JP6114518B2 - Noise reduction device - Google Patents

Description

  The present invention relates to a noise reduction device.

  A spectral subtraction method is known as a method for removing noise superimposed on an input acoustic signal. FIG. 12 shows a block diagram of a conventional noise reduction apparatus 901 using the spectral subtraction method. The noise reduction device 901 subtracts a noise spectrum signal 920 held in advance by a subtraction coefficient α from an input spectrum signal 922 based on an input acoustic signal 921 including speech and noise, thereby subtracting a spectrum signal 931. , And the subtracted spectrum signal 931 is used to generate an output spectrum signal 933 and an output acoustic signal 934 with reduced noise. The noise to be reduced is, for example, lens driving sound in the imaging apparatus. Here, the subtraction coefficient α takes a large value in order to suppress noise. For this reason, depending on the level difference between the voice and noise in the input acoustic signal 921, the subtraction result may become zero or less. If such a subtraction result is included in the output as it is, the output spectrum is greatly distorted. Distortion leads to abnormal noise (so-called musical noise), and makes the listener of the acoustic signal feel uncomfortable.

  In order to prevent this, a signal 932 obtained by multiplying the input spectrum signal 922 by a flooring coefficient β (0.1 or the like) is generated, and if the signal 932 is larger than the signal 931, the signal 932 is output (933, 934). To include. FIGS. 13A to 13C show spectrum examples of the signals 922, 920, and 933, respectively. In the portion within the broken line frame 940 in FIG. 13C, the signal 932 based on the input spectrum signal 922 is reflected in the output spectrum signal 933.

  Note that a first conventional method is also proposed in which an imaging apparatus calculates a power value of an input acoustic signal immediately before a zoom operation and determines a flooring coefficient based on the power value (see Patent Document 1 below). A second conventional method is also known in which a device for collecting a noise signal in synchronization with the sound collection of an input acoustic signal is prepared and noise is removed by an adaptive filter. This method is generally referred to as feedforward ANC (Active Noise Control).

JP 2008-252389 A

  Although the distortion is somewhat suppressed by using the signal 932 using the flooring coefficient β, the suppression effect may be insufficient. For example, when the original input acoustic signal 921 is small, the output level in the band within the broken line frame 940 becomes low enough to be abnormal, and the listener of the output acoustic signal 934 may feel uncomfortable.

  In the first conventional method, the adjustment of the flooring coefficient may have a certain effect on distortion suppression. However, if the level of the input acoustic signal fluctuates during the zoom operation, Adjustment does not work effectively. Further, when the second conventional method is used, noise reduction processing according to the noise signal at that time can be performed, but between the noise component included in the actual input acoustic signal and the noise component collected as the noise signal. However, since the noise transmission paths are different, a phase shift or the like may occur between them, and noise may not be reduced appropriately.

  Therefore, the present invention has an object to provide a noise reduction device that contributes to optimization of noise reduction, in particular, a noise reduction device that can suppress distortion (discomfort of the listener) that can be caused by noise reduction using spectral subtraction, for example. To do.

  A first noise reduction apparatus according to the present invention subtracts a signal obtained by multiplying a noise spectrum signal by a predetermined coefficient from an input spectrum signal based on an input acoustic signal collected by a sound collection unit. In a noise reduction apparatus that generates a spectrum signal and generates an output acoustic signal using the subtracted spectrum signal, a background noise signal holding unit that holds a background noise spectrum signal acquired in advance, the subtracted spectrum signal, and the background noise An output generation unit configured to generate the output acoustic signal using a spectrum signal.

  As a result, for example, in the output acoustic signal, at least a signal of the level of the background noise spectrum can be secured, and signal distortion that causes the output level to be abnormally low (such as making the listener feel uncomfortable) is suppressed. Is possible.

  Specifically, for example, the output generation unit includes a first spectrum signal as the subtracted spectrum signal, a second spectrum signal obtained by multiplying the input spectrum signal by a predetermined second coefficient, and the background noise spectrum signal. The output acoustic signal may be generated based on a third spectrum signal obtained by multiplying a predetermined third coefficient.

  More specifically, for example, the output generation unit compares the magnitudes of the first, second, and third spectrum signals in each of a plurality of frequency bands, and the first, second, and third spectrum signals. Of these, the output acoustic signal may be generated using a spectrum signal having the maximum magnitude.

  Also, for example, based on the magnitude relationship between the magnitudes of the first, second, and third spectral signals, the first coefficient, the second coefficient, and the third coefficient as the coefficients to be multiplied by the noise spectrum signal. A coefficient updating unit that updates at least one coefficient as an update target coefficient may be further provided in the first noise reduction device.

  As a result, the coefficients can be optimized, and as a result, noise reduction can be optimized.

  Specifically, for example, the coefficient updating unit updates the second time after the first time based on a magnitude relationship between the magnitudes of the first, second, and third spectrum signals at the first time. The update may be performed so that the value of the target coefficient is determined.

  Further, for example, based on the magnitude relationship between the magnitude of the subtracted spectrum signal and the magnitude of the spectrum signal obtained by multiplying the background noise spectrum signal by a predetermined coefficient, a coefficient for updating the coefficient multiplied by the noise spectrum signal An update unit may be further provided in the first noise reduction device.

  As a result, it is possible to optimize the coefficient to be multiplied by the noise spectrum signal, and as a result, noise reduction is optimized.

  Further, for example, the noise is collected in synchronization with the collection of the input acoustic signal by the sound collecting unit to generate an acoustic signal in the time domain of the noise, and the generated acoustic signal is converted into a signal in the frequency domain. Thus, a signal generation unit that generates the noise spectrum signal may be further provided in the first noise reduction device.

  Thereby, noise reduction suitable for the noise at that time becomes possible. If the acoustic signal of noise collected in synchronization with the input acoustic signal is converted into a signal on the frequency domain and a subtracted spectrum signal is generated based on the converted signal (noise spectrum signal), feed forward Unlike type ANC, appropriate noise reduction is possible regardless of phase. Proper noise reduction leads to distortion reduction.

  The second noise reduction apparatus according to the present invention subtracts a signal obtained by multiplying a noise spectrum signal by a predetermined coefficient from an input spectrum signal based on the input acoustic signal collected by the sound collection unit. In the noise reduction device that generates a spectrum signal and generates an output acoustic signal using the subtracted spectrum signal, the noise time is acquired by collecting the noise in synchronization with the collection of the input acoustic signal by the sound collection unit. A signal generation unit is provided that generates an acoustic signal on a region and converts the generated acoustic signal into a signal on a frequency region to generate a spectrum signal of the noise.

  Thereby, noise reduction suitable for the noise at that time becomes possible. If the acoustic signal of noise collected in synchronization with the input acoustic signal is converted into a signal on the frequency domain and a subtracted spectrum signal is generated based on the converted signal (noise spectrum signal), feed forward Unlike type ANC, appropriate noise reduction is possible regardless of phase. Proper noise reduction leads to distortion reduction.

  Further, for example, based on the magnitude relationship between the magnitude of the first spectrum signal as the subtracted spectrum signal and the magnitude of the second spectrum signal obtained by multiplying the input spectrum signal by a predetermined second coefficient, the spectrum signal of the noise A coefficient updating unit that updates the coefficient multiplied by may be further provided in the second noise reduction device.

  As a result, it is possible to optimize the coefficient to be multiplied by the noise spectrum signal, and as a result, noise reduction is optimized.

  ADVANTAGE OF THE INVENTION According to this invention, it is possible to provide the noise reduction apparatus which contributes to optimization of noise reduction, especially the noise reduction apparatus which can suppress the distortion (listener's uncomfortable feeling) which may arise by noise reduction using spectrum subtraction, for example. It is.

It is a figure which shows the site | part connected to the noise reduction apparatus which concerns on 1st Embodiment of this invention, and a noise reduction apparatus. It is a figure which shows the electronic device provided with the noise reduction apparatus. It is a figure which shows the structure of three spectrum signals in a noise reduction apparatus. It is a figure which shows the structure of the output spectrum signal in a noise reduction apparatus. It is a figure which shows the example of the two spectrum signals in a noise reduction apparatus. It is a figure which shows the result signal obtained by applying a spectrum subtraction process with respect to two spectrum signals of Fig.5 (a) and (b), when the presence of a background noise spectrum signal is disregarded. It is a figure which shows the example of the background noise spectrum signal in a noise reduction apparatus. It is a figure which shows the example of an output spectrum signal based on three spectrum signals of Fig.5 (a) and (b) and FIG. It is a figure which shows the temporal relationship of two frames. It is a figure which shows the site | part connected to the noise reduction apparatus which concerns on 2nd Embodiment of this invention, and a noise reduction apparatus. It is a figure which shows the internal structural example of the non-target sound signal generation part of FIG. It is an internal block diagram of the conventional noise reduction apparatus. It is a figure which shows the input in the noise reduction apparatus of FIG. 12, a noise, and an output spectrum signal.

  Hereinafter, an example of an embodiment of the present invention will be specifically described with reference to the drawings. In each of the drawings to be referred to, the same part is denoted by the same reference numeral, and redundant description regarding the same part is omitted in principle. In this specification, for simplification of description, a symbol or reference that refers to information, signal, physical quantity, state quantity, member, or the like is written to indicate information, signal, physical quantity, state quantity or Names of members and the like may be omitted or abbreviated.

<< First Embodiment >>
In FIG. 1, the noise reduction apparatus 1 which concerns on 1st Embodiment of this invention, and the site | part connected to the noise reduction apparatus 1 are shown. The noise reduction apparatus 1 of FIG. 1 includes each part referred to by reference numerals 11 to 20. The noise reduction apparatus 1 is mounted on an arbitrary electronic device 100 having a sound collection function together with the sound collection unit 2 and the preprocessing unit 3 (see FIG. 2). The electronic device 100 includes, for example, a sound collection device such as an imaging device or a voice recorder having a sound collection function. The imaging device is a digital video camera that can capture and record still images and moving images, or a digital still camera that can capture and record only still images. An imaging apparatus as the electronic apparatus 100 includes a solid-state imaging device, a focus lens for forming a subject image on the solid-state imaging device, and a zoom lens for adjusting a shooting angle of view, and a focus lens, a zoom lens, or an arbitrary lens A motor for moving the element is provided (both not shown). The noise reduction device 1 has a function of reducing noise in an acoustic signal. As noise to be reduced, for example, driving sound accompanying movement of the focus lens or zoom lens and driving sound of the motor are assumed. .

The sound collection unit 2 is composed of one or more microphones. The sound collection unit 2 collects the peripheral sound of the electronic device 100 including the target sound, converts the collected sound into an analog sound signal, and outputs the analog sound signal. The preprocessing unit 3 performs predetermined signal processing on the output signal of the sound collection unit 2 to generate an input acoustic signal S _IN [t] that is an acoustic signal in the time domain. The signal processing includes an A / D conversion process for converting an analog sound signal output from the sound collection unit 2 into a digital sound signal, a frame division process for dividing the digital sound signal into frame units having a predetermined time length, and the like. including.

The FFT unit 11 performs a discrete Fourier transform, which is a type of time / frequency conversion, on the input acoustic signal S _IN [t] divided in units of frames for each frame, whereby the input spectrum signal S _IN [f ] Is generated. The input spectrum signal S _IN [f] is obtained by converting the input acoustic signal S _IN [t] into a signal on the frequency domain.

The input sound signal S _IN [t] and the input spectrum signal S _IN [f] include the sound signal of the target sound to be collected by the sound collecting unit 2 and the sound of the non-target sound different from the target sound. It also includes a signal.
The target sound is a sound to be collected by the sound collection unit 2, such as a human voice, an animal call, or music. Of the sound signal components included in the input acoustic signal S _IN [t] and the input spectrum signal S _IN [f], signal components other than the non-target sound signal components may be considered to be the target sound signal components. Good.
The non-target sound is a kind of noise for the target sound, and is reduction target noise (specific noise) to be reduced in the noise reduction device 1. The non-target sound may be a sound generated from a specific noise source (in this case, it can be said that the specific noise source is a source of non-target sound). The sound generated by the specific noise source is, for example, a driving sound accompanying the movement of the focus lens or the zoom lens or a driving sound of the motor.

Note that the output sound signal of the sound collection unit 2, the digital sound signal, the input sound signal S _IN [t], or the input spectrum signal S _IN [f] is passed through a recording medium (not shown) for recording any of them. The electronic device 100 or the noise reduction device 1 may be provided. That is, for example, the input acoustic signal S _IN [t] generated by the preprocessing unit 3 is recorded on a recording medium (not shown), and the input acoustic signal S _IN [t] is converted into a noise reduction device at an arbitrary timing thereafter. 1 may be provided.

The non-target sound signal holding unit 12 is a memory that holds the spectrum signal S _A [f] corresponding to the non-target sound. In the present embodiment, it is assumed that the spectrum signal S _A [f] is prepared in advance and held in the holding unit 12 before the input acoustic signal S _IN [t] is input to the noise reduction device 1. For example, an acoustic signal of a non-target sound (driving sound of the focus lens or the like) is acquired in advance, and a spectrum signal S _A [f] is acquired in advance by performing discrete Fourier transform on the acquired acoustic signal. I can leave. In order to acquire the acoustic signal of the non-target sound and the spectrum signal S _A [f], the sound collecting unit 2 may be used, or a sound collecting device (microphone) different from the sound collecting unit 2 may be used. . Alternatively, the vibration component of the non-target sound source is measured using a measurement unit (acceleration sensor or the like), and the non-target sound signal and the spectrum signal S _A [f] are acquired from the measurement result. good.

The multiplier 13 generates a spectrum signal S _A ′ [f] by multiplying the spectrum signal S _A [f] by a predetermined coefficient α. The subtractor 14 subtracts the spectrum signal S _A ′ [f] from the input acoustic signal S _IN [t], thereby generating a spectrum signal S ₁ [f] that can be called a subtracted spectrum signal. The multiplier 15 generates a spectrum signal S ₂ [f] by multiplying the input spectrum signal S _IN [f] by a predetermined coefficient β. The coefficient α is also called a subtraction coefficient, and the coefficient β is also called a flooring coefficient.

The background noise signal holding unit 16 is a memory that holds the spectrum signal S _B [f] corresponding to the background noise. Although the background noise is a kind of noise, the background noise assumed in the present embodiment is different from the non-target sound. Background noise refers to noise that is collected by the sound collection unit 2 when there is no target sound and non-target sound, and is also referred to as background noise or background sound. For example, prior to actual operation of the noise reduction apparatus 1, an acoustic signal of background noise is acquired in a predetermined sound collection environment, and the spectrum signal S _B [f is obtained by performing discrete Fourier transform on the acquired acoustic signal of background noise. ] Can be obtained in advance. The predetermined sound collection environment is, for example, a sufficiently quiet environment for the sound collection unit 2 and the electronic device 1. The sound collection unit 2 is used to acquire the acoustic signal of the background noise and the spectrum signal S _B [f].

The multiplier 17 multiplies the spectrum signal S _B [f], which is a background noise spectrum signal, by a predetermined coefficient (background noise coefficient) γ to generate the spectrum signal S ₃ [f].

As understood from the above description, the spectrum signals S ₁ [f] to S ₃ [f] satisfy the following expressions (A1) to (A3).
S ₁ [f] = S _IN [f] −S _A [f] × α (A1)
S ₂ [f] = S _IN [f] × β (A2)
S ₃ [f] = S _B [f] × γ (A3)

In this embodiment, the spectrum signal refers to an acoustic signal in the frequency domain, and the frequency sampling interval is Δf in an arbitrary spectrum signal. An arbitrary spectrum signal is composed of a total of M signal components from a signal component with frequency number 0 to a signal component with frequency number (M−1). M is an integer greater than or equal to 2, for example 128. For clarity of description, each frequency band divided by the frequency sampling interval Δf is called a unit band. Then, the entire frequency band of an arbitrary spectrum signal is formed from M unit bands (divided by M unit bands).
The m-th unit band refers to a frequency band from a frequency of (m × Δf) to a frequency of ((m + 1) × Δf) (m is an integer).

Therefore, as shown in FIGS. 3A to 3C, the input spectrum signal S _IN [f] is composed of signal components S _IN [0 · Δf] to S _IN [M−1 · Δf], Spectrum signal S _A [f] is composed of signal components S _A [0 · Δf] to S _A [M−1 · Δf], and background noise spectrum signal S _B [f] is signal component S _B [0]. Δf] to S _B [M−1 · Δf]. The signal components S _IN [m · Δf], S _A [m · Δf], and S _B [m · Δf] are included in the spectrum signals S _IN [f], S _A [f], and S _B [f], respectively. The signal component in the mth unit band is shown.

Since the multiplications of the multiplication units 13, 15 and 17 and the subtraction of the subtraction unit 14 are performed for each unit band, the following equations (A4) to (A6) corresponding to the above equations (A1) to (A3) are established. . When i = 1, i = 2, or i = 3, the spectrum signal S _i [f], which is the spectrum signal S ₁ [f], S ₂ [f], or S ₃ [f], is a signal component S _i [ 0 · Δf] to S _i [M−1 · Δf], and the signal component S _i [m · Δf] indicates a signal component in the m-th unit band included in the spectrum signal S _i [f].
S ₁ [m · Δf] = S _IN [m · Δf] −S _A [m · Δf] × α (A4)
S ₂ [m · Δf] = S _IN [m · Δf] × β (A5)
S ₃ [m · Δf] = S _B [m · Δf] × γ (A6)

Based on the spectrum signals S ₁ [f], S ₂ [f] and S ₃ [f], the comparison processing unit 18 generates an output spectrum signal S _OUT [f] through comparison of the signal levels. As shown in FIG. 4, the output spectrum signal S _OUT [f] is composed of signal components S _OUT [0 · Δf] to S _OUT [M−1 · Δf], and the signal component S _OUT [m · Δf] is output. The signal component in the m-th unit band included in the spectrum signal S _OUT [f] is shown.

The IFFT 19 unit converts the output spectrum signal S _OUT [f] into an output acoustic signal S _OUT [t] in the time domain using inverse discrete Fourier transform, which is a type of frequency / time conversion. The output acoustic signal S _OUT [t] is output from the noise reduction device 1 as an acoustic signal in which noise (including non-target sounds) included in the input acoustic signal S _IN [t] is suppressed. The coefficient setting unit 20 sets the values of the coefficients α, β, and γ. The coefficients α, β, and γ satisfy “0 <α”, “0 <β <1”, and “0 <γ <1”. The coefficient setting unit 20 also has a function as a coefficient update unit, which will be described later.

Consider the case where the spectral signals S _IN [f] and S _A [f] are the spectral signals 300 _IN [f] and 300 _A [f] in FIGS. 5 (a) and 5 (b), respectively. For convenience, in FIGS. 5A and 5B and FIGS. 6 to 8 described later, the spectrum signals to be illustrated are represented as continuous curves by connecting the discretized signal components. . Also, in the explanatory text referring to the signals 300 _IN [f] and 300 _A [f], it is assumed that α = β = 1 for the sake of simplification and convenience.

In order for noise reduction in the spectral subtraction method to function effectively, the signal level of the signal 300 _A [f] needs to be higher than the signal level of the signal 300 _IN [f]. However, the signal level of the signal 300 _A [f] is lower than the signal level of the signal 300 _IN [f] in the frequency band surrounded by the broken line frame 301 (hereinafter referred to as the frequency band 301) ( For convenience, this is called the reverse phenomenon).

While the spectrum signal 300 _A [f] is prepared in advance by evaluating the non-target sound, the input spectrum signal 300 _IN [f] corresponds to the sum of the acoustic signals of the target sound and the non-target sound. Therefore, originally, the signal level of the signal 300 _IN [f] should be larger than the signal level of the signal 300 _A [f].
However, the non-target sound generated at that time varies in time series. As a result, the non-target sound at the timing of interest (that is, the non-target sound in the input acoustic signal) and the non-target sound at the timing of the evaluation (that is, the holding unit 12). The difference from the non-target sound held in the sound may increase. The increase in the difference causes the above reversal phenomenon. In particular, for example, when the non-target sound is noise or the like due to the driving of the lens of the imaging apparatus, the non-target sound generated at that time varies over time due to the secular change of the lens unit.
Further, when the spectrum signal 300 _A [f] obtained through the evaluation of the non-target sound is commonly applied to a plurality of electronic devices 100, the above inversion phenomenon occurs due to the influence of individual differences in the electronic devices 100. (The commonly applied spectral signal 300 _A [f] is optimal for one electronic device 100, but may be too large for some other electronic device 100). In particular, for example, when the non-target sound is noise or the like due to the driving of the lens of the imaging device, the individual difference variation of the lens unit may cause the reverse phenomenon.
In addition, depending on the phase relationship between the target sound and the non-target sound, the power of the input acoustic signal may be considerably reduced in the local frequency band. In such a case, the reverse phenomenon occurs.

If the conventional spectral subtraction processing corresponding to the configuration of FIG. 12 is applied to the spectral signals 300 _IN [f] and 300 _A [f], the result signal 310 _OUT [f] as shown in FIG. ] Is obtained as a result signal of the spectral subtraction process (the signal 310 _OUT [f] corresponds to the signal S _OUT [f] obtained when it is assumed that there is no background noise spectrum signal for convenience). In the frequency band 301, when the signal obtained by multiplying the spectrum signal 300 _A [f] by the subtraction coefficient α is smaller than the signal obtained by multiplying the spectrum signal 300 _IN [f] by the flooring coefficient β, the former signal Instead of (300 _A [f] × α), the latter signal (300 _IN [f] × β) is included in the result signal 310 _OUT [f]. Thus, in the frequency band 301, than to include the former signal _{(300 A [f] × α} ) results in signal 310 _OUT [f] is, the distortion of the result signal 310 _OUT [f] (distortion leads to abnormal noise Although the sound signal listener feels uncomfortable, the suppression effect may be insufficient.

For example, when the original input acoustic signal S _IN [t] is small, the signal level of the result signal 310 _OUT [f] may be lower than the signal level of background noise in the frequency domain 301. The background noise is background noise included in the input sound signal S _IN [t] when there is no target sound and non-target sound, and thus the signal level of the result signal 310 _OUT [f] that should be a correction signal of the input sound signal Is inherently unnatural and the result signal 310 _OUT [f] is such that the signal level of the result signal 310 _OUT [f] is smaller than the signal level of the background noise. This makes the listener of the acoustic signal of the result signal 310 _OUT [f] feel uncomfortable.

Considering these, in the first embodiment, the background noise signal holding unit 16 is provided as described above, and the output spectrum signal S _OUT [f] is generated also using the background noise spectrum signal S _B [f]. Therefore, the comparison processing unit 18 refers to the spectrum signal S ₃ [f] based on the background signal S _B [f] in addition to the spectrum signals S ₁ [f] and S ₂ [f], An output spectrum signal S _OUT [f] is generated based on the magnitude relationship between the signal levels of the spectrum signals S ₁ [f] to S ₃ [f]. More specifically, the comparison processing unit 18 compares the signal levels of the spectrum signals S ₁ [f] to S ₃ [f] for each unit band, and the spectrum signals S ₁ [f] to S ₃ [f]. Of these, the spectrum signal having the maximum signal level is included in the output spectrum signal S _OUT [f].

Among spectrum signal _{S 1 [f] ~S 3 [} f], the condition signal level of the spectral signal S ₁ [f] is maximized is referred to as a first condition, the spectrum signal _{S 1 [f] ~S 3 [} f ], The condition where the signal level of the spectrum signal S ₂ [f] is maximized is called a second condition. Of the spectrum signals S ₁ [f] to S ₃ [f], the spectrum signal S ₃ [f] The condition that maximizes the signal level is called the third condition.

For the m-th unit band, an output spectrum according to the following equation (A7) when the first condition is satisfied, according to the following equation (A8) when the second condition is satisfied, and according to the following equation (A9) when the third condition is satisfied. signal S _OUT signal component S _OUT of the [f] [m · Δf] is generated, the comparison processing unit 18 executes the generation processing for each unit band. The i-th condition is satisfied for the m-th unit band when the signal component S _i [m · Δf] is selected from the signal components S ₁ [m · Δf], S ₂ [m · Δf] and S ₃ [m · Δf]. Is the maximum signal level (where i is 1, 2 or 3).
S _OUT [m · Δf] = S ₁ [m · Δf] (A7)
S _OUT [m · Δf] = S ₂ [m · Δf] (A8)
S _OUT [m · Δf] = S ₃ [m · Δf] (A9)

The spectral signals S _IN [f] and S _A [f] are the spectral signals 300 _IN [f] and 300 _A [f] of FIGS. 5A and 5B, respectively, and the background noise spectral signal S. Consider the case where _B [f] is the spectrum signal 300 _B [f] in FIG. In addition, it is assumed that the third condition is satisfied in the frequency band 301. Under this assumption, in the frequency band 301, the spectrum signal S ₃ [f] based on the spectrum signal S _B [f] (300 _B [f]) is included in the output spectrum signal S _OUT [f]. Spectrum signal 320 _OUT [f] is output from the comparison processing unit 18 as an output spectrum signal S _OUT [f]. That is, the result of replacing the signal in the frequency band 301 with the spectrum signal S ₃ [f] corresponding to the spectrum signal S _B [f] (300 _B [f]) using the signal 310 _OUT [f] in FIG. 6 as a reference. It can be seen that the signal level in the frequency band 301 in the output spectrum signal S _OUT [f] (320 _OUT [f]) is corrected to be increased.

As described above, distortion in the output sound signal can be reduced by generating the output sound signal using the spectrum signal of the background noise while using the spectrum subtraction method as a basis. That is, it is possible to compensate for distortion of the output acoustic signal (corresponding to signal distortion in the frequency band 301 in FIG. 6) such that a signal smaller than the background noise signal level is included in the output acoustic signal (signal in FIG. 6). The change from 310 _OUT [f] to the signal 320 _OUT [f] in FIG. 8 corresponds to the compensation), and it is possible to reduce the uncomfortable feeling of the listener of the output acoustic signal that may be caused by the distortion.

  Next, the coefficient update function of the coefficient setting unit 20 in FIG. 2 will be described. In the coefficient updating function, the coefficient setting unit 20 having a function as a coefficient updating unit sets at least one coefficient among the three coefficients α, β, and γ as the update target coefficient, and updates the value of the update target coefficient. can do. Although not particularly conscious in the above description, the coefficient setting unit 20 can individually set the values of the coefficients α, β, and γ for each unit band, and can perform the update for each unit band.

Here, for clarity of explanation, the first frame and the second frame shown in FIG. 9 are defined. As described above, the input acoustic signal S _IN [t] is divided in units of frames, and the input spectrum signal S _IN [f] is generated from the input acoustic signal S _IN [t] for each frame. , Any frame that is later in time than the first frame. The second frame may be a frame adjacent to the first frame. Part of the first and second frames may overlap each other. The coefficients α, β, and γ for the mth unit band applied to the first frame are represented by symbols α [m], β [m], and γ [m], respectively, and applied to the second frame. In particular, the coefficients α, β, and γ for the m-th unit band are represented by symbols α _NEW [m], β _NEW [m], and γ _NEW [m], respectively.

[How to update the subtraction coefficient α]
A coefficient update method when the coefficient α is set as the update target coefficient will be described. For the mth unit band of the first frame, the coefficient setting unit 20 follows the equation (B1) when the first condition is satisfied, follows the equation (B2) when the second condition is satisfied, and applies the equation when the third condition is satisfied. According to (B3), the value of the coefficient α _NEW [m] of the second frame is determined. However, the constants K1 to K3 in the formulas (B1) to (B3) satisfy “1 <K1” and “1 <K3 <K2 <1”.
α _NEW [m] = α [m] × K1 (B1)
α _NEW [m] = α [m] × K2 (B2)
α _NEW [m] = α [m] × K3 (B3)

The coefficient α should be as large as possible in order to remove non-target sound from the output acoustic signal as much as possible. When the first condition is satisfied, the signal S ₁ [f] is the largest of the signals S ₁ [f] to S ₃ [f], and therefore it is determined that the removal of the non-target sound can be further increased. In order to increase the subtraction amount in the subtraction unit 14, the coefficient α is increased by the equation (B1) to reduce the signal S ₁ [f]. On the other hand, when the second condition is satisfied, there is a possibility that the subtraction amount in the subtraction unit 14 is too large. Therefore, the coefficient α is decreased by the equation (B2) to increase the signal S ₁ [f]. The state where the third condition is satisfied is a state where the signal S ₁ [f] is smaller than the signal S ₃ [f] corresponding to the background noise, and the subtraction amount in the subtracting unit 14 is smaller than that when the second condition is satisfied. It is considered to be too large. Therefore, when the third condition is satisfied, the coefficient α is updated by the equation (B3) so as to further decrease the coefficient α compared to when the second condition is satisfied, and the signal S ₁ [f] is further increased.

[How to update the flooring coefficient β]
A coefficient update method when the coefficient β is set as the update target coefficient will be described. For the mth unit band of the first frame, the coefficient setting unit 20 follows the equation (C1) when the first condition is satisfied, follows the equation (C2) when the second condition is satisfied, and applies the equation when the third condition is satisfied. According to (C3), the value of the coefficient β _NEW [m] of the second frame is determined. However, the constants K1 to K3 in the equations (C1) to (C3) satisfy “1 <K1”, “0 <K2 <1”, and “1 <K3”.
β _NEW [m] = β [m] × K1 (C1)
β _NEW [m] = β [m] × K2 (C2)
β _NEW [m] = β [m] × K3 (C3)

When the first condition is satisfied, the signal S ₂ [f] is too small and the significance of inputting the signal S ₂ [f] to the comparison processing unit 18 (when S ₁ [f] is too small, S ₁ [f] is There is a possibility that the distortion that may occur if S _OUT [f] is included, is compensated with S ₂ [f]). Therefore, when the first condition is satisfied, the coefficient S is increased by the equation (C1) to increase the signal S ₂ [f]. During the second condition is established, the signal S ₂ [f] is because it may not be the original noise reduction using the signal S ₁ [f] is too large, the coefficient β and the signal S ₂ by the equation (C2) [F] is reduced. The state in which the third condition is satisfied is a state in which the signal S ₂ [f] is smaller than the signal S ₃ [f] corresponding to background noise. A state where the signal level of background noise that does not include the target sound should not be lower than the signal level of the signal that includes the target sound, and it can be said that this state is caused by a small coefficient β. Therefore, when the third condition is satisfied, the coefficient β and the signal S ₂ [f] are increased by the equation (C3).

[How to update coefficient γ for background noise]
A coefficient updating method when the coefficient γ is set as the update target coefficient will be described. For the mth unit band of the first frame, the coefficient setting unit 20 follows the equation (D1) when the first condition is satisfied, follows the equation (D2) when the second condition is satisfied, and applies the equation when the third condition is satisfied. According to (D3), the value of the coefficient γ _NEW [m] of the second frame is determined. However, the constants K1 to K3 in the equations (D1) to (D3) satisfy “1 <K1”, “1 <K2”, and “0 <K3 <1”.
γ _NEW [m] = γ [m] × K1 (D1)
γ _NEW [m] = γ [m] × K2 (D2)
γ _NEW [m] = γ [m] × K3 (D3)

When the first condition is satisfied, the signal S ₃ [f] is too small, and the significance of inputting the signal S ₃ [f] to the comparison processing unit 18 (the distortion in the frequency band 301 in FIG. 6 is changed to the signal S ₃ [f]). The meaning of compensation as shown in FIG. Therefore, when the first condition is satisfied, the coefficient γ is increased by the equation (D1) to increase the signal S ₃ [f]. Even when the second condition is satisfied, similarly to the case when the first condition is satisfied, the signal S ₃ [f] may be too small. Therefore, the coefficient γ and the signal S ₃ [f] are increased according to the equation (D2). When the third condition is established, there is a possibility that the signal S ₃ [f] is too large to reduce the non-target sound that should be originally achieved (reflection of S ₁ [f] to S _OUT [f]). Therefore, the coefficient γ and the signal S ₃ [f] are reduced by the equation (D3).

[Common items for coefficient update]
The coefficient setting unit 20 can regard each frame that is visited one after another as the first frame, and can update the update target coefficient for each frame. However, there may be a frame in which the update target coefficient is not updated.

The method of obtaining α _NEW [m] by multiplying α [m] by a constant (K1, K2 or K3) is illustrated, but as long as the above update is followed, α _NEW [m] The arithmetic expression to be determined can be arbitrarily changed (the same applies to the coefficients β and γ). For example, when the coefficient α is increased, α _NEW [m] is determined by adding a positive predetermined value to α [m], and when the coefficient α is decreased, a positive value is determined from α [m]. Α _NEW [m] may be determined by subtracting a predetermined value (the same applies to the coefficients β and γ).

Further, if the update process is limited so that the coefficients α, β, and γ are updated within the range where “0 <α”, “0 <β <1”, and “0 <γ <1” are satisfied. good. That is, since the signal corresponding to the non-target sound is “subtracted” from the input spectrum signal S _IN [f], the coefficient α is always positive. When the signal S ₂ [f] becomes larger than the signal S _IN [f], the signal S ₂ [f] is always larger than the signal S ₁ [f], and the inherent noise reduction becomes impossible. A positive value less than 1. It is assumed that the signal S ₃ [f] functions effectively in a situation where the input acoustic signal is close to the background noise level. In such a situation, the signal S ₃ [f] is actually dark. It should not be louder than noise (because a signal larger than background noise is mixed in the output acoustic signal). Therefore, the coefficient γ is a positive value less than 1.

As described above, the coefficient setting unit 20, spectrum signal S ₁ [f] ~S ₃ of on a relationship in magnitude between the signal levels of the [f] (signal _{S 1 [f] ~S 3 [} f], any signal The value of the update target coefficient can be sequentially updated based on whether it is included in S _OUT [f]. That is, the coefficient setting unit 20 determines the second time after the first time (based on the magnitude relationship between the signal levels of the spectrum signals S ₁ [f] to S ₃ [f] at the first time (first frame). The update target coefficient can be updated so that the value of the update target coefficient in the second frame) is determined.

By this update process, compensation by spectral subtraction or background noise signal suitable for the actually used signal S _IN [f], S _A [f] or S _B [f] (distortion in the frequency band 301 in FIG. 6). , Compensation by the signal S ₃ [f]; see FIG. 8). The amount of subtraction in the subtracting unit 14 is optimized by updating the coefficient α, and it is possible to cope with variations in time series of non-target sounds that actually occur (aging of non-target sounds) and a holding unit. It is possible to cope with the difference (individual difference variation) between the 12 held signals and the actually generated non-target sound signal. Further, the method of adjusting the flooring coefficient based on the power value of the input acoustic signal immediately before the zoom operation described in Japanese Patent Application Laid-Open No. 2008-252389 supports a case where the level of the input acoustic signal varies during the zoom operation. Although it is not possible (although the adjustment contents become invalid), such a case can be appropriately handled by sequentially updating the flooring coefficient β as described above.

  However, it is possible not to update any or all of the coefficients α, β and γ. In other words, in the noise reduction apparatus 1, it is possible to set one or more of the three coefficients α, β, and γ to a fixed value.

<< Second Embodiment >>
A second embodiment of the present invention will be described. The second embodiment is an embodiment based on the first embodiment. Regarding matters not specifically described in the second embodiment, the description of the first embodiment is the second embodiment unless there is a description and no contradiction. Also applies. When the description of the first embodiment is applied to the second embodiment, the noise reduction device 1 in the description of the first embodiment is replaced with the noise reduction device 1a in the second embodiment.

  FIG. 10 shows a noise reduction device 1a according to the second embodiment of the present invention and parts connected to the noise reduction device 1a. The noise reduction apparatus 1a of FIG. 10 includes each part referred to by reference numerals 11, 13 to 20, 31 and 32. The noise reduction device 1a of FIG. 10 is obtained by replacing the holding unit 12 in the noise reduction device 1 of FIG. 1 with a non-target sound signal generation unit 31 and an FFT unit 32, except for the replacement and the contents described later. 1a is the same as the apparatus 1.

The non-target sound signal generation unit 31 collects the non-target sound in synchronization with the collection (sound collection) of the input acoustic signal S _IN [t] using the sound collection unit 2, and the collected non-target sound acoustic signal S _A [t] is generated. In the generation unit 31 or in a frame division unit (not shown) provided between the generation unit 31 and the FFT unit 32, the acoustic signal S _A [t], which is an acoustic signal in the time domain, is divided in units of frames. The The FFT unit 32 performs a discrete Fourier transform, which is a type of time / frequency conversion, for each frame on the acoustic signal S _A [t] divided in units of frames, so that the spectrum signal S _A of the non-target sound is obtained. [F] is generated. The spectrum signal S _A [f] is obtained by converting the acoustic signal S _A [t] into a signal on the frequency domain.

As shown in FIG. 11A, the generating unit 31 may be provided with a sound collecting unit 31a including one or more microphones, and the sound signal S _A [t] may be generated using the sound collecting unit 31a. The sound collection unit 31a is arranged and formed so as to collect only the non-target sound, and converts the collected non-purpose sound into an acoustic signal and outputs it. The generating unit 31 can generate the acoustic signal S _A [t] by digitizing the output acoustic signal of the sound collecting unit 31a. Or you may provide the measurement part 31b in the production | generation part 31, as shown in FIG.11 (b). The measurement unit 31b is formed by an acceleration sensor or the like that measures the vibration component of the non-target sound source, and the generation unit 31 converts the measurement result of the measurement unit 31b into an acoustic signal, thereby converting the acoustic signal S _A [ t] may be generated.

By performing spectral subtraction using a non-target sound signal collected in real time, it is possible to perform optimal spectral subtraction suitable for the non-target sound at that time. Optimal spectral subtraction is to optimize the subtracted spectral signal S ₁ [f] (for example, a situation where S ₁ [f] <S ₂ [f] is avoided as much as possible). This also leads to a reduction in distortion as described with reference to FIG. 6 (a reduction in the uncomfortable feeling of the listener of the output acoustic signal).

Further, in the conventional feedforward type ANC (Active Noise Control), a phase shift or the like between a noise component in an input acoustic signal and a noise component collected as a noise signal becomes a problem. Noise can be reduced regardless of the phase by performing time / frequency conversion of the noise (non-target sound) acoustic signal S _A [t] and performing spectrum calculation (spectral power comparison or the like) in the frequency domain.

  Also in the second embodiment, the coefficient α, β, and γ update processing by the coefficient setting unit 20 described in the first embodiment may be performed. However, in the noise reduction apparatus 1a, it is also possible to set a value of one or more of the three coefficients α, β and γ to a fixed value.

It is also possible to delete the background noise signal holding unit 16 and the multiplication unit 17 from the noise reduction device 1a. In this case, comparison processing unit 18, for each unit band, to compare the signal level of the spectral signal S ₁ [f] and S ₂ [f] for each unit band, the spectrum signal S ₁ [f] and S ₂ [ f], a spectrum signal having a larger signal level is included in the output spectrum signal S _OUT [f] (that is, of the signal components S ₁ [m · Δf] and S ₂ [m · Δf], The larger signal component is substituted into the signal component S _OUT [m · Δf]).

<< Deformation, etc. >>
The embodiment of the present invention can be appropriately modified in various ways within the scope of the technical idea shown in the claims. The above embodiment is merely an example of the embodiment of the present invention, and the meaning of the term of the present invention or each constituent element is not limited to that described in the above embodiment. The specific numerical values shown in the above description are merely examples, and as a matter of course, they can be changed to various numerical values. As annotations applicable to the above-described embodiment, notes 1 to 3 are described below. The contents described in each comment can be arbitrarily combined as long as there is no contradiction.

[Note 1]
“Signal level” in the description of each embodiment may be read as “size” or “power”. Regarding an arbitrary acoustic signal in the time domain or the frequency domain, the signal level is a kind of index indicating the magnitude of the acoustic signal. Regarding an arbitrary acoustic signal in the time domain or the frequency domain, the power may be an index representing the magnitude of the acoustic signal. That is, regarding an arbitrary acoustic signal in the time domain or the frequency domain, the magnitude of the acoustic signal represents the signal level or power of the acoustic signal.

[Note 2]
The target device that is the noise reduction device 1 or 1a or the electronic device 100 can be configured by hardware or a combination of hardware and software. Arbitrary specific functions that are all or part of the functions realized in the target device are described as a program, the program is stored in a flash memory that can be mounted on the target device, and the program is executed by the program execution device. The specific function may be realized by executing on a microcomputer (for example, a microcomputer that can be mounted on the target device). The program can be stored and fixed in an arbitrary recording medium (not shown). A recording medium (not shown) for storing and fixing the program may be mounted or connected to a device (such as a server device) different from the target device.

[Note 3]
For example, it can be considered as follows. The noise reduction device 1 or 1a subtracts the signal S _A '[f] obtained by multiplying the spectrum signal S _A [f] of noise (noise to be reduced) by the subtraction coefficient α from the input spectrum signal S _IN [f]. Thus, a subtracted spectrum signal S ₁ [f] can be generated, and an output acoustic signal in which noise (noise to be reduced) is suppressed can be generated using the subtracted spectrum signal S ₁ [f]. In each of the above-described embodiments, the noise to be reduced is called a non-target sound.

DESCRIPTION OF SYMBOLS 1, 1a Noise reduction apparatus 12 Non-purpose sound holding part 16 Background noise signal holding part 18 Comparison processing part 20 Coefficient setting part 31 Non-purpose signal generation part 100 Electronic device

Claims (5)

A subtracted spectrum signal is generated by subtracting a signal obtained by multiplying a noise spectrum signal by a predetermined coefficient from an input spectrum signal based on the input acoustic signal collected by the sound collecting unit, and the subtracted spectrum signal is used. In the noise reduction device that generates the output acoustic signal,
A background noise signal holding unit that holds a background noise spectrum signal acquired in advance;
An output generation unit that generates the output acoustic signal using the subtracted spectrum signal and the background noise spectrum signal , and
The output generation unit includes a first spectrum signal as the subtracted spectrum signal, a second spectrum signal obtained by multiplying the input spectrum signal by a predetermined second coefficient, and a predetermined third coefficient to the background noise spectrum signal. Based on the third spectral signal obtained by multiplying by
In each of a plurality of frequency bands, the magnitudes of the first, second, and third spectrum signals are compared, and the spectrum signal having the maximum magnitude among the first, second, and third spectrum signals is used. The noise reduction device is characterized by generating the output acoustic signal . Based on the magnitude relationship between the magnitudes of the first, second, and third spectral signals, at least one of the first coefficient, the second coefficient, and the third coefficient as the coefficient multiplied by the noise spectrum signal. The noise reduction apparatus according to claim 1 , further comprising a coefficient updating unit that updates one coefficient as an update target coefficient. The coefficient updating unit determines a value of the update target coefficient at a second time after the first time based on a magnitude relationship between the magnitudes of the first, second, and third spectrum signals at the first time. The noise reduction apparatus according to claim 2 , wherein the update is performed. A coefficient updating unit that updates the coefficient multiplied by the noise spectrum signal based on a magnitude relationship between the magnitude of the subtracted spectrum signal and the magnitude of the spectrum signal obtained by multiplying the background noise spectrum signal by a predetermined coefficient; The noise reduction device according to claim 1 , further comprising: By generating the acoustic signal in the time domain of the noise by collecting the noise in synchronization with the collection of the input acoustic signal by the sound collection unit, and converting the generated acoustic signal into a signal in the frequency domain noise reduction device according to any one of claims 1 to 4, further comprising a signal generator for generating a spectral signal of the noise.

Images