Noise reduction method and system for a tram driver's cab

Abstract

This application provides a noise reduction method and system for a tram driver's cab. The method includes: acquiring at least two noise signals; using the signals from the at least two noise signals arriving at an error sensor as desired signals; filtering the at least two noise signals respectively through a filter to obtain at least two secondary signals; passing the at least two secondary signals respectively through a secondary channel to obtain at least two cancellation signals; and interacting the at least two cancellation signals and the desired signals in the driver's cab to achieve noise reduction in the tram driver's cab.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of noise reduction technology, in particular to a noise reduction method and system for a trolleybus cab. BACKGROUND

[0002] The noise in the trolleybus cab is one of the important factors that makes the speech signal-to-noise ratio decrease and affects the speech interaction system, and the trolleybus cab is a closed space sound field, generally has multiple noise sources, and the noise signal is more difficult to spread.

[0003] The noise reduction method adopted by the prior art is mostly single-source active noise reduction. There are generally multiple noise sources in the trolleybus cab, but there is no corresponding noise reduction scheme for multiple noise sources at present, and after the active noise reduction system reduces the noise, there is still noise residue. In addition, the trolleybus cab is a closed space sound field, and the residual noise will reduce the signal-to-noise ratio, thereby affecting the interaction effect of the speech interaction system. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a noise reduction method and system for a trolleybus cab. Based on the above purpose, the present application provides a noise reduction method for a trolleybus cab, characterized in that it comprises:

[0005] obtaining at least two noise signals;

[0006] filtering at least two noise signals through a filter to obtain at least two secondary signals;

[0007] passing at least two secondary signals through a secondary channel to obtain at least two cancellation signals;

[0008] interacting at least two cancellation signals and an expected signal in the cab to achieve noise reduction in the trolleybus cab; the expected signal is the signal of at least two noise signals reaching an error sensor.

[0009] Optionally, it further comprises:

[0010] obtaining at least two error signals after at least two cancellation signals and the expected signal interact in the cab;

[0011] obtaining at least two filter-X signals in combination with at least two noise signals and the transfer function of the secondary channel;

[0012] obtaining a target function by using a least mean square error algorithm in combination with at least two error signals;

[0013] minimizing the target function according to an iterative formula by using a steepest descent method in combination with at least two filter-X signals, to adjust the coefficients of the filter.

[0014] Optionally, further comprising:

[0015] mixing the error signal and the speech to obtain an amplitude spectrum of the noisy speech;

[0016] inputting the amplitude spectrum of the noisy speech into a noise reduction model to obtain the noise-reduced speech.

[0017] Optionally, the at least two error signals are respectively collected by at least two error sensors;

[0018] The at least two secondary signals are respectively filtered by at least two sub-filters;

[0019] The error signal is calculated in the following manner:

[0020]

[0021] wherein, represents an error signal obtained by an l e th error sensor, represents an expected signal corresponding to the l e th error sensor, represents a secondary channel transfer function, y j represents a signal output by an j h th sub-filter, L represents a number of sub-filters, and L h represents a length of the secondary channel.

[0022] Optionally, the target function is calculated in the following manner:

[0023]

[0024] wherein, J represents the target function, represents an error signal obtained by an l e th error sensor, L represents a number of sub-filters, and E represents an average value of the independent variable.

[0025] Optionally, the iteration formula is as follows:

[0026]

[0027] wherein, represents a filter coefficient of an l e th sub-filter, represents an error signal obtained by an l e th error sensor, represents a filtered-X signal, μ represents a convergence step size, λ represents a weight coefficient momentum factor, Γ is a gamma function, and v represents an order of a fractional order.

[0028] Optionally, the feature extraction process comprises:

[0029] Logarithmic spectrum of the noisy speech is obtained to obtain the amplitude spectrum of the noisy speech.

[0030] Optionally, the process of obtaining the noise-reduced speech comprises:

[0031] The amplitude spectrum of the noisy speech is subjected to a noise reduction model to obtain a noise-reduced speech amplitude spectrum;

[0032] The noise-reduced speech amplitude spectrum is subjected to inverse Fourier transform combined with the phase to obtain the noise-reduced speech;

[0033] The phase is the phase of the noisy speech.

[0034] Optionally, the process of training the noise reduction model comprises:

[0035] A DNN deep neural network is established;

[0036] A training noise signal is collected;

[0037] A pure speech is collected;

[0038] The training noise signal and the pure speech are mixed to obtain a training noisy speech;

[0039] Feature extraction is performed on the training noisy speech to obtain an amplitude spectrum of the training noisy speech;

[0040] Feature extraction is performed on the pure speech to obtain an amplitude spectrum of the pure speech;

[0041] The amplitude spectrum of the pure speech is input into the DNN deep neural network;

[0042] The amplitude spectrum of the training noisy speech is input into the DNN deep neural network to obtain a training noise-reduced speech;

[0043] The training noise-reduced speech and the pure speech are compared to obtain an error function, and the DNN deep neural network is updated through back propagation to obtain the noise reduction model.

[0044] Based on the same inventive concept, the application also provides a noise reduction system for a trolleybus cab, comprising: at least two microphones, at least two error sensors, a multi-channel controller, and at least two loudspeakers;

[0045] The at least two microphones are respectively used to obtain the at least two noise signals;

[0046] The multi-channel controller is used to filter the at least two noise signals through a filter to obtain at least two secondary signals;

[0047] The at least two speakers are respectively used to pass the at least two secondary signals through secondary channels to obtain at least two cancellation signals;

[0048] The at least two error sensors are respectively used to obtain signals after the at least two cancellation signals and the expected signal interact with each other in the cab; the expected signal is a signal of the at least two noise signals reaching the error sensor.

[0049] As can be seen from the above, the noise reduction method and system of the tram cab provided by the present application filter at least two noise signals through a filter to obtain at least two secondary signals, pass the at least two secondary signals through secondary channels to obtain at least two cancellation signals, and realize noise reduction of the tram cab by interaction of the at least two cancellation signals and an expected signal. For a tram cab full of various noises, by collecting noises of multiple noise sources, the noise reduction method combines deep learning noise reduction method on the basis of active noise reduction, effectively reduces residual noise after noise reduction by only using the active noise reduction method. At the same time, the multiple noise sources are collected in a targeted manner, the noise reduction is more directional, the noise reduction effect is better, the signal-to-noise ratio is improved, and the voice interaction effect is improved. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the present application or related art, the drawings needed to be used in the embodiments or related art description will be briefly introduced. Obviously, the drawings in the following description are only embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0051] Figure 1 The noise reduction method flowchart of the tram cab of the embodiment of the present application;

[0052] Figure 2 The noise reduction model training flowchart of the embodiment of the present application;

[0053] Figure 3 The noise reduction system schematic diagram of the tram cab of the embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to specific embodiments and drawings.

[0055] It should be noted that, unless otherwise defined, technical terms or scientific terms used in the embodiments of the present application shall have the common meaning understood by one of ordinary skill in the art to which the embodiments of the present application belong. The terms "first", "second", and similar terms used in the embodiments of the present application do not denote any order, quantity, or importance, but are merely used to distinguish different components. The terms "include", "contain", and similar terms mean that the elements or objects before the terms encompass the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connect" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "upper", "lower", "left", "right", and the like are merely used to represent relative positional relationships, and when the absolute positions of the described objects change, the relative positional relationships can also change accordingly.

[0056] As described in the background section, the technical solutions in the related art for noise reduction by active noise reduction technology are generally directed to noise reduction of a single noise source. In addition, the related art also uses active noise reduction and deep learning to jointly reduce noise, but it is also a solution for a single noise source. However, the noise in a normal electric train cab is generally not a single noise source, and obviously using the prior art to reduce noise cannot meet the noise reduction needs of the electric train cab.

[0057] In view of the above considerations, the present application proposes a noise reduction method for an electric train cab, which is based on an active noise reduction system and a pre-trained noise reduction model, and reduces the noise in the electric train cab by collecting multiple noise sources, thereby improving the voice interaction effect.

[0058] In the following, the technical solutions of the present application will be described in detail through specific embodiments.

[0059] Reference Figure 1 The noise reduction method for the electric train cab of the present application includes the following steps:

[0060] Step S101, acquiring at least two noise signals.

[0061] In this step, at least two noise signals are first acquired, and subsequent noise reduction processing is based on the noise signals.

[0062] In the process of implementing the present application, the inventors found that there are at least two noise sources in the electric train cab during actual driving of the electric train. In view of this situation, the inventors designed a method that can reduce at least two noise signals, so that at least two noise signals can be acquired when acquiring noise signals.

[0063] The noise signals include at least motor noise signals, air conditioner noise signals, outdoor wind noise signals, tire noise signals, and the like.

[0064] In some optional embodiments, the motor noise signals and the air conditioner noise signals are taken as examples to describe the noise reduction method and system of the tram driver's room in detail. In the following steps, the motor noise signals and the air conditioner noise signals are also taken as examples to describe the steps in detail. In the embodiment, step S101 is specifically implemented as obtaining the motor noise signals and the air conditioner noise signals.

[0065] It is easy to understand that the motor noise signals and the air conditioner noise signals only represent one embodiment of the noise reduction method and system of the tram driver's room, and the noise reduction method and system of the tram driver's room are not limited to the motor noise signals and the air conditioner noise signals. The noise reduction method and system of the tram driver's room can also be used to process other noise signals such as outdoor wind noise signals and tire noise signals. The noise reduction method and system of the tram driver's room can be applied to other embodiments different from the above-mentioned embodiments in actual application, that is, the number of noise signals does not substantially affect the implementation of the noise reduction method and system of the tram driver's room.

[0066] In step S102, at least two noise signals are filtered by a filter respectively to obtain at least two secondary signals.

[0067] In this step, each noise signal is filtered by a corresponding sub-filter to obtain a corresponding cancellation signal. The calculation formula of the signal is as follows:

[0068] y(n) = X(n)W(n) T (n)W(n)

[0069] wherein y(n) represents a secondary signal, X(n) represents a noise signal, and W(n) represents a vector of filter coefficients.

[0070] In this embodiment, the motor noise signal is filtered by a corresponding sub-filter to obtain a motor noise secondary signal. The air conditioner noise signal is filtered by a corresponding sub-filter to obtain an air conditioner noise secondary signal.

[0071] In step S103, at least two secondary signals are respectively passed through a secondary channel to obtain at least two cancellation signals.

[0072] In this step, the secondary channel includes three parts: electro-acoustic devices (such as a loudspeaker, an error sensor, and the like), a sound field, and an electronic circuit.

[0073] In the embodiment, the motor noise secondary signal and the air conditioner noise secondary signal need to pass through a physical path to reach the error sensor, the physical path is a secondary channel, the motor noise secondary signal obtains a motor noise cancellation signal after passing through the secondary channel, and the air conditioner noise secondary signal obtains an air conditioner noise cancellation signal after passing through the secondary channel; the expected signal is at least two noise signals reaching the error sensor.

[0074] In step S104, the at least two cancellation signals and the expected signal interact with each other in the cab to achieve noise reduction in the cab of the tram.

[0075] In this step, the noise signal reaching the error sensor is different from the noise signal near the noise source, the noise signal needs to pass through a physical path to reach the error sensor, and in this process, the noise signal is lost accordingly, therefore, the noise signal reaching the error sensor is referred to as an expected signal, and the noise signal and the expected signal are different from each other. The error sensor is located in the cab.

[0076] In the embodiment, the motor noise signal and the air conditioner noise signal reaching the error sensor are expected signals.

[0077] In the embodiment, the motor noise signal and the air conditioner noise signal reach the error sensor to obtain expected signals, and the motor noise cancellation signal and the air conditioner noise cancellation signal interact with the expected signals, which is equivalent to the interaction between the motor noise cancellation signal and the motor noise expected signal and the interaction between the air conditioner noise cancellation signal and the air conditioner noise expected signal, and the noise can be effectively cancelled.

[0078] As an optional embodiment, after the noise reduction method in the cab of the tram in the foregoing embodiment effectively cancels the noise, the method further includes:

[0079] After the at least two cancellation signals and the expected signal interact with each other in the cab, at least two error signals are obtained.

[0080] At least two filter-X signals are obtained in combination with the at least two noise signals and a transfer function of the secondary channel.

[0081] In combination with the at least two error signals, a target function is obtained by using a least mean square error algorithm.

[0082] In combination with the at least two filter-X signals, the target function is minimized by using a steepest descent method according to an iterative formula, so as to adjust coefficients of a filter.

[0083] It should be noted that the noise reduction system of the present application only updates the filter coefficients when there is no voice in the driver's room, and if there is voice in the driver's room, the filter coefficients stop updating. During the whole process of the noise reduction system working, the cancellation signal will continue to be output, and when the filter coefficients are updated, the cancellation signal will also change.

[0084] In the above steps, the presence of the secondary channel causes the cancellation signal output during active noise reduction to have a time delay from the noise signal, making the active noise reduction system unstable, so the secondary channel needs to be compensated to offset its influence. Therefore, the present application is provided with a compensation channel of the secondary channel, and after the noise signal passes through the compensation channel, a filter-X signal is obtained to compensate for the delay error caused by the secondary channel to the system. Among them, at least two error signals are respectively collected by at least two error sensors; at least two signals are respectively filtered by at least two sub-filters.

[0085] The filter-X signal is LxL h Matrix:

[0086]

[0087] Among them,

[0088]

[0089] Among them, L represents the number of sub-filters, L h represents the length of the secondary channel, represents the filter-X signal after the l e th sub-filter, represents the filter coefficient vector of the finite impulse response FIR filter, represents the vector of the l e th noise signal.

[0090] Optionally, the calculation method of the error signal comprises:

[0091]

[0092] Among them, represents the error signal obtained by the l e th error sensor, represents the expected signal corresponding to the l e th error sensor, represents the secondary channel transfer function, y j represents the signal output by the jth sub-filter, L represents the number of sub-filters, L h represents the length of the secondary channel.

[0093] In this embodiment, the motor noise error sensor acquires the signal after the interaction of the motor noise cancellation signal and the desired signal. However, the air conditioning noise signal also exists in the driver's cab at this time and is acquired by the motor noise error sensor. Therefore, the motor noise error sensor acquires the signal after the interaction of the motor noise cancellation signal, the air conditioning noise cancellation signal and the desired signal.

[0094] In this step, the objective function is calculated as follows:

[0095]

[0096] Where J represents the objective function, Indicates the lth e The error signal acquired by the error sensor, where L represents the number of sub-filters and E represents the average value of the independent variable.

[0097] In this step, the iterative formula includes:

[0098]

[0099] in, Indicates the lth e The filter coefficients of each sub-filter Indicates the lth e Error signals acquired by each error sensor Let denot -X, μ denote the convergence step size, λ denote the momentum factor of the weighting coefficients, Γ be the gamma function, and v denote the order of the fractional order.

[0100] As an optional embodiment, the noise reduction method for the tram driver's cab in the foregoing embodiments, after effectively canceling the noise, further includes:

[0101] The error signal and speech are mixed and then feature extracted to obtain the amplitude spectrum of the noisy speech.

[0102] The amplitude spectrum of the noisy speech is input into the denoising model to obtain the denoised speech.

[0103] Optional, the feature extraction process includes:

[0104] The amplitude spectrum of the noisy speech is obtained by taking the logarithmic spectrum of the noisy speech.

[0105] Optionally, the process of acquiring denoised speech includes:

[0106] The amplitude spectrum of the noisy speech is obtained by passing it through a noise reduction model.

[0107] The amplitude spectrum of the denoised speech is subjected to inverse Fourier transform by combining phase information to obtain the denoised speech;

[0108] The phase is a phase of the noisy speech.

[0109] In this step, there will still be residual noise signals after the cancellation of noise by the active noise reduction system, which is difficult to avoid, so the application adopts DNN deep neural network noise reduction after active noise reduction, which will be better.

[0110] In this step, a microphone is separately arranged near the error sensor in the active noise reduction system to collect the mixed signal of the error signal and the speech signal, and then the feature extraction is performed.

[0111] In some optional embodiments, the training process of the noise reduction model is introduced in detail in this embodiment taking motor noise, air conditioner noise and speech as examples. As shown in Figure 2 , the training process of the model includes the following steps:

[0112] Step S201, establishing a DNN deep neural network.

[0113] In this step, it is assumed that the DNN deep neural network has L layers, the number of neurons in each layer is M, the input layer and the output layer are 1 layer respectively, and the hidden layer is L-2 (L>2) layer, then the number of neurons in any layer l (1 l l ) of the L layers is M

[0114]

[0115] , wherein, represents the output function of the i-th (1 l ) neuron of the l-th layer of the neural network, represents the weight parameter connecting the j-th neuron of the l-1-th layer and the i-th neuron of the l-th layer of the neural network, represents the activation function value of the j-th neuron of the l-1-th layer of the neural network, represents the bias parameter of the i-th neuron of the l-th layer of the neural network.

[0116] In addition, the output function of the i-th neuron of the first layer of the neural network is the i-th input speech of the input layer of the neural network, at the same time, the activation function value of the i-th (1 l ) neuron of the first layer The input speech here is the amplitude spectrum obtained by performing Fourier transform on the actual speech to extract the features.

[0117] ​The activation function is a Sigmoid function, and its specific expression is:

[0118]

[0119] Wherein, x is the independent variable, and f(x) is the activation function.

[0120] In this embodiment, the deep neural network DNN comprises an input layer, a hidden layer and an output layer, wherein the hidden layer has 2 layers, and each hidden layer has 256 neurons.

[0121] Step S202, collecting training noise signals.

[0122] In this step, when collecting training noise signals, any position in the driver's room can be collected, but the effect is best near the error sensor.

[0123] Because the effect of collecting training noise signals near the error sensor is best, in this embodiment, the training motor noise signals and the training air conditioner noise signals are collected near the error sensor in the driver's room. It should be noted that the training noise signals collected in this embodiment are mixed noise signals of the training motor noise signals and the training air conditioner noise signals near the error sensor, and in other embodiments, they are mixed noise signals of other noise signals that need to be denoised.

[0124] Step S203, collecting pure speech.

[0125] In this step, the pure speech is collected in a relatively quiet environment.

[0126] Step S204, mixing the training noise signals and the pure speech to obtain training noisy speech.

[0127] In this embodiment, the training motor noise signals, the training air conditioner noise signals and the training pure speech are mixed to obtain the training noisy speech.

[0128] Step S205, extracting features from the training noisy speech to obtain the amplitude spectrum of the training noisy speech.

[0129] In this step, the feature extraction process includes:

[0130] Taking the logarithmic spectrum of the training noisy speech, the amplitude spectrum of the training noisy speech is obtained.

[0131] Step S206, extracting features from the pure speech to obtain the amplitude spectrum of the pure speech.

[0132] In this step, the feature extraction process includes:

[0133] Logarithmic spectrum of the pure speech is obtained to obtain an amplitude spectrum of the pure speech.

[0134] In step S207, the amplitude spectrum of the pure speech is input into the DNN deep neural network.

[0135] In this step, the pure speech is used for comparison with the training denoised speech, and therefore needs to be input into the DNN deep neural network.

[0136] In step S208, the amplitude spectrum of the training noisy speech is input into the DNN deep neural network to obtain the training denoised speech.

[0137] In this step, the process of obtaining the training denoised speech includes:

[0138] The amplitude spectrum of the training noisy speech is input into the denoising model to obtain the amplitude spectrum of the training denoised speech;

[0139] The amplitude spectrum of the training denoised speech is combined with the phase to perform inverse Fourier transform to obtain the training denoised speech.

[0140] The phase is the phase of the training noisy speech.

[0141] In step S209, the training denoised speech and the pure speech are compared, and the DNN deep neural network is updated through back propagation to obtain the denoising model.

[0142] In this step, the training denoised speech and the pure speech are compared, and the error is calculated through an error function, and the error function adopts a mean square error algorithm.

[0143]

[0144] wherein E represents an error function, M L represents the number of neurons of the output layer of the neural network, i.e., the dimension of the output data, y k represents the training pure speech amplitude spectrum corresponding to the training noisy speech amplitude spectrum corresponding to the kth neuron. represents the training denoised speech amplitude spectrum corresponding to the training noisy speech amplitude spectrum corresponding to the kth neuron.

[0145] Taking the specific back propagation update process of the weight parameter l connecting the jth neuron of the l-1th layer and the ith neuron of the lth (1 layer (1 ) and the bias parameter of the ith neuron of the lth layer as an example, the process includes:

[0146]

[0147]

[0148] wherein,

[0149]

[0150]

[0151] In addition,

[0152]

[0153] wherein, E represents an error function, represents a weight parameter connecting the jth neuron of the l-1th layer and the ith neuron of the lth (1 l ) layer of the deep neural network DNN, η represents a proportional coefficient, and represents a learning rate of the deep neural network DNN, represents a bias parameter of the ith neuron of the lth layer, represents an activation function value of the jth neuron of the l-1th layer of the neural network, represents an output function of the ith (1 l ) neuron of the lth layer of the neural network.

[0154] The embodiment realizes noise reduction of the driver's room of the electric train. Firstly, the obtained at least two noise signals are filtered by a filter, and then pass through a secondary channel to obtain a cancellation signal. The cancellation signal and the expected signal of the noise signal reaching the error sensor interact with each other to realize noise reduction of the driver's room of the electric train. After the interaction, there is still residual noise, which is together with the voice and then passes through a noise reduction model. The voice after noise reduction is sent to a voice interaction system.

[0155] In the past noise reduction method, the person skilled in the art usually only uses the active noise reduction method to reduce noise, and only reduces noise for one noise source. Although this method can reduce the noise in the driver's room to a certain extent, the driver's room generally has multiple noise sources. Obviously, reducing noise for only one noise source cannot meet the noise reduction demand of the driver's room. The present application reduces noise for multiple noise sources, and each noise source has a corresponding cancellation signal for targeted noise reduction. This is specifically reflected in that the active noise reduction system of the present application is provided with multiple microphones to respectively obtain noise of different noise sources.

[0156] In addition, even if there is only one noise in the driver's room, only using the active noise reduction method to process it will also have residual noise. Therefore, the present application makes the residual noise pass through a noise reduction model together with the voice after the active noise reduction system, so as to perform secondary noise reduction on the residual noise, thereby effectively reducing the residual noise in the driver's room.

[0157] In conclusion, the noise reduction method and system for the trolley driver's room can reduce noise from multiple noise sources and deal with residual noise more thoroughly, thereby reducing the noise in the driver's room and improving the effect of the voice interaction system.

[0158] Based on the same inventive concept, the present application also provides a noise reduction system for a trolley driver's room. Referring to Figure 3 , the noise reduction system for the trolley driver's room comprises:

[0159] at least two microphones S301, at least two error sensors S304, a multi-channel controller S302, and at least two speakers S303;

[0160] The at least two microphones S301 are respectively configured to obtain the at least two noise signals;

[0161] The multi-channel controller S302 is configured to filter the at least two noise signals through a filter to obtain at least two secondary signals;

[0162] The at least two speakers S303 are respectively configured to pass the at least two secondary signals through a secondary channel to obtain at least two cancellation signals;

[0163] The at least two error sensors S304 are respectively configured to obtain signals after the at least two cancellation signals and an expected signal interact in the driver's room; the expected signal is a signal obtained when the at least two noise signals reach the error sensor.

[0164] It should be noted that the method of the present application can be executed by a single device, such as a computer or a server, etc. The method of the present application can also be applied to a distributed scenario, and completed by multiple devices in cooperation. In this distributed scenario, one of the multiple devices can only execute one or more steps in the method of the present application, and the multiple devices can interact with each other to complete the method.

[0165] It should be noted that some embodiments of the present application have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than the order described above and still achieve the desired result. In addition, the processes depicted in the accompanying drawings do not necessarily require the particular order shown or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous or required.

[0166] Those of ordinary skill in the art will realize that the foregoing discussion of any of the embodiments has been presented for the purpose of illustration and description and is not intended to be exhaustive or to limit the application to the precise forms described, and that various adaptations and modifications are possible within the scope and spirit of the application. For example, while the embodiments discussed above have been described in the context of a memory device, other memory architectures (e.g., dynamic RAM (DRAM)) can use the embodiments discussed.

[0167] In addition, to simplify the description and discussion, and so as not to make the embodiments of the application difficult to understand, the well-known power / ground connections to integrated circuit (IC) chips and other components can or can not be shown in the provided drawings. Further, devices can be shown in block diagram form so as not to make the embodiments of the application difficult to understand, and this also takes into account the fact that the details regarding the implementation of these block diagram devices are highly dependent on the platform in which the embodiments of the application are to be implemented (i.e., these details should be well within the understanding of one of ordinary skill in the art). Where specific details (e.g., circuitry) are set forth in order to describe an illustrative embodiment of the application, it should be apparent to one of ordinary skill in the art that the embodiments of the application can be practiced without or with variations of these specific details. Thus, the description should not be viewed as limiting the application, but rather as merely describing illustrative embodiments.

[0168] While the application has been described in connection with specific embodiments thereof, it will be understood that many modifications, variations and alternatives will be apparent to those skilled in the art as a result of the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) can use the embodiments discussed.

[0169] It is therefore intended that the embodiments of the application embrace all such alternatives, modifications and variations as falling within the broad scope of the appended claims. Accordingly, any and all departures from the above described embodiments are intended to be included within the scope of the application as defined by the following claims.

Claims

1. A method of reducing noise in a cab of an electric vehicle, characterized in that, The method comprises the following steps: acquiring at least two noise signals; filtering the at least two noise signals through filters respectively to obtain at least two secondary signals; passing the at least two secondary signals through secondary channels respectively to obtain at least two cancellation signals; interacting the at least two cancellation signals and an expected signal in a driver's room to realize noise reduction of the driver's room; the expected signal is a signal of the at least two noise signals reaching error sensors; wherein the method further comprises: obtaining at least two error signals after the at least two cancellation signals and the expected signal interact in the driver's room; obtaining at least two filter-X signals by combining the at least two noise signals and a transfer function of the secondary channels; the filter-X signals are used to compensate for delay errors caused by the secondary channels; obtaining a target function by combining the at least two error signals and using a least mean square error algorithm; minimizing the target function by using a steepest descent method according to an iterative formula to adjust coefficients of the filters in combination with the at least two filter-X signals.

2. The method of claim 1, wherein, The method further comprises: extracting features from a mixture of the error signals and speech to obtain an amplitude spectrum of noisy speech; inputting the amplitude spectrum of the noisy speech into a noise reduction model to obtain noise-reduced speech.

3. The method of claim 1, wherein, The at least two error signals are respectively collected by the at least two error sensors; The at least two secondary signals are respectively filtered by at least two sub-filters; The error signal is calculated in the following manner: wherein, represents an error signal acquired by the lth e error sensor, represents a desired signal corresponding to the lth e error sensor, represents a secondary path transfer function, y j represents a signal output by the jth sub-filter, and L represents the number of sub-filters, L h represents the length of the secondary path.

4. The method of claim 1, wherein, The target function is calculated in the following manner: where J denotes the objective function, denotes the error signal acquired by the lth e error sensor, L denotes the number of sub-filters, and E denotes the averaging of the argument.

5. The method of claim 1, wherein, The iterative formula is as follows: wherein, represents filter coefficients of an lth e sub-filter, represents an error signal acquired by an lth e error sensor, represents a filtered-X signal, μ represents a convergence step size, λ represents a weight coefficient momentum factor, Γ is a gamma function, and v represents an order of a fractional order.

6. The method of claim 2, wherein, The feature extraction process comprises: taking a logarithmic spectrum of the noisy speech to obtain an amplitude spectrum of the noisy speech.

7. The method of claim 2, wherein, The noise-reduced speech is obtained in the following manner: The amplitude spectrum of the noisy speech is inputted into the noise reduction model to obtain an amplitude spectrum of noise-reduced speech; The noise-reduced speech is obtained by inverse Fourier transform of the amplitude spectrum of the noise-reduced speech in combination with phase information; The phase is a phase of the noisy speech.

8. The method of claim 2, wherein, The noise reduction model is trained in the following manner: a DNN deep neural network is established; training noise signals are collected; pure speech is collected; the training noise signals and the pure speech are mixed to obtain training noisy speech; features of the training noisy speech are extracted to obtain an amplitude spectrum of the training noisy speech; features of the pure speech are extracted to obtain an amplitude spectrum of the pure speech; the amplitude spectrum of the pure speech is inputted into the DNN deep neural network; the amplitude spectrum of the training noisy speech is inputted into the DNN deep neural network to obtain training noise-reduced speech; the training noise-reduced speech and the pure speech are compared to obtain an error function, and the DNN deep neural network is updated through back propagation to obtain the noise reduction model.

9. A noise reduction system for a trolley car cab, the noise reduction system applying the method of noise reduction of any one of claims 1 to 8, wherein, The method comprises the following steps: at least two microphones, at least two error sensors, a multi-channel controller, and at least two loudspeakers; the at least two microphones are used to acquire the at least two noise signals respectively; the multi-channel controller is used to filter the at least two noise signals through filters respectively to obtain at least two secondary signals; At least two loudspeakers are used to respectively pass at least two said secondary signals through secondary channels to obtain at least two cancellation signals; At least two error sensors are used to respectively obtain signals of mutual interaction of at least two said cancellation signals and said expected signal in the driver's room; said expected signal is a signal of arrival of at least two said noise signals at the error sensor.

Images