined that QT is not the target user voice; γ' is the preset weight threshold.

[0070] S530, obtaining the direction angle corresponding to the face direction of the rear row user in each rear row area image in E, so as to obtain a second direction angle list θ corresponding to E=(θ 1 ,θ 2 ,…,θ x ,…,θ y ); where θ xFor E x The corresponding direction angle.

[0071] S540: If the rear-seat user is located behind the front passenger seat, the number ND of direction angles in θ that are within a second preset direction angle range is obtained.

[0072] In this embodiment, if the rear-seat user is located behind the front passenger seat, the direction angle corresponding to the orientation of the user's face when talking to the driver will be within the second preset direction angle range; it can be understood that since the user is sitting behind the front passenger seat, the direction angle corresponding to the orientation of the user's face when talking to the driver is smaller than the direction angle corresponding to the front passenger seat user; for example: the second preset direction angle range is 40° to 60°.

[0073] S550: If the rear-seat user is located behind the driver, the number NE of direction angles in θ that are within a third preset direction angle range is obtained.

[0074] In this embodiment, if the rear-seat user is located behind the front passenger seat, the direction angle corresponding to the orientation of the user's face when talking to the driver will be within the third preset direction angle range; it can be understood that since the user is sitting behind the main driver's seat, when talking to the driver, the direction angle corresponding to the orientation of the user's face is almost parallel to the orientation of the front of the car; for example: the third preset direction angle range is 0° to 40°.

[0075] S560, if ND / y>β or NE / y>β, determine that QT is the target user voice; otherwise, determine that QT is not the target user voice; wherein β is a preset direction angle quantity weight threshold.

[0076] In this embodiment, if the rear user is located behind the front passenger seat and ND / y>β, QT is determined as the target user voice; if the rear user is located behind the front passenger seat and NE / y>β, QT is determined as the target user voice.

[0077] S600: If the user corresponding to QT is in the back row area and the number of back row users is greater than 1, then obtain the audio TE to be compared with a preset duration after QR.

[0078] In this embodiment, if the user corresponding to QT is in the back row area and the number of back row users is greater than 1, then QT may be the voice talking to other users in the back row area. Therefore, the audio TE to be compared with a preset length after QR is obtained to further determine the object of QT's conversation.

[0079] S700, obtaining the similarity η between QR and TE.

[0080] Further, step S700 may include the following steps:

[0081] S710, obtaining the speech feature vector XL corresponding to QR 1 The speech feature vector XL corresponding to TE2 .

[0082] In this embodiment, the speech feature vector can be generated according to the Mel-frequency cepstral coefficient, LPC coefficient, short-time energy parameter, short-time autocorrelation coefficient, etc. of the speech; it should be noted that those skilled in the art can use the existing speech feature vector generation method according to actual needs to obtain the speech feature vector XL corresponding to QR 1 and TE's speech feature vector XL 2 , which will not be elaborated here.

[0083] S720, XL 1 and XL 2 The similarity between them is determined as η.

[0084] It should be noted that those skilled in the art can use existing vector similarity determination methods to obtain η according to actual needs, which will not be elaborated here.

[0085] S800: Determine whether QT is the target user's voice based on η, A and the rear area image.

[0086] Further, step S800 may include the following steps:

[0087] S810, obtaining QT as the weight γ of the target user's speech.

[0088] In this embodiment, in the back row area scenario, the method for determining QT as the weight γ of the target user's voice is the same as the method in steps S410-S440 in the above embodiment, and is not repeated here.

[0089] S820: If η<η', determine that QT is not the target user's voice; wherein η' is a preset voice feature vector similarity threshold.

[0090] In this embodiment, if η<η', it means that the user corresponding to QR is different from the user corresponding to TE, and the voice is a conversation between two people. Therefore, it is determined that QT is not the target user voice. η' can be set according to actual needs, for example: η'=0.6.

[0091] S830: If η≥η', proceed to S520.

[0092] In this embodiment, if η≥η', it means that the user corresponding to QR is the same as the user corresponding to TE, and one person is speaking. Then, the judgment can be made through the method in steps S520-S560.

[0093] S900: If QT is the target user voice, QT is played through an audio device set in the main driving area.

[0094] In this embodiment, an audio device is provided in the main driving area, for example, an external speaker; if QT is the target user's voice, QT is played through the audio device provided in the main driving area, so that the driver can hear the voice spoken by the user to the driver without turning off the shielding function.

[0095] The vehicle-mounted multi-zone system based on DNN noise reduction technology of this embodiment obtains the initial audio in the shielding area of ​​the target vehicle, and uses a preset DNN noise reduction model to reduce the noise on the initial audio to obtain the initial user voice; extracts each keyword in the initial user voice, and then determines whether the initial user voice is the target user voice based on each keyword in the initial user voice and the seat in the car where the user corresponding to the initial user voice sits, and the image of the area corresponding to the seat, and combined with the number of users in the rear area; if it is the target user voice, the target user voice is played through an audio device set in the main driving area, so that without turning off the shielding function, the voice spoken by the user to the driver is played to the driver through the audio device in a targeted manner, thereby simplifying the user voice communication process and improving the efficiency of voice communication between drivers and passengers in the car.

[0096] Furthermore, when the user is in the back row area and there are multiple users in the back row area, the similarity of the voice is combined to further determine whether QT is the target user's voice, thereby enriching the scenario of target user voice judgment and further improving the accuracy of target user voice judgment.

[0097] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0098] Although some specific embodiments of the present invention have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are only for illustration, not for limiting the scope of the present invention. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention.

Claims

1. A vehicle-mounted multi-zone system based on DNN noise reduction technology, characterized in that: The system includes: a processor and a storage medium; the storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the following steps: S100, obtaining initial audio QR in a shielded area of ​​a target vehicle; wherein the shielded area includes a passenger area and a rear area; QR includes a voice emitted by a user; S200, inputting QR into a preset DNN noise reduction model to obtain a noise-reduced initial user speech QT corresponding to QR; S300, obtaining each keyword corresponding to QT and performing deduplication processing to obtain a keyword list A corresponding to QT = (A1, A2, ..., A i , …, A n ), i=1, 2,...,n; among them, A i is the i-th keyword corresponding to QT, and n is the number of keywords corresponding to QT; S400, if the user corresponding to QT is in the passenger seat area, determine whether QT is the target user voice according to the passenger seat area image and A; S500, if the user corresponding to QT is in the back row area, and the number of back row users is 1, determine whether QT is the target user voice according to the back row area image and A; S600: If the user corresponding to QT is in the back row area, and the number of back row users is greater than 1, then obtain the audio TE to be compared with a preset duration after QR; S700, obtaining the similarity η between QR and TE; S800, determining whether QT is the target user's voice according to η, A and the rear area image; S900, if QT is the target user voice, QT is played through an audio device set in the main driving area; Step S400 includes the following steps: S410, obtaining a first preset value W=1 and a second preset value NUM=0; S420, if W<n, then determine A W Whether it is in the preset target keyword library; proceed to S430; S430, if A W If the target keyword database is in the preset target keyword database, NUM is updated to NUM+1; otherwise, NUM is not updated; W is updated to W+1; and the process proceeds to S420; S440, if W=n, then according to NUM and n, determine QT as the weight γ=NUM / n of the target user's voice; S450, determining whether QT is the target user voice according to γ ​​and the passenger area image; Step S450 includes the following steps: S451, if γ≥γ', obtain s images of the passenger area taken in the time period corresponding to QR to obtain a list of passenger area images corresponding to QR, D=(D1, D2, ..., D r , …, D s ), r = 1, 2, …, s; where D r is the rth passenger area image taken in the time period corresponding to QR; otherwise, it is determined that QT is not the target user voice; γ' is the preset weight threshold; S452, obtaining the direction angle corresponding to the direction of the front passenger user's face in each front passenger area image in D, so as to obtain a first direction angle list α=(α1, α2, ..., α r , …, α s ); where α r D r The corresponding direction angle; S453, obtaining the number NA of direction angles in α that are within a first preset direction angle range; S454, if NA / s>β, determine QT as the target user voice; otherwise, determine QT is not the target user voice; wherein β is a preset direction angle quantity weight threshold.

2. The vehicle-mounted multi-zone system based on DNN noise reduction technology according to claim 1 is characterized in that: QR is obtained by the following steps: S110, obtaining the volume of each preset time point in the preset sliding time window TA, so as to obtain a collection point volume list B corresponding to TA = (B1, B2, ..., B j , …, B m ), j = 1, 2, …, m; where B j is the volume at the jth preset time point in TA, m is the number of preset time points in TA; the time interval between two adjacent preset time points is the same; S120, obtain the average volume corresponding to B YB = (1 / m) × ∑ m j=1 B j and the average volume YB' corresponding to the volume at each preset time point in the previous adjacent sliding time window of TA; S130, if YB-YB'≥ΔY, then sequentially obtain the average volume corresponding to each sliding time window in the q sliding time windows after TA to obtain an average volume list C=(C1, C2, ..., C p , …, C q ), p = 1, 2, …, q; where C p is the average volume corresponding to the pth sliding time window among the q sliding time windows after TA; ΔY is the preset average volume difference threshold; S140, obtaining the difference between each average volume in C and YB', so as to obtain a list of average volume differences corresponding to C ΔC = (ΔC1, ΔC2, ..., ΔC p ,…,ΔC q ); where ΔC p C p Difference from YB'; ΔC p =C p -YB'; S150, traverse ΔC, if ΔC1 to ΔCp-1 are all greater than or equal to ΔY, and ΔC p <ΔY, then the starting time point of TA and C p The audio between the end time points is determined as QR.

3. The vehicle-mounted multi-zone system based on DNN noise reduction technology according to claim 1 is characterized in that: Step S500 includes the following steps: S510, obtaining QT as the weight γ of the target user's voice; S520, if γ≥γ', then obtain y rear area images taken in the time period corresponding to QR to obtain a rear area image list E=(E1, E2, ..., E x ,…,E y ), x=1, 2, …, y; where E x is the xth rear area image captured in the time period corresponding to QR; otherwise, it is determined that QT is not the target user voice; γ' is the preset weight threshold; S530, obtaining the direction angle corresponding to the face direction of the rear row user in each rear row area image in E, so as to obtain a second direction angle list θ=(θ1, θ2, …, θ x ,…,θ y ); where θ x For E x The corresponding direction angle; S540, if the rear-seat user is located behind the front passenger seat, obtaining the number ND of direction angles in θ that are within a second preset direction angle range; S550, if the rear-seat user is located behind the driver, obtaining the number NE of direction angles in θ that are within a third preset direction angle range; S560, if ND / y>β or NE / y>β, determine that QT is the target user voice; otherwise, determine that QT is not the target user voice; wherein β is a preset direction angle quantity weight threshold.

4. The vehicle-mounted multi-zone system based on DNN noise reduction technology according to claim 3 is characterized in that: Step S800 includes the following steps: S810, obtaining QT as the weight γ of the target user's voice; S820, if η<η', determine that QT is not the target user's voice; wherein η' is a preset voice feature vector similarity threshold; S830: If η≥η', proceed to S520.

5. The vehicle-mounted multi-zone system based on DNN noise reduction technology according to claim 1 is characterized in that: Step S700 includes the following steps: S710, obtaining a speech feature vector XL1 corresponding to QR and a speech feature vector XL2 corresponding to TE; S720, determine the similarity between XL1 and XL2 as η.

Images