Signal processing device, signal processing method, and non-transitory recording medium

Abstract

A signal processing device according to one embodiment of the present invention includes: a filter circuit capable of generating a second signal by performing filter processing on a first signal using a filter coefficient; a first Fourier transform circuit capable of generating a first spectrum signal by performing a discrete Fourier transform on the basis of the first signal; a subtraction circuit capable of generating a fourth signal by performing subtraction processing on the basis of the second signal and a third signal supplied from a sensor and corresponding to the first signal; a second Fourier transform circuit capable of generating a second spectrum signal by performing a discrete Fourier transform on the basis of the fourth signal; a spectrum subtraction processing circuit capable of generating a third spectrum signal by performing spectrum subtraction processing using a noise spectrum signal indicated by a noise model prepared in advance on the basis of the second spectrum signal; and a control unit capable of calculating the filter coefficient on the basis of the first spectrum signal and the third spectrum signal.

Description

Signal processing device, signal processing method, and non-transitory recording medium

[0001] The present invention relates to a signal processing device and a signal processing method for reducing stationary noise, and a non-transitory recording medium on which software for reducing stationary noise is recorded.

[0002] Some signal processing devices perform processing based on the output signal of a sensor. For example, Patent Document 1 discloses a technique for calculating an error signal by subtracting noise contained in the output signal of a sensor.

[0003] Japanese Patent Application Laid-Open No. 2020-10803

[0004] It is desirable for a signal processing device to be able to reduce stationary noise, and more effective reduction of stationary noise is expected.

[0005] It is desirable to provide a signal processing device, a signal processing method, and a non-transitory recording medium that can effectively reduce stationary noise.

[0006] A signal processing device according to an embodiment of the present invention includes a filter circuit, a first Fourier transform circuit, a subtraction circuit, a second Fourier transform circuit, a spectral subtraction processing circuit, and a control unit. The filter circuit is capable of generating a second signal by performing filtering on a first signal using filter coefficients. The first Fourier transform circuit is capable of generating a first spectral signal by performing a discrete Fourier transform on the first signal. The subtraction circuit is capable of generating a fourth signal by performing subtraction on the second signal and a third signal supplied from a sensor and corresponding to the first signal. The second Fourier transform circuit is capable of generating the second spectral signal by performing a discrete Fourier transform on the fourth signal. The spectral subtraction processing circuit is capable of generating a third spectral signal by performing spectral subtraction on the second spectral signal using a noise spectral signal represented by a previously prepared noise model. The control unit is capable of calculating filter coefficients based on the first spectral signal and the third spectral signal so as to reduce the signal component of the third spectral signal.

[0007] A signal processing method according to an embodiment of the present invention includes generating a second signal by filtering a first signal using a filter coefficient; generating a first spectral signal by performing a discrete Fourier transform on the first signal; generating a fourth signal by performing a subtraction process on the second signal and a third signal supplied from a sensor that corresponds to the first signal; generating a second spectral signal by performing a discrete Fourier transform on the fourth signal; generating a third spectral signal by performing a spectral subtraction process on the second spectral signal using a noise spectral signal indicated by a previously prepared noise model; and calculating a filter coefficient on the first spectral signal and the third spectral signal so as to reduce a signal component of the third spectral signal.

[0008] A non-transitory recording medium according to an embodiment of the present invention has recorded thereon software that causes a processor to perform the following operations: generating a second signal by performing a filter process on a first signal using a filter coefficient; generating a first spectral signal by performing a discrete Fourier transform based on the first signal; generating a fourth signal by performing a subtraction process on the second signal and a third signal supplied from a sensor that corresponds to the first signal; generating a second spectral signal by performing a discrete Fourier transform on the fourth signal; generating a third spectral signal by performing a spectral subtraction process on the second spectral signal using a noise spectral signal indicated by a previously prepared noise model; and calculating a filter coefficient based on the first spectral signal and the third spectral signal so as to reduce a signal component of the third spectral signal.

[0009] According to a signal processing device, a signal processing method, and a non-transitory recording medium according to an embodiment of the present invention, stationary noise can be effectively reduced.

[0010] FIG. 2 is a block diagram showing an example configuration of a signal processing device according to an embodiment of the present invention. FIG. 3 is an explanatory diagram showing an example operation of a Fourier transform unit shown in FIG. 1. FIG. 4 is an explanatory diagram showing an example operation of a spectral subtraction processing unit shown in FIG. 1. FIG. 5 is an explanatory diagram showing an operating state of the signal processing device shown in FIG. 1. FIG. 6 is an explanatory diagram showing an operating example of a noise model generation unit shown in FIG. 1. FIG. 7 is an explanatory diagram showing another operating example of the noise model generation unit shown in FIG. 1. FIG. 8 is an explanatory diagram showing another operating state of the signal processing device shown in FIG. 1. FIG. 9 is an explanatory diagram showing an example of spectral subtraction processing in a signal processing device according to a reference example. FIG. 10 is an explanatory diagram showing an example of spectral subtraction processing in the signal processing device shown in FIG.

[0011] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0012] <Embodiment> [Configuration Example] Fig. 1 shows a configuration example of a signal processing device (signal processing device 1) according to an embodiment of the present invention. This signal processing device 1 is configured to detect the transfer characteristics of an acoustic signal in a system B including a certain space A, for example. This signal processing device 1 includes a signal generation unit 11, a DA (Digital to Analog) conversion unit 12, a speaker 13, a microphone 14, an AD (Analog to Digital) conversion unit 15, a Fourier transform unit 21, an adaptive filter 22, a subtraction unit 23, a Fourier transform unit 24, a spectral subtraction processing unit 25, an adaptive algorithm calculation unit 26, an inverse Fourier transform unit 27, a noise model generation unit 31, a noise model selection unit 32, and a user interface 33. The signal processing device 1 is configured to include, for example, a processor and a memory, and is configured to perform processing by executing a program.

[0013] The signal generating unit 11 is configured to generate a signal S11, which is audio data. The signal generating unit 11 sequentially generates data x at a predetermined processing frequency f (e.g., 16 kHz). This processing frequency f is a frequency that is sufficiently higher than the frequency range of the acoustic signal handled by the signal processing device 1. Data x(n) shown in FIG. 1 indicates the nth data x. The signal generating unit 11 then outputs this series of data x as a signal S11.

[0014] The DA conversion unit 12 is configured to perform DA conversion based on the signal S11 supplied from the signal generation unit 11, thereby generating an analog signal.

[0015] The speaker 13 is configured to convert the analog signal supplied from the DA converter 12 into sound. The sound emitted from the speaker 13 propagates through the space A and reaches the microphone 14.

[0016] The microphone 14 is configured to convert the sound emitted from the speaker 13 and propagated through the space A into an analog signal, which is an electrical signal. In this example, the microphone 14 is configured using MEMS (Micro Electro Mechanical Systems) technology. In this example, the microphone 14 is formed integrally with a semiconductor chip of the signal processing device 1.

[0017] Since the microphone 14 may generate stationary noise, the output signal of the microphone 14 contains this stationary noise component in addition to the sound component propagated through the space A. By reducing this stationary noise, the signal processing device 1 can detect with higher accuracy the transfer characteristics of the acoustic signal in the system B including the space A, for example, and can speed up the convergence speed of the adaptive algorithm calculation described below.

[0018] The AD conversion unit 15 is configured to generate a signal S15 by performing AD conversion based on the analog signal supplied from the microphone 14. The AD conversion unit 15 performs AD conversion at a processing frequency f to sequentially generate data, and outputs this series of data as the signal S15.

[0019] The Fourier transform unit 21 is configured to perform a discrete Fourier transform using a fast Fourier transform algorithm based on the signal S11 supplied from the signal generating unit 11, thereby generating a spectrum signal SS21.

[0020] FIG. 2 illustrates an example of the operation of the Fourier transform unit 21. Each time a predetermined number (e.g., 512) of data x is supplied via the signal S11, the Fourier transform unit 21 generates spectral data X by performing a discrete Fourier transform on the data x. The spectral data X includes amplitude spectral data and phase spectral data. Note that FIG. 2 illustrates the amplitude spectral data, but omits the phase spectral data. In the amplitude spectral data illustrated in FIG. 2, the horizontal axis represents frequency and the vertical axis represents amplitude. During a processing period T, the Fourier transform unit 21 receives a predetermined number (e.g., 512) of data x from the signal generating unit 11. During each processing period T, the Fourier transform unit 21 generates spectral data X by performing a discrete Fourier transform on the data x. Spectral data X(k) illustrated in FIG. 1 indicates the kth spectral data X. The Fourier transform unit 21 then outputs this series of spectral data X as a spectral signal SS21.

[0021] The adaptive filter 22 ( FIG. 1 ) is configured to generate a signal S22 by performing filtering on the signal S11 supplied from the signal generating unit 11 using filter coefficients supplied from the inverse Fourier transform unit 27. The adaptive filter 22 is an FIR (Finite Impulse Response) filter, and performs a convolution operation on the signal S11 using the filter coefficients supplied from the inverse Fourier transform unit 27. As a result, the adaptive filter 22 adjusts the amplitude and phase of the signal S11 and outputs the amplitude- and phase-adjusted signal as the signal S22. As will be described later, the filter coefficients of the adaptive filter 22 when the negative feedback operation has converged correspond to the transfer characteristics of the acoustic signal in the system B including the space A. The system B includes the DA converter 12, the speaker 13, the space A, the microphone 14, and the AD converter 15.

[0022] The subtraction unit 23 is configured to generate a signal S23 by performing subtraction processing on the signal S15 supplied from the AD conversion unit 15 and the signal S22 supplied from the adaptive filter 22. The subtraction unit 23 is sequentially supplied with the data contained in the signal S15 from the AD conversion unit 15 and sequentially supplied with the data contained in the signal S22 from the adaptive filter 22. The subtraction unit 23 subtracts the data contained in the signal S22 from the data contained in the signal S15 to sequentially generate data m, and outputs this series of data m as the signal S23. Data m(n) shown in FIG. 1 indicates the nth data m.

[0023] This signal S23 is a so-called error signal when the signal processing device 1 performs a negative feedback operation. Data m(n) is expressed using the following equation EQ1. m(n) is error signal data. e(n) is error signal data when it is assumed that no noise occurs in the microphone 14. no(n) is data corresponding to noise occurring in the microphone 14. In this way, data m(n) includes data no(n) corresponding to noise occurring in the microphone 14.

[0024] The Fourier transform unit 24 is configured to generate a spectrum signal SS24 by performing a discrete Fourier transform using a fast Fourier transform algorithm on the signal S23 supplied from the subtraction unit 23. The operation of the Fourier transform unit 24 is similar to that of the Fourier transform unit 21. That is, for example, each time a predetermined number (e.g., 512) of data m is supplied via the signal S23, the Fourier transform unit 24 performs a discrete Fourier transform on the data m to generate spectrum data M. The spectrum data M includes amplitude spectrum data and phase spectrum data. The spectrum data M(k) shown in FIG. 1 indicates the kth spectrum data M. The Fourier transform unit 21 then outputs this series of spectrum data M as the spectrum signal SS24.

[0025] The spectrum data M(k) is expressed using the following equation EQ2. E1(k) is the spectrum data of data e(n). That is, the spectrum data E1(k) is the spectrum data of the error signal when it is assumed that no noise occurs in the microphone 14. NO(k) is the spectrum data of data no(n). That is, the spectrum data NO(k) is the spectrum data of noise occurring in the microphone 14.

[0026] The spectral subtraction processing unit 25 is configured to generate a spectrum signal SS25 by performing spectral subtraction processing on the spectrum signal SS24 supplied from the Fourier transform unit 24 and the noise spectrum signal SN32 related to the noise model supplied from the noise model selection unit 32. The noise spectrum signal SN32 includes amplitude spectrum data |NM|. The amplitude spectrum data |NM(k)| shown in FIG. 1 indicates the kth amplitude spectrum data |NM|. This amplitude spectrum data |NM| is amplitude spectrum data indicating noise generated in the microphone 14. This amplitude spectrum data |NM| is generated in advance by measuring the characteristics of the microphone 14 and is stored in the noise model selection unit 32.

[0027] 3 shows an example of the operation of the spectral subtraction process, where (A) shows the amplitude spectral data |M(k)| contained in the spectral data M(k) supplied from the Fourier transform unit 24, (B) shows the amplitude spectral data |NM(k)| supplied from the noise model selection unit 32, and (C) shows the amplitude spectral data |E2(k)| contained in the spectral data E2(k) resulting from the spectral subtraction process. In this Figure 3, the horizontal axis represents frequency and the vertical axis represents amplitude. In this Figure 3, the dotted and shaded areas represent error signal components, and the diagonal-shaded areas represent noise components of the microphone 14.

[0028] The spectral subtraction processing unit 25 performs spectral subtraction processing based on the spectral data M(k) ( FIG. 3A ) supplied from the Fourier transform unit 24 and the amplitude spectral data |NM(k)| ( FIG. 3B ) supplied from the noise model selection unit 32. As a result, the spectral subtraction processing unit 25 generates spectral data E2(k) ( FIG. 3C ) in which noise components have been reduced. Note that the amplitude spectral data |NM(k)| ( FIG. 3B ) is generated by measuring the characteristics of the microphone 14 in advance. Therefore, the noise components indicated by this amplitude spectral data |NM(k)| are similar to, but not identical to, the noise components contained in the spectral data M(k) ( FIG. 3A ). Therefore, the generated spectral data E2 may contain components corresponding to the differences in these noise components in addition to the error signal components, as shown in FIG. 3C .

[0029] The spectral subtraction process can use various calculation methods, such as the spectral subtraction method, the MAP (maximum a posteriori) estimation method, and the MMSE-ETSA (minimum mean-square-error short-time spectral amplitude) estimation method. The spectral subtraction processing unit 25 can select the calculation method to be used based on, for example, an instruction from the user interface 33. For example, when the spectral subtraction method is used, the spectral data E2(k) is expressed by, for example, the following equations EQ3 and EQ4. In these equations EQ3 and EQ4, the part before the multiplication sign is the amplitude term, and the part after the multiplication sign is the phase term.

[0030] The spectral subtraction processing unit 25 is sequentially supplied with the spectral data M and the amplitude spectral data |NM| related to the noise model. The spectral subtraction processing unit 25 performs spectral subtraction processing to sequentially generate spectral data E2. The spectral subtraction processing unit 25 then outputs this series of spectral data E2 as a spectral signal SS25.

[0031] The adaptive algorithm calculation unit 26 is configured to perform adaptive algorithm calculation based on the spectrum signal SS21 supplied from the Fourier transform unit 21 and the spectrum signal SS25 supplied from the spectral subtraction processing unit 25. Specifically, the adaptive algorithm calculation unit 26 sets the filter coefficients of the adaptive filter 22 via the inverse Fourier transform unit 27 based on the spectrum signal SS21 and the spectrum signal SS25 so that the spectral data E2 included in the spectrum signal SS25 is reduced. In other words, the signal processing device 1 performs negative feedback operation so as to reduce the spectral data E2. The adaptive algorithm calculation unit 26 sequentially generates spectral data of the filter coefficients by performing adaptive algorithm calculation. This spectral data includes amplitude spectral data and phase spectral data. The adaptive algorithm calculation unit 26 outputs this series of spectral data as the spectrum signal SS26.

[0032] The inverse Fourier transform unit 27 is configured to generate a signal S27 by performing an inverse discrete Fourier transform based on the spectrum signal SS26 supplied from the adaptive algorithm calculation unit 26. Spectral data of the filter coefficients is sequentially supplied to the inverse Fourier transform unit 27. The inverse Fourier transform unit 27 performs an inverse discrete Fourier transform based on the spectral data of the filter coefficients to sequentially generate filter coefficients. The inverse Fourier transform unit 27 then supplies these filter coefficients to the adaptive filter 22.

[0033] The noise model generation unit 31 is configured to generate a noise model for noise occurring in the microphone 14. The noise model generation unit 31 generates the noise model before the signal processing device 1 performs normal processing operations. The noise model is, for example, amplitude spectrum data |NM| of the noise.

[0034] When the signal processing device 1 generates a noise model, the signal processing device 1 is placed, for example, in a quiet space with almost no surrounding noise. The signal processing device 1 is configured to operate six blocks: the microphone 14, the AD conversion unit 15, the subtraction unit 23, the Fourier transform unit 24, the noise model generation unit 31, and the noise model selection unit 32. Under these conditions, the data e(n) related to the error signal and the spectral data E1(k) related to the error signal are zero. Therefore, the noise model generation unit 31 is able to generate amplitude spectral data |NM| for the noise occurring at the microphone 14, based on the spectral signal SS24 including the spectral data N0.

[0035] The noise model selection unit 32 is configured to select a noise model based on an instruction from the user interface 33. The noise model selection unit 32 stores the amplitude spectrum data |NM| generated by the noise model generation unit 31. The noise model selection unit 32 can store multiple pieces of amplitude spectrum data |NM|. Based on an instruction from the user interface 33, the noise model selection unit 32 selects one of the multiple pieces of amplitude spectrum data |NM| and outputs the selected amplitude spectrum data |NM| as the noise spectrum signal SN32.

[0036] The user interface 33 is configured to accept operations by a user who uses this signal processing device 1. The user interface 33 is configured using, for example, a liquid crystal display device, various indicators, a touch panel, buttons, etc. In this example, the user interface 33 is configured to accept an operation by the user to select a noise model and an operation by the user to select a calculation method for the spectral subtraction processing.

[0037] Here, the adaptive filter 22 corresponds to a specific example of a "filter circuit" in an embodiment of the present disclosure. The Fourier transform unit 21 corresponds to a specific example of a "first Fourier transform circuit" in an embodiment of the present disclosure. The subtraction unit 23 corresponds to a specific example of a "subtraction circuit" in an embodiment of the present disclosure. The Fourier transform unit 24 corresponds to a specific example of a "second Fourier transform circuit" in an embodiment of the present disclosure. The spectral subtraction processing unit 25 corresponds to a specific example of a "spectral subtraction processing circuit" in an embodiment of the present disclosure. The adaptive algorithm calculation unit 26 and the inverse Fourier transform unit 27 correspond to specific examples of a "control unit" in an embodiment of the present disclosure. The noise model generation unit 31 corresponds to a specific example of a "noise model generation circuit" in an embodiment of the present disclosure. The noise model selection unit 32 corresponds to a specific example of a "noise model selection circuit" in an embodiment of the present disclosure. The user interface 33 corresponds to a specific example of a "user interface" in an embodiment of the present disclosure. The signal S11 corresponds to a specific example of a "first signal" in an embodiment of the present disclosure. The signal S22 corresponds to a specific example of a "second signal" in an embodiment of the present disclosure. The spectrum signal SS21 corresponds to a specific example of a "first spectrum signal" in an embodiment of the present disclosure. The signal S15 corresponds to a specific example of a "third signal" in an embodiment of the present disclosure. The signal S23 corresponds to a specific example of a "fourth signal" in an embodiment of the present disclosure. The spectrum signal SS24 corresponds to a specific example of a "second spectrum signal" in an embodiment of the present disclosure. The spectrum signal SS25 corresponds to a specific example of a "third spectrum signal" in an embodiment of the present disclosure. The noise spectrum signal SN32 corresponds to a specific example of a "noise spectrum signal" in an embodiment of the present disclosure.

[0038] [Operation and Function] Next, the operation and function of the signal processing device 1 of this embodiment will be described.

[0039] (Overall Operation Overview) First, an overview of the overall operation of the signal processing device 1 will be described with reference to Fig. 1. First, the signal generation unit 11 generates a signal S11, which is audio data. The DA conversion unit 12 generates an analog signal by performing DA conversion based on the signal S11. The speaker 13 converts the analog signal supplied from the DA conversion unit 12 into sound. The sound emitted from the speaker 13 propagates through space A and reaches the microphone 14. The microphone 14 converts the sound emitted from the speaker 13 and propagated through space A into an analog signal, which is an electrical signal. The AD conversion unit 15 generates a signal S15 by performing AD conversion based on the analog signal supplied from the microphone 14.

[0040] The Fourier transform unit 21 generates a spectrum signal SS21 by performing a discrete Fourier transform on the signal S11. The adaptive filter 22 generates a signal S22 by performing a filter process on the signal S11 using filter coefficients supplied from the inverse Fourier transform unit 27. The subtraction unit 23 generates a signal S23 by performing a subtraction process on the signal S15 supplied from the AD conversion unit 15 and the signal S22 supplied from the adaptive filter 22. The Fourier transform unit 24 generates a spectrum signal SS24 by performing a discrete Fourier transform on the signal S23. The spectral subtraction processing unit 25 generates a spectrum signal SS25 by performing a spectral subtraction process on the spectrum signal SS24 supplied from the Fourier transform unit 24 and the noise spectrum signal SN32 related to the noise model supplied from the noise model selection unit 32. The adaptive algorithm calculation unit 26 performs adaptive algorithm calculation based on the spectrum signal SS21 supplied from the Fourier transform unit 21 and the spectrum signal SS25 supplied from the spectral subtraction processing unit 25. By performing this adaptive algorithm calculation, the adaptive algorithm calculation unit 26 generates a spectrum signal SS26. The inverse Fourier transform unit 27 performs an inverse discrete Fourier transform based on the spectrum signal SS26 supplied from the adaptive algorithm calculation unit 26, thereby generating a signal S27. The inverse Fourier transform unit 27 then supplies the filter coefficients included in the signal S27 to the adaptive filter 22.

[0041] The noise model generation unit 31 generates a noise model (amplitude spectrum data |NM|) for noise generated at the microphone 14. The noise model selection unit 32 stores a plurality of noise models and selects one of these plurality of noise models based on an instruction from the user interface 33. The noise model selection unit 32 then supplies the selected noise model to the spectrum subtraction processing unit 25 as a noise spectrum signal SN32. The user interface 33 accepts operations by the user of this signal processing device 1.

[0042] (Detailed Operation) In a normal processing operation, the signal processing device 1 detects the transfer characteristics of an acoustic signal in a system B including a space A. Before performing the normal processing operation, the signal processing device 1 first generates a noise model (amplitude spectrum data |NM|) for noise generated in the microphone 14.

[0043] Fig. 4 shows an example of the operating state of the signal processing device 1 when the signal processing device 1 generates a noise model. In Fig. 4, blocks drawn with solid lines are blocks that operate when the signal processing device 1 generates a noise model, and blocks drawn with dashed lines are blocks that do not operate when the signal processing device 1 generates a noise model.

[0044] When the signal processing device 1 generates a noise model, the signal processing device 1 is placed, for example, in a quiet space with almost no surrounding noise. In this example, the signal processing device 1 is set to operate six blocks: a microphone 14, an AD conversion unit 15, a subtraction unit 23, a Fourier transform unit 24, a noise model generation unit 31, and a noise model selection unit 32.

[0045] The microphone 14 converts sound into an analog signal, which is an electrical signal. Since the signal processing device 1 is placed in a quiet space, most of the output signal of the microphone 14 is noise components generated by the microphone 14. The AD conversion unit 15 generates a signal S15 by performing AD conversion based on the analog signal supplied from the microphone 14. Since the adaptive filter 22 is not operating, the subtraction unit 23 outputs the signal S15 as is as a signal S23. Under these conditions, the data e(n) related to the error signal is zero, and therefore the data m(n) included in this signal S23 is expressed using the following equation EQ5. This data no(k) is data on noise generated at microphone 14. Fourier transform unit 24 generates spectrum signal SS24 by performing a discrete Fourier transform on signal S23 containing this series of data m. Under these conditions, spectrum data E1(k) related to the error signal is zero, and therefore spectrum data M(k) contained in this spectrum signal SS24 is expressed using the following equation EQ6. The spectrum data NO(k) is spectrum data of noise generated at the microphone 14. In this way, when the signal processing device 1 generates a noise model, the signal S15, the signal S23, and the spectrum signal SS24 are signals related to noise generated at the microphone 14 and are substantially equivalent to each other. The noise model generation unit 31 generates a noise model (amplitude spectrum data |NM|) for the noise generated at the microphone 14 based on the series of spectrum data NO.

[0046] 5A and 5B show an example of the operation of the noise model generating unit 31. FIG.

[0047] 5A , the noise model generation unit 31 can generate a noise model (amplitude spectrum data |N|) by multiplying one piece of amplitude spectrum data |N| included in one piece of spectrum data N| among a series of spectrum data N| by a subtraction coefficient α. Specifically, the noise model generation unit 31 can generate a noise model by multiplying the magnitude of the amplitude at each frequency in the amplitude spectrum data |N| by, for example, 0.7. This is not a limitation, and the noise model generation unit 31 may also generate a noise model by multiplying the magnitude of the amplitude at each frequency in the amplitude spectrum data |N| by, for example, 1.2.

[0048] 5B , the noise model generation unit 31 can generate a noise model (amplitude spectral data |NM|) by calculating the average value of two amplitude spectral data |NO| included in two spectral data sets out of the series of spectral data NO. However, the present invention is not limited to this, and the noise model generation unit 31 may generate a noise model by, for example, calculating the average value of three amplitude spectral data |NO| included in three spectral data sets out of the series of spectral data NO.

[0049] Furthermore, for example, the example of FIG. 5A and the example of FIG. 5B may be combined, and the average value of the two amplitude spectrum data may be multiplied by a subtraction coefficient.

[0050] In this way, the noise model generation unit 31 generates a plurality of noise models (amplitude spectrum data |NM|) based on the series of spectrum data NO. The noise model generation unit 31 then supplies these noise models to the noise model selection unit 32. The noise model selection unit 32 stores these noise models.

[0051] This allows the signal processing device 1 to detect the transfer characteristics of an acoustic signal in a system B including a space A using these noise models.

[0052] Fig. 6 shows an example of the operating state of the signal processing device 1 when the signal processing device 1 detects the transfer characteristics. In Fig. 6, blocks drawn with solid lines are blocks that operate when the signal processing device 1 detects the transfer characteristics, and blocks drawn with dashed lines are blocks that do not operate when detecting the transfer characteristics.

[0053] In this operation, the signal processing device 1 is placed in a space A in which the transfer characteristics of an acoustic signal are to be detected. In this example, the signal processing device 1 is set so that all blocks other than the noise model generation unit 31 are operational.

[0054] The user operates the user interface 33 to select a noise model and a calculation method for the spectral subtraction process. The user interface 33 accepts these user operations. The noise model selection unit 32 selects one of a plurality of noise models (amplitude spectrum data |NM|) based on an instruction from the user interface 33, and outputs the selected amplitude spectrum data |NM| as a noise spectrum signal SN32. The spectral subtraction processing unit 25 selects a calculation method to use based on an instruction from the user interface 33. As a result, the spectral subtraction processing unit 25 performs the spectral subtraction process using the selected calculation method.

[0055] First, the signal generating unit 11 generates a signal S11, which is audio data. The DA converting unit 12 generates an analog signal by performing DA conversion based on the signal S11. The speaker 13 converts the analog signal supplied from the DA converting unit 12 into sound. The sound emitted from the speaker 13 propagates through space A and reaches the microphone 14. The microphone 14 converts the sound emitted from the speaker 13 and propagated through space A into an analog signal, which is an electrical signal. The AD converting unit 15 generates a signal S15 by performing AD conversion based on the analog signal supplied from the microphone 14.

[0056] Since microphone 14 may generate stationary noise, the output signal of microphone 14 contains this stationary noise component in addition to the sound component propagated through space A. Thus, signal S15 similarly contains the stationary noise component in addition to the sound component propagated through space A.

[0057] The Fourier transform unit 21 generates a spectrum signal SS21 by performing a discrete Fourier transform on the signal S11. The adaptive filter 22 generates a signal S22 by performing a filter process on the signal S11 using filter coefficients supplied from the inverse Fourier transform unit 27. The subtraction unit 23 generates a signal S23 by performing a subtraction process on the signal S15 supplied from the AD conversion unit 15 and the signal S22 supplied from the adaptive filter 22.

[0058] This signal S23 is a so-called error signal when the signal processing device 1 performs negative feedback operation. This signal S23 includes a series of data m. As shown in equation EQ1, the data m(n) includes data e(n) of the error signal when it is assumed that no noise occurs in the microphone 14, and data no(n) corresponding to noise occurring in the microphone 14.

[0059] The Fourier transform unit 24 generates a spectrum signal SS24 by performing a discrete Fourier transform on the signal S23 supplied from the subtraction unit 23. This spectrum signal SS24 includes a series of spectrum data M. As shown in equation EQ2, the spectrum data M(k) includes spectrum data E1(k) of the error signal when it is assumed that no noise occurs in the microphone 14, and spectrum data NO(k) corresponding to noise occurring in the microphone 14.

[0060] 3, the spectral subtraction processing unit 25 generates a spectral signal SS25 by performing spectral subtraction processing based on the spectral signal SS24 supplied from the Fourier transform unit 24 and the noise spectral signal SN32 related to the noise model supplied from the noise model selection unit 32. As shown in FIG. 3, the spectral subtraction processing unit 25 generates spectral data E2(k) (FIG. 3C) in which the noise component has been reduced based on the spectral data M(k) (FIG. 3A) supplied from the Fourier transform unit 24 and the amplitude spectral data |NM(k)| (FIG. 3B) supplied from the noise model selection unit 32. The spectral subtraction processing unit 25 then generates a spectral signal SS25 including a series of spectral data E2(k).

[0061] The adaptive algorithm calculation unit 26 performs adaptive algorithm calculation based on the spectrum signal SS21 supplied from the Fourier transform unit 21 and the spectrum signal SS25 supplied from the spectral subtraction processing unit 25. By performing this adaptive algorithm calculation, the adaptive algorithm calculation unit 26 sets the filter coefficients of the adaptive filter 22 via the inverse Fourier transform unit 27 so that the spectral data E2 included in the spectrum signal SS25 is reduced. That is, the signal processing device 1 performs negative feedback operation so as to reduce the spectral data E2. By performing this adaptive algorithm calculation, the adaptive algorithm calculation unit 26 sequentially generates spectral data of the filter coefficients. This spectral data includes amplitude spectral data and phase spectral data. This amplitude spectral data and phase spectral data are sequentially updated. The amount of change in the amplitude spectral data and phase spectral data is limited to a predetermined step size. The adaptive algorithm calculation unit 26 outputs this series of spectral data as the spectrum signal SS26.

[0062] The adaptive algorithm computing unit 26 performs processing in the frequency domain, generating spectral data of filter coefficients by processing each of a plurality of frequencies. Therefore, the convergence speed may vary depending on the frequency. For example, the adaptive algorithm computing unit 26 may change the step size for each of a plurality of frequencies based on the convergence speed at each of the plurality of frequencies. In this way, the adaptive algorithm computing unit 26 can improve the convergence speed at a frequency with a slow convergence speed by changing the step size at that frequency.

[0063] The inverse Fourier transform unit 27 generates filter coefficients by performing an inverse discrete Fourier transform based on the spectrum data of the filter coefficients included in the spectrum signal SS26 supplied from the adaptive algorithm calculation unit 26. The adaptive filter 22 performs filtering using these filter coefficients.

[0064] The signal processing device 1 performs negative feedback operation so as to reduce the spectral data E2. When the negative feedback operation has converged, the filter coefficients of the adaptive filter 22 correspond to the transfer characteristics of the acoustic signal in the system B including the space A. The user can grasp the transfer characteristics of the acoustic signal in the space A from the filter coefficients.

[0065] As described above, the signal processing device 1 includes a filter circuit (adaptive filter 22) capable of generating a second signal (signal S22) by performing filter processing on a first signal (signal S11) using a filter coefficient, a first Fourier transform circuit (Fourier transform unit 21) capable of generating a first spectrum signal (spectrum signal SS21) by performing Fourier transform based on the first signal (signal S11), a subtraction circuit (subtraction unit 23) capable of generating a fourth signal (signal S23) by performing subtraction processing based on the second signal (signal S22) and a third signal (signal S15) supplied from the sensor (microphone 14) corresponding to the first signal, and a fourth signal (signal S23) by performing Fourier transform based on the fourth signal (signal S23). The signal processing device 1 includes a second Fourier transform circuit (Fourier transform unit 24) capable of generating a noise spectrum signal (SN32) based on the second spectrum signal (spectrum signal SS24), a spectral subtraction processing circuit (spectrum subtraction processing unit 25) capable of performing spectral subtraction processing using a noise spectrum signal (noise spectrum signal SN32) indicated by a previously prepared noise model, to generate a third spectrum signal (spectrum signal SS25), and a control unit (adaptive algorithm calculation unit 26 and inverse Fourier transform unit 27) capable of calculating filter coefficients based on the first spectrum signal (spectrum signal SS21) and the third spectrum signal (spectrum signal SS25) so as to reduce the signal component of the third spectrum signal. In this way, the signal processing device 1 performs adaptive algorithm calculation in the frequency domain, and performs processing individually at each frequency. This allows, for example, an increased convergence speed of the adaptive algorithm calculation. Furthermore, the signal processing device 1 uses fast Fourier transform to reduce the amount of calculation required for the adaptive algorithm calculation. As a result, stationary noise can be effectively reduced. In this way, the signal processing device 1 suppresses the disturbance components contained in the error signal by reducing the stationary noise, and as a result, the signal processing device 1 can increase the convergence speed of the adaptive algorithm calculation and improve the calculation accuracy of the adaptive algorithm calculation.

[0066] Furthermore, in the signal processing device 1, noise components are reduced by performing the spectral subtraction processing in the frequency domain, thereby reducing the amount of calculation. That is, for example, when an adaptive algorithm operation operating in the time domain is used, performing the spectral subtraction processing in the frequency domain requires a Fourier transform, spectral subtraction processing, and an inverse Fourier transform, as shown in FIG. 7 , which increases the amount of calculation. On the other hand, in the signal processing device 1, the adaptive algorithm operation is performed in the frequency domain, and the spectral subtraction processing is performed in the frequency domain, so there is no need to add a Fourier transform and an inverse Fourier transform, as shown in FIG. 8 . That is, in the signal processing device 1, a Fourier transform and an inverse Fourier transform are performed to perform the adaptive algorithm operation in the frequency domain. Therefore, there is no need to add a Fourier transform and an inverse Fourier transform to perform the spectral subtraction processing in the frequency domain. Therefore, the signal processing device 1 can reduce the amount of calculation and effectively reduce stationary noise.

[0067] Furthermore, in the signal processing device 1, noise components are reduced by performing spectral subtraction processing in the frequency domain, thereby reducing the amount of calculation and reducing noise over a wide frequency range. That is, for example, if a high-pass filter is provided in a circuit portion that operates in the time domain, this signal processing device can reduce noise in the low frequency range but cannot reduce noise in the high frequency range. Furthermore, this signal processing device reduces signal components in the low frequency range among the signal components of the acoustic signal that propagated through space A. On the other hand, in the signal processing device 1, noise components are reduced by performing spectral subtraction processing using a noise model in the frequency domain, thereby reducing noise over a wide frequency range and effectively reducing stationary noise.

[0068] Furthermore, the signal processing device 1 further includes a noise model generation circuit (noise model generation unit 31) that can generate a noise model based on the third signal (signal S15). This allows the signal processing device 1 to generate a noise model and perform normal processing operations using a single device, thereby simplifying the configuration and improving user convenience.

[0069] Furthermore, in the signal processing device 1, the noise model generation circuit (noise model generation unit 31) is capable of generating multiple noise models based on the third signal (signal S15) by changing generation parameters (such as the subtraction coefficient α shown in FIG. 5A and the number of spectral data NO shown in FIG. 5B). Note that, as shown in FIG. 4, the noise model generation unit 31 generates multiple noise models based on the spectral signal SS24, but when generating the noise models, the signals S15, S23, and SS24 are equivalent to one another. This allows the signal processing device 1 to use an optimal noise model depending on, for example, the application or various situations, thereby effectively reducing stationary noise.

[0070] The signal processing device 1 also includes a user interface 33 that can accept user operations, and the noise model selection circuit (noise model selection unit 32) can select one of a plurality of noise models based on the user operation. This allows the user to select the optimal noise model depending on, for example, the application or various situations. Therefore, the signal processing device 1 can effectively reduce stationary noise.

[0071] Furthermore, in the signal processing device 1, the spectral subtraction processing circuit (spectral subtraction processing unit 25) can perform spectral subtraction processing using a plurality of spectral subtraction methods, and one of the plurality of spectral subtraction methods can be selected as the spectral subtraction method to be used. This allows the signal processing device 1 to use the optimal spectral subtraction method depending on, for example, the application or various situations, thereby effectively reducing stationary noise.

[0072] The signal processing device 1 further includes a user interface 33 that can accept user operations, and the spectral subtraction processing circuit (spectral subtraction processing unit 25) can select one of multiple spectral subtraction methods as the spectral subtraction method to be used based on the user's operation. This allows the user to select the optimal spectral subtraction method depending on, for example, the application or various situations. Therefore, the signal processing device 1 can effectively reduce stationary noise.

[0073] [Effects] As described above, this embodiment includes a filter circuit capable of generating a second signal by filtering a first signal using a filter coefficient, a first Fourier transform circuit capable of generating a first spectral signal by performing a Fourier transform on the first signal, a subtraction circuit capable of generating a fourth signal by performing a subtraction on the second signal and a third signal supplied from a sensor corresponding to the first signal, a second Fourier transform circuit capable of generating a second spectral signal by performing a Fourier transform on the fourth signal, a spectral subtraction processing circuit capable of generating a third spectral signal by performing spectral subtraction on the second spectral signal using a noise spectral signal represented by a pre-prepared noise model, and a control unit capable of calculating filter coefficients based on the first spectral signal and the third spectral signal so as to reduce the signal component of the third spectral signal. This makes it possible to effectively reduce stationary noise.

[0074] Furthermore, in this embodiment, the noise model generation circuit is capable of generating a plurality of noise models based on the third signal by changing the generation parameters, and therefore, stationary noise can be effectively reduced.

[0075] Furthermore, this embodiment further includes a user interface that can accept user operations, and the noise model selection circuit is capable of selecting one of a plurality of noise models based on the user operation, thereby making it possible to effectively reduce stationary noise.

[0076] Furthermore, in this embodiment, the spectral subtraction processing circuit is capable of performing spectral subtraction processing using a plurality of spectral subtraction methods, and one of the plurality of spectral subtraction methods can be selected as the spectral subtraction method to be used, thereby making it possible to effectively reduce stationary noise.

[0077] In addition, this embodiment further includes a user interface capable of accepting user operations, and the spectral subtraction processing circuit is capable of selecting one of a plurality of spectral subtraction methods as the spectral subtraction method to be used based on the user operation, thereby enabling effective reduction of stationary noise.

[0078] [Variation 1] In the above embodiment, the noise model selection unit 32 selected one of the multiple noise models (amplitude spectrum data |NM|) based on a user operation. However, this is not limited to this. Instead, for example, the noise model selection unit 32 may select one of the multiple noise models (amplitude spectrum data |NM|) by performing a calibration operation. Specifically, the signal processing device 1, for example, detects the transfer characteristics of an acoustic signal in a system B including a space A by sequentially using one of the multiple noise models (amplitude spectrum data |NM|) during a calibration operation. In this calibration operation, the noise model selection unit 32 selects one of the multiple noise models as the noise model to be used based on the convergence characteristics of the negative feedback operation. Specifically, the noise model selection unit 32 can select, for example, the noise model that has the largest reduction value as the noise model to be used. Here, the reduction value is a value indicating the amount of reduction of the error signal based on the ratio between the power of the microphone output signal and the power of the error signal. For example, the smaller the spectrum data E2(k), the larger the reduction value. In the signal processing device 1, by using a noise model that maximizes the reduction value in this way, it is possible to obtain transfer characteristics more accurately.

[0079] In this example, the signal processing device 1 detects the transfer characteristics of the acoustic signal in the system B including the space A by sequentially using one of the multiple noise models (amplitude spectrum data |NM|) in the calibration operation, but this is not limited to this. For example, the signal processing device 1 may select two or more noise models that are part of the multiple noise models (amplitude spectrum data |NM|) in accordance with an application specified by a user operation, and detect the transfer characteristics by sequentially using one of the two or more noise models. Even in this case, the signal processing device 1 may select, for example, the noise model that has the largest reduction value as the noise model to be used.

[0080] [Variation 2] In the above embodiment, the spectral subtraction processing unit 25 selects one of a plurality of calculation methods based on a user operation. However, this is not limiting. Instead, for example, the spectral subtraction processing unit 25 may select one of a plurality of calculation methods by performing a calibration operation. Specifically, the signal processing device 1, for example, detects the transfer characteristics of an acoustic signal in a system B including a space A by sequentially using one of a plurality of calculation methods during the calibration operation. In this calibration operation, the spectral subtraction processing unit 25 selects one of the plurality of calculation methods as the calculation method to be used based on the convergence characteristics of the negative feedback operation. Specifically, in this calibration operation, the spectral subtraction processing unit 25 can select the calculation method that maximizes the reduction value as the calculation method to be used. In the signal processing device 1, by using the calculation method that maximizes the reduction value in this manner, the transfer characteristics can be obtained more accurately.

[0081] In this example, the signal processing device 1 detects the transfer characteristics of the acoustic signal in the system B including the space A by sequentially using one of a plurality of calculation methods in the calibration operation, but this is not limited to this. For example, the signal processing device 1 may select two or more calculation methods that are part of the plurality of calculation methods according to an application specified by a user operation, and detect the transfer characteristics by sequentially using one of the two or more calculation methods. Even in this case, the signal processing device 1 may select, for example, the calculation method that maximizes the reduction value as the calculation method to be used.

[0082] [Variation 3] In the above embodiment, the noise model is described as amplitude spectrum data, but this is not limited to this. Instead, the noise model may be, for example, data representing a power spectrum. In this case, the spectral subtraction processing unit 25 may perform the spectral subtraction processing based on the data representing the power spectrum.

[0083] [Other Modifications] Two or more of these modifications may be combined.

[0084] Although the present invention has been described above by way of embodiments and modifications, the present invention is not limited to these embodiments and can be modified in various ways.

[0085] For example, in the above embodiment, a microphone constructed using MEMS technology is used as the microphone 14, but this is not limited to this, and instead, for example, a condenser microphone may be used as the microphone 14.

[0086] For example, in the above embodiments, the present technology is used to detect the transmission characteristics of an acoustic signal in a system B including a space A, but the present technology is not limited to this and can be used for various purposes.

[0087] For example, in the above-described embodiment, the microphone 14 is used to reduce stationary noise generated in the microphone 14, but this is not limited to this. Instead, a magnetic sensor may be used to reduce stationary noise generated in the magnetic sensor. In this case, the signal processing device 1 may be used in an application that detects biological information using a magnetic sensor, for example. The present technology may also be applied to devices that use various other sensors.

[0088] For example, in the above-described embodiments, the present technology is used to reduce noise generated in the microphone 14, but the present technology is not limited to this and can be used to reduce various stationary noises, such as background noise and environmental noise.

[0089] The effects described in this specification are merely examples, and the effects of the present disclosure are not limited to the effects described in this specification. Therefore, other effects may be obtained with respect to the present disclosure.

[0090] Furthermore, the present disclosure may take the following aspects.

[0091] (1) A signal processing device comprising: a filter circuit capable of generating a second signal by filtering a first signal using a filter coefficient; a first Fourier transform circuit capable of generating a first spectral signal by performing a discrete Fourier transform on the first signal; a subtraction circuit capable of generating a fourth signal by performing subtraction on the second signal and a third signal supplied from a sensor corresponding to the first signal; a second Fourier transform circuit capable of generating a second spectral signal by performing a discrete Fourier transform on the fourth signal; a spectral subtraction processing circuit capable of generating a third spectral signal by performing spectral subtraction on the second spectral signal using a noise spectral signal indicated by a previously prepared noise model; and a control unit capable of calculating the filter coefficient based on the first spectral signal and the third spectral signal so as to reduce a signal component of the third spectral signal. (2) The signal processing device according to (1), further comprising a noise model generation circuit capable of generating the noise model based on the third signal. (3) The signal processing device according to (2), further comprising a noise model selection circuit, wherein the noise model generation circuit is capable of generating a plurality of the noise models based on the third signal by changing a generation parameter, and the noise model selection circuit is capable of selecting one of the plurality of noise models as the noise model to be used. (4) The signal processing device according to (3), further comprising a user interface capable of accepting a user operation, and the noise model selection circuit is capable of selecting one of the plurality of noise models based on the user operation. (5) The signal processing device according to (3), further comprising a calibration mode, and the noise model selection circuit is capable of selecting the noise model to be used based on convergence characteristics when each of the plurality of noise models is used in the calibration mode.(6) The signal processing device according to any one of (1) to (5), wherein the spectral subtraction processing circuit is capable of performing the spectral subtraction processing using a plurality of spectral subtraction methods and is capable of selecting one of the plurality of spectral subtraction methods as the spectral subtraction method to be used. (7) The signal processing device according to (6), further comprising a user interface capable of accepting user operations, wherein the spectral subtraction processing circuit is capable of selecting one of the plurality of spectral subtraction methods as the spectral subtraction method to be used based on the user operation. (8) The signal processing device according to (6), wherein the signal processing device has a calibration mode, wherein the spectral subtraction processing circuit is capable of selecting the spectral subtraction method to be used in the calibration mode based on convergence characteristics when each of the plurality of spectral subtraction methods is used. (9) The signal processing device according to any one of (1) to (8), further comprising the sensor, wherein the sensor is a microphone configured using MEMS technology. (10) The signal processing device according to any one of (1) to (9), wherein the control unit is capable of generating a plurality of coefficients corresponding to a plurality of frequencies, calculating the filter coefficients by performing an inverse discrete Fourier transform based on the plurality of coefficients, and setting amounts of change of the plurality of coefficients based on convergence speeds of the plurality of coefficients.(11) A signal processing method including: generating a second signal by filtering a first signal using a filter coefficient; generating a first spectral signal by performing a discrete Fourier transform based on the first signal; generating a fourth signal by performing a subtraction process based on the second signal and a third signal supplied from a sensor and corresponding to the first signal; generating a second spectral signal by performing a discrete Fourier transform based on the fourth signal; generating a third spectral signal by performing a spectral subtraction process based on the second spectral signal using a noise spectral signal indicated by a previously prepared noise model; and calculating the filter coefficient based on the first spectral signal and the third spectral signal so that a signal component of the third spectral signal is reduced. (12) A non-transitory recording medium having recorded thereon software that causes a processor to perform the following operations: generating a second signal by filtering a first signal using a filter coefficient; generating a first spectral signal by performing a discrete Fourier transform based on the first signal; generating a fourth signal by performing a subtraction process based on the second signal and a third signal supplied from a sensor that corresponds to the first signal; generating a second spectral signal by performing a discrete Fourier transform based on the fourth signal; generating a third spectral signal by performing a spectral subtraction process based on the second spectral signal using a noise spectral signal indicated by a previously prepared noise model; and calculating the filter coefficient based on the first spectral signal and the third spectral signal so that a signal component of the third spectral signal is reduced.

Claims

1. A signal processing apparatus comprising: a filter circuit capable of generating a second signal by performing a filter process on a first signal using a filter coefficient; a first Fourier transform circuit capable of generating a first spectral signal by performing a discrete Fourier transform on the first signal; a third signal supplied from a sensor and corresponding to the first signal; a subtraction circuit capable of generating a fourth signal by performing a subtraction process based on the third signal and the second signal; a second Fourier transform circuit capable of generating a second spectral signal by performing a discrete Fourier transform on the fourth signal; a spectral subtraction process circuit capable of generating a third spectral signal by performing a spectral subtraction process using a noise spectral signal indicated by a noise model prepared in advance based on the second spectral signal; and a control unit capable of calculating the filter coefficient so that a signal component of the third spectral signal becomes small based on the first spectral signal and the third spectral signal.

2. The signal processing apparatus according to claim 1, further comprising a noise model generation circuit capable of generating the noise model based on the third signal.

3. The signal processing apparatus according to claim 2, further comprising a noise model selection circuit, wherein the noise model generation circuit is capable of generating a plurality of the noise models based on the third signal by changing generation parameters, and the noise model selection circuit is capable of selecting one of the plurality of noise models as the noise model to be used.

4. The signal processing apparatus according to claim 3, further comprising a user interface capable of receiving a user operation, wherein the noise model selection circuit is capable of selecting one of the plurality of noise models based on the user operation.

5. The signal processing apparatus has a calibration mode, and the noise model selection circuit according to claim 3 is capable of selecting the noise model to be used based on convergence characteristics when each of the plurality of noise models is used in the calibration mode.

6. The spectrum subtraction processing circuit is capable of performing the spectrum subtraction processing using a plurality of spectrum subtraction methods, and one of the plurality of spectrum subtraction methods can be selected as the spectrum subtraction method to be used. The signal processing apparatus according to claim 1.

7. The signal processing apparatus further includes a user interface capable of receiving a user operation, and the spectrum subtraction processing circuit can select one of the plurality of spectrum subtraction methods as the spectrum subtraction method to be used based on the user operation. The signal processing apparatus according to claim 6.

8. The signal processing apparatus has a calibration mode, and the spectrum subtraction processing circuit can select the spectrum subtraction method to be used based on the convergence characteristics when each of the plurality of spectrum subtraction methods is used in the calibration mode. The signal processing apparatus according to claim 6.

9. The signal processing apparatus further includes the sensor, and the sensor is a microphone configured using MEMS technology. The signal processing apparatus according to claim 1.

10. The control unit can generate a plurality of coefficients respectively corresponding to a plurality of frequencies, can calculate the filter coefficient by performing an inverse discrete Fourier transform based on the plurality of coefficients, and can set the change amount of each of the plurality of coefficients based on the convergence speed of the plurality of coefficients. The signal processing apparatus according to claim 1.

11. Generating a second signal by performing a filtering process on a first signal using a filter coefficient; generating a first spectral signal by performing a discrete Fourier transform on the first signal; generating a fourth signal by performing a subtraction process based on a third signal supplied from a sensor and corresponding to the first signal and the second signal; generating a second spectral signal by performing a discrete Fourier transform on the fourth signal; generating a third spectral signal by performing a spectral subtraction process using a noise spectral signal indicated by a noise model prepared in advance based on the second spectral signal; and calculating the filter coefficient so that the signal component of the third spectral signal becomes smaller based on the first spectral signal and the third spectral signal. A signal processing method including this.

12. A non-transitory recording medium in which software is recorded to cause a processor to perform: generating a second signal by performing a filtering process on a first signal using a filter coefficient; generating a first spectral signal by performing a discrete Fourier transform on the first signal; generating a fourth signal by performing a subtraction process based on a third signal supplied from a sensor and corresponding to the first signal and the second signal; generating a second spectral signal by performing a discrete Fourier transform on the fourth signal; generating a third spectral signal by performing a spectral subtraction process using a noise spectral signal indicated by a noise model prepared in advance based on the second spectral signal; and calculating the filter coefficient so that the signal component of the third spectral signal becomes smaller based on the first spectral signal and the third spectral signal.

Images