Active noise control method, active noise control device, and program

Abstract

This invention seeks to improve noise control performance even in bands where there is little suppression of noise with conventional active noise control. A reference microphone 10 detects noise x(n) of a noise source N. A noise prediction unit 20 takes the noise x(n) as input, and predicts the noise that will propagate through a primary path P and reach a specific location Z (noise x'(n)). A noise control filter 30 uses the noise x'(n) and a coefficient K to generate canceling sound y(n). A secondary sound source 40 emits the canceling sound y(n). A secondary path model 50 takes the noise x'(n) as input, and estimates the noise that will propagate through a secondary path S and reach the specific location Z (noise x''(n)). An error microphone 60 detects interference sound e(n) composed of noise d(n) and noise d'(n). A coefficient update unit 70 takes the estimated noise x''(n) and the interference sound e(n) as inputs, and updates the coefficient K.

Description

[Technical Field]

[0001] This disclosure relates to active noise control technology. [Background technology]

[0002] An active noise control (ANC) device attenuates noise at a specific location by adding a sound of opposite phase to the noise at that location. Active noise control devices often consist of a microphone called an error microphone that measures the amount of noise attenuation, a speaker that generates a sound called a secondary sound source that cancels out the noise (a cancellation sound for noise control), and a noise control filter that estimates the sound to be canceled. In addition to the above configuration, a configuration that includes a reference microphone installed in a different location from the error microphone and that records the noise is also commonly used. There are two types of noise control filters: one that remains fixed while the active noise control is in operation, and one that adaptively updates the noise control filter using a coefficient update unit based on the noise suppression results, etc. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Yoshinobu Kajikawa, "Recent Topics and Applications of Active Noise Control", Research Report Music Information Science (MUS) 2015.3 (2015): 1-6 Summary of the Invention [Problem to be solved by the invention]

[0004] As Non-Patent Document 1 indicates, active noise control is generally used to control noise with low-frequency components, while passive noise control techniques that physically attenuate noise, such as covering the ears with in-ear earphones, are used to control noise with high-frequency components. In spaces with fixed layouts such as public transportation, active noise control techniques are useful for providing users with a comfortable, quiet space. However, there is a problem in that it is difficult to completely eliminate sound across all frequency bands.

[0005] The purpose of this disclosure has been made in consideration of the above-mentioned technical problems, and is to provide an active noise control method that is expected to improve noise control performance even in frequency bands where noise suppression is low with conventional active noise control. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, one aspect of the active noise control method disclosed herein is an active noise control method executed by an active noise control device. A reference microphone detects noise from a noise source. A noise prediction unit predicts the detected noise that propagates along a primary path from the noise source to a specific position and reaches the specific position. A noise control filter generates a cancellation sound for noise control using the predicted noise and predetermined coefficients. A secondary sound source radiates the generated cancellation sound. A secondary path model estimates a predicted noise that propagates along a secondary path from the secondary sound source to the specific position and reaches the specific position. An error microphone placed at a specific position detects interference sound between the noise from the noise source that propagates along the primary path and reaches the specific position, and the radiated cancellation sound that propagates along the secondary path and reaches the specific position. A coefficient update unit receives the estimated noise and the detected interference sound as inputs and updates the predetermined coefficients used by the noise control filter. [Effects of the Invention]

[0007] According to this disclosure, noise predictions from noise sources can be notified to the noise control filter at an early stage, which is expected to improve noise control performance even in frequency bands where conventional active noise control achieves little noise suppression. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram illustrating the functional configuration of an active noise control device according to this embodiment. [Figure 2] FIG. 2 is a process flow diagram illustrating the process procedure of the active noise control method of this embodiment. [Figure 3] FIG. 3 is a diagram illustrating an example of the functional configuration of the noise prediction learning device of this embodiment. [Figure 4] FIG. 4 is a process flow diagram illustrating the process steps of the noise prediction learning method of this embodiment. [Figure 5] FIG. 5 is a diagram illustrating an example of a functional configuration of an active noise control device according to a modified example of this embodiment. [Figure 6] FIG. 6 is a process flow diagram illustrating a process procedure of an active noise control method according to a modified example of this embodiment. [Figure 7] FIG. 7 is a diagram illustrating an example of the functional configuration of a computer. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present disclosure will be described in detail. Note that components having the same functions in the drawings will be assigned the same reference numerals, and redundant explanations will be omitted.

[0010] [Active noise control system] As shown in Fig. 1, the active noise control device 1 according to this embodiment includes a reference microphone 10, a noise prediction unit 20, a noise control filter 30, a secondary sound source 40, a secondary path model 50, an error microphone 60, and a coefficient update unit 70. The active noise control method according to this embodiment is realized by the active noise control device 1 performing the processing of each step shown in Fig. 2.

[0011] Hereinafter, the processing procedure of the active noise control method executed by the active noise control device 1 of the embodiment will be described with reference to FIGS.

[0012] (Reference Mike 10) In the active noise control device 1 shown in Fig. 1, it is assumed that noise x(n) is radiated from noise source N. In this case, the reference microphone 10 detects the noise x(n) from noise source N, performs necessary amplification processing (not shown) and analog-to-digital conversion processing (not shown), and then outputs the detected noise to the noise prediction unit 20 (step S10). In the active noise control device 1 of the present disclosure, a vibration pickup may be installed in addition to the reference microphone 10. This is because vibrations propagate faster in solids than in gases, and therefore noise may be recorded in the form of vibrations faster than it propagates as sound.

[0013] In this disclosure, the path along which the noise x(n) propagates from the noise source N to the specific position Z is referred to as the primary path P.

[0014] (Noise Prediction Section 20) The noise prediction unit 20 predicts the noise x(n) detected by the reference microphone 10, which propagates through the primary path P and reaches the specific position Z. That is, the noise prediction unit 20 receives the noise x(n) from the reference microphone 10 as input, predicts the noise d(n) of the noise source N, which propagates through the primary path P and reaches the specific position Z, and outputs the predicted noise x'(n) to the noise control filter 30 and the secondary path model 50 (step S20). The noise prediction unit 20 may be configured using linear convolution, for example, or may be configured using a trained network (trained model) that is trained using neural network technology, as described below. The configuration of the noise prediction unit 20 is determined by receiving parameters stored in the noise prediction parameter p.

[0015] (Noise Control Filter 30) The noise control filter 30 generates a cancellation sound y(n) for noise control using the noise x'(n) predicted by the noise prediction unit 20 and a predetermined coefficient K. That is, the noise control filter 30 updates the coefficients (parameters) of the noise control filter 30 using the predetermined coefficient K received from the coefficient update unit 70 as described below, and then generates a cancellation sound y(n) for noise control using the noise x'(n) that is the noise prediction result received from the noise prediction unit 20, and outputs it to the secondary sound source 40 (step S30). The noise control filter 30 is, for example, a model represented by the convolution of a coefficient and a signal.

[0016] (Secondary sound source 40) The secondary sound source 40 performs necessary amplification processing (not shown) and digital-to-analog conversion processing (not shown) on the cancellation sound y(n) received from the noise control filter 30, and then emits the cancellation sound y(n) toward the error microphone 60 (step S40). In the present disclosure, the path along which the cancellation sound y(n) propagates from the secondary sound source 40 to the specific position Z is referred to as the secondary path S.

[0017] (Secondary Pathway Model 50) The secondary path model 50 estimates the noise x'(n) predicted by the noise prediction unit 20, which propagates through the secondary path S and reaches the specific position Z. That is, in order to compensate for the influence of the secondary path S, the secondary path model 50 estimates the noise that would occur if it were assumed that the noise x'(n), which is the prediction result received from the noise prediction unit 20, propagates through the secondary path S and reaches the specific position Z, and outputs the estimated noise X''(n) to the coefficient update unit 70 (step S50). The secondary path model 50 is, for example, a model represented by the convolution of a coefficient and a signal.

[0018] (Error microphone 60) The error microphone 60 is disposed at a specific position Z. The error microphone 60 detects an interference sound e(n) between noise d(n) from a noise source N that propagates through a primary path P and arrives at the specific position Z, and noise d'(n) from a secondary sound source 40 that propagates through a secondary path S and arrives at the specific position Z. That is, the error microphone 60 detects the interference sound e(n) consisting of two noises: noise d(n) from the noise source that propagates through a primary path P and arrives at the specific position Z, and noise d'(n) from the secondary sound source 40 that propagates through a secondary path S and arrives at the specific position Z (noise resulting from the cancellation sound y(n) that has been subjected to necessary amplification processing and digital-to-analog conversion processing being propagated through the secondary path S and arriving at the specific position Z), and outputs the interference sound e(n) to the coefficient update unit 70 after performing necessary amplification processing (not shown) and analog-to-digital conversion processing (not shown) (step S60). Instead of the error microphone 60, a virtual error microphone may be used to spatially predict noise at the error microphone based on the recording results of another microphone.

[0019] (Coefficient update unit 70) The coefficient update unit 70 receives as input the noise X''(n) estimated by the secondary path model 50 and the interference sound e(n) detected by the error microphone 60, updates the predetermined coefficient K used by the noise control filter 30 to generate a cancellation sound y(n) for noise control to a new coefficient K, and outputs the new coefficient K to the noise control filter 30 (step S70). Note that the method of updating the coefficients of the noise control filter 30 using the secondary path model 50 and the coefficient update unit 70 corresponds to the Filtered-x algorithm described in the above-mentioned Non-Patent Document 1.

[0020] The active noise control device 1 has been described above. The active noise control device 1 configured as described above is expected to have the following effects. Specifically, the secondary path model 50 and the coefficient update unit 70 compensate for the influence of the secondary path S and update the coefficients (parameters) used in the noise control filter 30. Here, the secondary path model 50 and the noise control filter 30 acquire noise x'(n), which is a prediction result of noise d(n), before the error microphone 60 detects the noise d(n). This x'(n) is a result obtained by the noise prediction unit 20 taking into account the influence of the primary path P. Therefore, compared to a case where the noise prediction unit 20 is not provided, a cancellation sound y(n) for noise control with a higher attenuation effect can be generated. That is, the active noise control device 1 can notify the noise control filter 30 of the noise prediction (noise x'(n)) of the noise x(n) from the noise source N at an earlier timing than the error microphone 60. As a result, improved noise control performance is expected even in frequency bands where noise suppression is low in conventional active noise control.

[0021] The present disclosure is expected to improve noise control performance, particularly in the high-frequency band. Because high-frequency sounds have short wavelengths, even a slight phase difference between the noise observed at the error microphone 60 and the cancellation sound prevents the sound waves from canceling each other, making it impossible to suppress the noise. Without the noise prediction unit 20, the noise control filter 30 would not be able to generate the cancellation sound y(n) in time, and the noise d(n) would arrive at the error microphone 60 before the noise d'(n), resulting in performance degradation. In contrast, by providing the second estimator 20 described in the present disclosure, it is possible to predict the noise d(n) at the time the noise d'(n) arrives at the error microphone 60, eliminating performance degradation due to the cancellation sound y(n) not being generated in time. In other words, the noise prediction unit 20 makes it possible to predict, from the signal observed by the reference microphone 10, a noise signal (noise d(n)) that has not yet been observed but will soon arrive at the error microphone 60 (i.e., a future noise signal) from the noise signal (noise d(n)) observed at the reference microphone 10.

[0022] [Noise prediction learning device] As described above, the noise prediction unit 20 may be configured to learn using neural network technology and use the learned network (trained model). When the noise prediction unit 20 of this embodiment uses a neural network, the trained model may be configured to be learned by a noise prediction learning device as described below. As shown in FIG. 3, the noise prediction learning device 300 of the present disclosure includes a parameter storage unit 310, a prediction model unit 320, an objective function calculation unit 330, and a parameter update unit 340. The noise prediction learning method of this embodiment is realized by the noise prediction learning device 300 performing the processing of each step shown in FIG. 4.

[0023] Hereinafter, the processing procedure of the noise predictive learning method executed by the noise predictive learning device 300 of this embodiment will be described with reference to FIGS.

[0024] (Parameter storage unit 310) In FIG. 3, the parameter storage unit 310 stores noise prediction parameters p used by the prediction model unit 320, which will be described later, and outputs the latest noise prediction parameters p to the prediction model unit 320 (step S310).

[0025] (Prediction model unit 320) The prediction model unit 320 updates its own parameters using the noise prediction parameter p received from the parameter storage unit 310, and then predicts the signal sequence received from the training dataset D as it propagates along the primary path P and reaches a specific position Z, and outputs the predicted signal sequence to the objective function calculation unit 330 (step S320).

[0026] The training data included in the training dataset D includes, for example, K pairs of data: a signal sequence x(t) (hereinafter also referred to as "x_t") of length L from time tL-1 to time t, and a signal sequence x(t) (hereinafter also referred to as "y_t") of length N from time t+M to time t+M+N-1. x(t) may be a one-dimensional signal recorded from a single microphone, or a multidimensional signal recorded from multiple microphones. The prediction model unit 320 receives the above-mentioned x_t as input and outputs y'_t, which is the prediction result of y_t, to the objective function calculation unit 330.

[0027] (Objective function calculation unit 330) The objective function calculation unit 330 calculates an objective function using as input the signal sequence of the prediction result received from the prediction model unit 320 and the signal sequence of the correct data received from the training dataset D, and outputs the calculation result to the parameter update unit 340 (step S330). Explaining this using the example of the dataset above, the objective function calculation unit 330 receives as input the signal sequence y'_t of the prediction result and the signal sequence y_t of the correct data received from the training dataset, calculates the distance between them using an appropriate method, calculates the objective function, and outputs the calculation result to the parameter update unit 340 (step S330).

[0028] (Parameter update unit 340) If the parameter update unit 340 determines that the objective function received from the objective function calculation unit 330 does not satisfy a predetermined condition, it updates the noise prediction parameter p and outputs it to the parameter storage unit 310. To update the parameters, for example, the parameters of a neural network are updated using a method such as a gradient method. On the other hand, if the parameter update unit 340 determines that the objective function received from the objective function calculation unit 330 satisfies a predetermined condition, it outputs the prediction model unit having the current noise prediction parameter p as the trained model W (step S340). The trained model W is stored in the noise prediction parameter p shown in FIG. 1.

[0029] [Modification of active noise control device] The active noise control device 1 described above may be configured as a modified example such as an active noise control device 1′ shown in FIG. 5. The active noise control device 1′ according to this modified example differs from the active noise control device 1 described above in the following ways: In the active noise control device 1′, the reference microphone 10 is no longer an essential component. Accordingly, the noise prediction unit 20 has been changed to a noise prediction unit 21. The secondary path model 50 has been changed to a first secondary path model 50A. A second secondary path model 50B having the same function as the first secondary path model 50A has been added. In addition, a reference sound generation unit 80 has been newly added. The active noise control method according to this modified example is realized by the active noise control device 1′ performing the processing of the steps shown in FIG. 6.

[0030] 5 and 6, the processing procedure of the active noise control method executed by the active noise control device 1' will be described, focusing on the differences from the active noise control device 1. Note that step S50A performed by the first secondary path model 50A is equivalent to step S50 performed by the secondary path model 50 described above, and therefore a description thereof will be omitted.

[0031] (Noise Prediction Section 21) The noise prediction unit 21 predicts noise that propagates through the primary path P and reaches a specific position Z from noise received as input (noise x'''(n) as a reference sound, which will be described later). That is, the noise prediction unit 21 receives noise x'''(n) from a reference sound generation unit 80, which will be described later, as input, and predicts noise d(n) from a noise source N that propagates through the primary path P and reaches the specific position Z, and outputs the predicted noise x'(n) to the noise control filter 30 and the secondary path model 50 (step S21). The noise prediction unit 20 may be configured using linear convolution, for example, or may be configured using a trained network (trained model) obtained by training using neural network technology, as will be described later.

[0032] (Second secondary pathway model 50B) The second secondary path model 50B estimates the cancellation sound y(n) generated by the noise control filter 30, which propagates through the secondary path S and arrives at the specific position Z. That is, in order to compensate for the influence of the secondary path S, the second secondary path model 50B estimates the noise that arrives at the specific position Z when the cancellation sound y(n) propagates through the secondary path S. Push the sound The noise d''(n) that is the estimation result is output to the reference sound generation unit 80 (secondary path S50B).

[0033] (Reference sound generation unit 80) The reference sound generation unit 80 receives as input the noise d''(n) estimated by the second secondary path model 50B and the interference sound e(n) detected by the error microphone 60, and generates noise x'''(n) that the noise prediction unit 21 accepts as input. Here, the noise x'''(n) is noise made up of noise d''(n), noise d(n), and noise d'(n). The reference sound generation unit 80 outputs this noise x'''(n) to the noise prediction unit 21 as a reference sound (equivalent to the noise x(n) in the active noise control device 1). The noise prediction unit 21 executes the processing of the noise prediction unit 21 using this reference sound (noise x'''(n)) as the noise accepted as input.

[0034] A modification of this embodiment has been described above. The active noise control device 1′ in this modification does not have the reference microphone 10. While this allows the overall device to be smaller in size than the active noise control device 1, it results in a delay between the reference sound (noise x′″(n)) input to the noise prediction unit 21 of the active noise control device 1′ and the noise x(n) input to the noise prediction unit 20 of the active noise control device 1. However, the active noise control device 1′ not only compensates for the influence of the secondary path S in the application algorithm (first secondary path model 50A), but also uses the secondary path model 50 to generate the reference sound (second secondary path model 50B). For example, by employing a trained model that has been trained in advance to take this delay into account when constructing the trained model for the noise prediction unit 21, it is expected that the noise attenuation effect at a specific location will be comparable to that of the active noise control device 1.

[0035] In this modification, the second secondary path model 50B and the reference sound generation unit 80 are added, but it is also possible to provide the error microphone 60 with a function for switching directivity without adding these elements. The error microphone can be configured to acquire the noise d(n) that has arrived from the noise source N via the primary path P, and output this noise d(n) to the noise prediction unit 21 after performing appropriate amplification processing and analog-to-digital conversion processing. If the noise prediction unit 21 employs neural network technology, it is necessary to prepare a trained model that has been trained in advance to match the above-mentioned configuration.

[0036] Although the embodiments and modifications of this disclosure have been described above, the specific configurations are not limited to these embodiments and modifications, and it goes without saying that appropriate design changes, etc., are included in this disclosure as long as they do not deviate from the spirit of this disclosure. The various processes described in the embodiments and modifications may not only be executed chronologically in the order described, but may also be executed in parallel or individually depending on the processing capacity of the device executing the processes or as needed.

[0037] [Programs, recording media] The various processes described above can be implemented by loading a program that executes each step of the above method into the recording unit 2020 of the computer 2000 shown in Figure 7, and operating the control unit 2010, input unit 2030, output unit 2040, display unit 2050, etc.

[0038] The program describing the processing contents can be recorded on a computer-readable recording medium, which may be, for example, a magnetic recording device, an optical disk, a magneto-optical recording medium, a semiconductor memory, or any other suitable recording medium.

[0039] The program may be distributed, for example, by selling, transferring, lending, etc. a portable recording medium such as a DVD or CD-ROM on which the program is recorded. Furthermore, the program may be stored in a storage device of a server computer, and then transferred from the server computer to another computer via a network, thereby distributing the program.

[0040] A computer that executes such a program may first temporarily store the program recorded on a portable recording medium or transferred from a server computer in its own storage device. Then, when executing a process, the computer reads the program stored on its own recording medium and executes the process in accordance with the read program. Alternatively, the computer may read the program directly from a portable recording medium and execute the process in accordance with the program. Furthermore, the computer may execute the process in accordance with the received program each time a program is transferred from a server computer to the computer. Alternatively, the server computer may not transfer the program to the computer, but may execute the process through a so-called ASP (Application Service Provider) service, which realizes the processing function by issuing an execution instruction and obtaining the results. In this embodiment, the program includes information used for processing by a computer that is equivalent to a program (such as data that is not a direct instruction to the computer but has properties that define computer processing).

[0041] Furthermore, in this embodiment, the device is configured by executing a predetermined program on a computer, but at least a part of the processing contents may be realized by hardware.

Claims

1. An active noise control method executed by an active noise control device, comprising: A reference microphone detects the noise from the noise source; a noise prediction unit predicts the detected noise propagating along a primary path from the noise source to a specific position and reaching the specific position; a noise control filter that generates a cancellation sound for noise control using the predicted noise and a predetermined coefficient; a secondary sound source radiating the generated cancellation sound; a secondary path model estimating the predicted noise traveling a secondary path from the secondary sound source to the specific location and reaching the specific location; an error microphone arranged at the specific position detects an interference sound between the noise from the noise source that has propagated through the primary path and reached the specific position, and the radiated cancellation sound that has propagated through the secondary path and reached the specific position; a coefficient updating unit receiving the estimated noise and the detected interference sound as inputs and updating the predetermined coefficient used by the noise control filter; Active noise control methods.

2. 2. The active noise control method according to claim 1, wherein the noise prediction unit is configured by linear convolution.

3. 2. The active noise control method according to claim 1, wherein the noise prediction unit is configured with a trained model trained by a neural network.

4. the trained model is trained by a noise prediction learning method performed by a noise prediction learning device, The noise prediction learning method by the noise prediction learning device includes: The parameter storage unit outputs the stored noise prediction parameters, a prediction model unit that uses the output noise prediction parameters to predict a received signal sequence that propagates through the primary path and arrives at a specific location; an objective function calculation unit calculates an objective function using the predicted signal sequence and a signal sequence that is correct data as inputs; If the parameter update unit determines that the calculated objective function does not satisfy a predetermined condition, it updates the noise prediction parameters stored in the parameter storage unit, and if it determines that the calculated objective function satisfies the predetermined condition, it outputs a prediction model unit having the current noise prediction parameters as the trained model.

4. The active noise control method according to claim 3.

5. An active noise control method executed by an active noise control device, comprising: a noise prediction unit predicts, from the noise received as input, noise that propagates along a primary path from a noise source to a specific position and reaches the specific position; a noise control filter that generates a cancellation sound for noise control using the predicted noise and a predetermined coefficient; a secondary sound source radiating the generated cancellation sound; a first secondary path model estimating the predicted noise propagating along a secondary path from the secondary sound source to the specific location and arriving at the specific location; an error microphone arranged at the specific position detects an interference sound between the noise from the noise source that has propagated through the primary path and reached the specific position, and the radiated cancellation sound that has propagated through the secondary path and reached the specific position; a coefficient updating unit that receives the estimated noise and the detected interference sound as input and updates the predetermined coefficient used by the noise control filter; a second secondary path model that estimates noise that the generated cancellation sound propagates through the secondary path and reaches the specific location; a reference sound generation unit receives the noise estimated by the second secondary path model and the interference sound detected by the error microphone as inputs and generates the noise to be received as an input by the noise prediction unit; Active noise control methods.

6. a reference microphone for detecting the noise of the noise source; a noise prediction unit that predicts the detected noise that propagates along a primary path from the noise source to a specific position and reaches the specific position; a noise control filter that generates a cancellation sound for noise control using the predicted noise and predetermined coefficients; a secondary sound source that radiates the generated cancellation sound; a secondary path model that estimates the predicted noise propagating along a secondary path from the secondary sound source to the specific location and arriving at the specific location; an error microphone disposed at the specific position and configured to detect an interference sound between the noise from the noise source that has propagated through the primary path and reached the specific position, and the radiated cancellation sound that has propagated through the secondary path and reached the specific position; a coefficient updating unit that receives the estimated noise and the detected interference sound as inputs and updates the predetermined coefficients used by the noise control filter; An active noise control device having:

7. A program for causing a computer to execute the active noise control method according to any one of claims 1 to 5.

Images