JP2003140700A - Method and device for noise removal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device and method for noise removal which can obtain a stressed voice of superior quality. SOLUTION: The device has an injected noise calculation part 55 which calculates noise to be injected from a deteriorated voice power spectrum and an estimated noise power spectrum, two adders 56 and 57 which add the obtained noise to the deteriorated voice power spectrum and estimated noise power spectrum, and a noise suppression coefficient generation part 8 which determines a suppression coefficient according to the noise-added deteriorated voice power spectrum and estimated noise power spectrum. Further, the device has a windowing processing part 22 which performs windowing processing for a signal sample extracted from two adjacent frames of a reverse Fourier transform output.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise removing method and device, and more particularly, to a noise removing method and device for removing noise superimposed on a desired audio signal.

[0002]

2. Description of the Related Art Noise elimination device (noise suppressor)
Is to remove noise superimposed on a desired speech signal. The power spectrum of the noise component is estimated using the input signal transformed from the time domain to the frequency domain, and this estimated power spectrum is input. By subtracting from the signal, it operates so as to suppress the noise mixed in the desired audio signal. The power spectrum of the noise component can be applied to suppression of non-stationary noise by detecting and updating the silent section of the voice. As a noise eliminator, for example, “December 1984, I E E E Transactions on Aqueous Speech and Signal Processing, Volume 32, No. 6 (IEEE TRANSACTIONS ON ACOUS
TICS, SPEECH, AND SIGNAL PROCESSING, VOL.32, NO.6,
PP.1109-1121, DEC, 1984), pages 1109 to 1121 ”(reference 1). This is known as the minimum mean square error short time spectral amplitude method. FIG. 48 shows the configuration of the noise eliminator described in Document 1.

A deteriorated voice signal (a signal in which a desired voice signal and noise are mixed) is supplied to the input terminal 11 as a time domain sample value sequence. The deteriorated audio signal sample is supplied to the frame division unit 1 and divided into K / 2 sample frames. Here, K is an even number of 2 or more. The deteriorated audio signal sample divided into frames is supplied to the windowing processing unit 2 and is multiplied by the window function w (t). The input signal y _n (t) (t = 0,
The signal y _n (t) bar windowed by w (t) for 1, ...., K / 2−1) is given by equation (1).

[0004]

[Equation 1]

It is also widely practiced to overlap a part of two consecutive frames to form a window. Assuming 50% of the frame length as the overlap length, for t = 0, 1, ..., K / 2-1,
Y _n (t) bar (t = 0, 1, ...,
K / 2-1) is the output of the windowing processing unit 2.

[0006]

[Equation 2]

A symmetric window function is used for real signals. Further, the window function is designed so that the input signal and the output signal when the suppression coefficient, which will be described later, is set to 1 match each other except for a calculation error. This is w (t) + w (t +
It means that K / 2) = 1. After that, 2 consecutive times
The description will be continued by taking as an example the case of opening windows by overlapping 50% of the frames. As the window function w (t), for example, the Hanning window shown in Expression (3) can be used.

[0008]

[Equation 3]

The windowed output y _n (t) bar is supplied to the Fourier transform unit 3 and transformed into a degraded speech spectrum (frequency domain signal) Y _n (k) in the frequency domain. The deteriorated speech spectrum Y _n (k) is separated into a phase and an amplitude, and the argY _n (k) of the deteriorated speech phase spectrum is inverse Fourier transform unit 9.
Then, the deteriorated voice amplitude spectrum | Y _n (k) | is supplied to the voice detection unit 4, the multiple multiplication unit 16 and the multiple multiplication unit 17.

The voice detection unit 4 detects the presence or absence of voice based on the deteriorated voice amplitude spectrum | Y _n (k) |, and the voice detection flag determined by the result is used as the estimated noise calculation unit 5
Propagate to 1. The multiple multiplication unit 17 squares the supplied deteriorated speech amplitude spectrum | Y _n (k) | by frequency, and the estimated noise calculation unit 51 and frequency-dependent SNR (signal-to-noise ratio) calculation unit 6 as a deteriorated speech power spectrum. Communicate to. The estimated noise calculation unit 51 estimates the power spectrum of noise (second noise) included in the deteriorated voice amplitude spectrum, using the voice detection flag, the deteriorated voice power spectrum, and the count value supplied from the counter 13. , Is transmitted to the frequency-dependent SNR calculation unit 6 as an estimated noise power spectrum. The frequency-based SNR calculation unit 6 divides by frequency using the input deteriorated speech power spectrum and estimated noise power spectrum, and obtains an acquired SNR (a posteriori SNR).
Is supplied to the estimated a priori SNR calculation unit 7 and the noise suppression coefficient generation unit 8. The acquired SNR is an estimate of the ratio of the pre-emphasized speech including noise and the noise.

The estimated a priori SNR calculator 7 receives the input
From the acquired SNR and the noise suppression coefficient generation unit 8 described later
Supplied suppression coefficient G_nUse the (k) bar to find the true voice pair
Estimate the a priori SNR indicating the noise ratio,
The estimated a priori SNR is fed back to the noise suppression coefficient generator 8.
Let The noise suppression coefficient generator 8 is supplied as an input
The noise suppression coefficient is calculated using the acquired SNR and the estimated a priori SNR.
And the suppression coefficient G _n(k) Estimated innate SN as bar
It returns to the R calculation unit 7 and at the same time is transmitted to the multiple multiplication unit 16.
It The multiple multiplication unit 16 is supplied from the Fourier transform unit 3.
Deteriorated speech amplitude spectrum | Y_n(k) | is the noise suppressor
Suppression coefficient G supplied from the number generation unit 8_n(k) Weight at bar
Emphasized speech amplitude spectrum | X
_n(k) | Obtains the bar and transfers it to the inverse Fourier transform unit 9.
｜ X_n(k) | bar is given by equation (4).

[0012]

[Equation 4]

[0013] The inverse Fourier transform unit 9, enhanced speech amplitude spectrum supplied from multiplexed multiplier 16 | multiply degraded supplied from the bar and a Fourier transform unit 3 audio phase spectrum _{argY n (k) | X n} (k) And emphasized speech spectrum X
_{Find n} (k) bars. That is, the equation (5) is executed.

[0014]

[Equation 5]

Then, the obtained emphasized speech spectrum X
_An inverse Fourier transform is applied to the _n (k) bar, and a time domain sample value sequence (time domain signal) x _n (t) bar (t = 0, 1, ..., K) in which one frame is composed of K samples -1) is transmitted to the frame synthesis unit 10. The frame composition unit 10 calculates K / 2 from two adjacent frames of the x _n (t) bar.
Samples are taken out and superposed, and the emphasized speech x _n (t) hat (t = 0, 1, ...
-1) is obtained. The obtained emphasized speech x _n (t) hat is transmitted to the output terminal 12 as the output of the frame synthesis unit 10.

[0016]

[Equation 6]

Next, the configuration and operation of each part of the noise eliminator shown in FIG. 48 will be further described. Document 1 does not disclose in detail how to realize the voice detection unit. However, as an implementation example of the voice detection unit, "2000
Since March 2013, a collection of lectures by the Acoustical Society of Japan, pp. 321 to 322 "(Reference 2), the method shown in Reference 2 will be described as a conventional method. FIG. 49 is a block diagram showing the configuration of the voice detection unit 4 in FIG. The voice detection unit 4 includes a threshold storage unit 401 and a comparison unit 40.
2, multiplier 404, logarithmic calculation unit 405, power calculation unit 4
06, a weighted addition unit 407, a weight storage unit 408, and a logical NOT circuit 409.

The deteriorated voice amplitude spectrum supplied from the Fourier transform unit 3 in FIG.
Is supplied to. The power calculation unit 406 calculates the sum of the power | Y _n (k) | ² of the degraded speech amplitude spectrum for k = 0 to K−1, and transmits the sum to the logarithmic calculation unit 405.
The logarithmic calculation unit 405 obtains the logarithm of the input deteriorated speech spectral power | Y _n (k) | ² and transmits it to the multiplier 404. The multiplier 404 multiplies the supplied logarithmic value by a constant (for example, 10 times) to obtain the deteriorated voice power Q _n , and the comparison unit 4
02 and the weighted addition unit 407. That is,
The deteriorated voice power Q _n of the nth frame is given by the equation (7).

[0019]

[Equation 7]

The voice detection unit disclosed in Document 2 is
Using the y _n (t) bar which is a time domain sample, Q _n is calculated according to the equation (8).

[0021]

[Equation 8]

However, for example, as shown in "1985, Theory of Digital Signal Processing, Corona Publishing Co., Ltd., pp. 75-76" (Reference 3), equations (8) and (7) are equivalent to each other. Known as the Parseval equation.

The threshold value TH _n is supplied from the threshold value storage unit 401 to the comparison unit 402. The comparing unit 402 compares the output Q _n of the multiplier 404 with the threshold value TH _n , and TH _n >
When Q _n , "1" indicating a voice is output, and when TH _n ≤Q _n , "0" indicating a silence is output. The output of the comparison unit 402 is supplied to the outside as a voice detection flag which is the output of the voice detection unit 4, and at the same time, is supplied to the negative operation circuit 409. The output of the negative operation circuit 409 is supplied to the weighted addition unit 407 as the weighted addition unit control signal 905. The weighted addition unit 407 also includes a threshold storage unit 40.
The threshold value (TH _n-1 ) 902 is supplied from 1 and the weight 903 is supplied from the weight storage unit 408.

The weighted addition unit 407 includes a threshold storage unit 40.
The threshold value (TH _n-1 ) 902 supplied from 1 is selectively updated based on the weighted addition unit control signal 905. The update threshold TH _n is obtained by weighted addition of the threshold (TH _n−1 ) 902 and the deteriorated voice power (Q _n ) 901 using the weight 903 supplied from the weight storage unit 408. The logical NOT circuit 40 calculates the update threshold TH _n.
The weighted addition unit control signal 905 which is the output of 9 is "1"
Is performed only when That is, the threshold value TH _n-1 is updated to TH _n only when there is no sound. The update threshold TH _n obtained by the update is fed back to the threshold storage unit 401 as the update threshold 904.

FIG. 50 is a block diagram showing the configuration of power calculation section 406 included in voice detection section 4 shown in FIG. The power calculation unit 406 includes a separation unit 4061, K multipliers 4062 _{0 to} 4062 _K−1 , and an adder 4063. The deteriorated speech amplitude spectrum | Y _n (k) supplied from the Fourier transform unit 3 in FIG. 48 in the multiplexed state.
| Is separated into frequency-of K samples in the separation unit 4061 are supplied to the multipliers 4062 ₀ ~4062 _K-1. The multipliers 4062 _{0 to} 4062 _K-1 square the respective input signals and transmit the squared signals to the adder 4063. The adder 4063 calculates and outputs the total sum of the input signals.

FIG. 51 is a block diagram showing the structure of the weighted addition unit 407 included in the voice detection unit 4 shown in FIG. The weighted addition unit 407 includes multipliers 4071, 4
073, constant multiplier 4075, adders 4072 and 407
Have 4. 49. The deteriorated voice power (Q _n ) 901 from the multiplier 404 in FIG.
01 to the threshold value (TH _n-1 ) 902, the weight storage unit 408 in FIG. 49 to the weight 903, the logical NOT circuit 409 in FIG. 49 to the weighted addition unit control signal 905,
Each is supplied as an input.

The weight 903 having the value β is the constant multiplier 4
075 and the multiplier 4073. Constant multiplier 40
75 is -β obtained by multiplying the input signal by -1 to adder 4
074 as one input. 1 is supplied to the other input of the adder 4074, and the adder 407
The output of 4 is the sum of the two, 1-β. 1-β is supplied as one input of the multiplier 4071 and is multiplied by the deteriorated voice power (Q _n ) 901 which is the other input, and the product (1-β) Q _n is transmitted to the adder 4072. It

On the other hand, the multiplier 4073, supplied β and the threshold value (TH _n-1) 902 is multiplied as the weight 903, βTH _n-1 which is a product is transmitted to the adder 4072. The adder 4072 outputs the sum of βTH _n−1 and (1−β) Q _n as the update threshold (TH _n ) 904. The update threshold value TH _n is calculated only when the weighted adder control signal 905 is equal to “1”. That is, the function of the weighted addition unit 407 is that the threshold value TH _{n −1} is used when there is no sound.
Is calculated to obtain TH _n , which can be expressed by equation (9).

[0029]

[Equation 9]

The multiplying unit 17 in FIG. 48 will be described. FIG. 52 is a block diagram showing the configuration of the multiple multiplication unit 17. The multiplex multiplication unit 17 includes K multipliers 1701.
_{It has 0 to} 1701 _K-1 , demultiplexing units 1702 and 1703, and a multiplexing unit 1704. The deteriorated voice amplitude spectrum supplied from the Fourier transform unit 3 in FIG. 48 in the multiplexed state is separated into K samples for each frequency in the separation units 1702 and 1703, and each of them is multiplied by 1701 ₀ to 1701.
Supplied to 1701 _K-1 . Multipliers 1701 _{0 to} 170
1 _K-1 squares the respective input signals and transmits them to multiplexing section 1704. Multiplexing section 1704 multiplexes the input signal and outputs it as a degraded voice power spectrum.

The estimated noise calculation unit 51 in FIG. 48 will be described. FIG. 53 is a block diagram showing the configuration of the estimated noise calculation unit 51. The estimated noise calculation unit 51 includes the separation unit 5
02, a multiplexing unit 503, and K frequency _- dependent estimated noise calculation units 514 _{0 to} 514 _K−1 . The voice detection flag supplied from the voice detection unit 4 in FIG. 48 and the count value supplied from the counter 13 in FIG. 48 are transmitted to the frequency _- dependent estimated noise calculation units 514 _{0 to} 514 _K-1 . Figure 4
The deteriorated voice power spectrum supplied from the multiplex multiplier 17 in 8 is transmitted to the separator 502.

The separating unit 502 separates the deteriorated speech power spectrum supplied in the multiplexed state into components corresponding to K frequencies, and the estimated noise calculating units 514 _{0 to} 514 _K-1 for each frequency. introduce. The frequency _- dependent estimated noise calculation units 514 _{0 to} 514 _K-1 calculate a noise power spectrum using the deteriorated speech power spectrum supplied from the separation unit 502, and transmit the noise power spectrum to the multiplexing unit 503. The calculation of the noise power spectrum is controlled by the count value and the value of the voice detection flag, and is executed only when a predetermined condition is satisfied. The multiplexing unit 503 multiplexes the supplied K noise power spectrum values and outputs them as an estimated noise power spectrum.

FIG. 54 is a block diagram showing the configuration of the frequency-dependent estimated noise calculation unit 514 included in the estimated noise calculation unit 51 shown in FIG. The noise estimation disclosed in Reference 2 updates the noise estimation value in a silent section, and uses the instantaneous value of the estimated noise averaged by the recursive filter as the noise estimation value. On the other hand, "1998
May, IEE Transactions on Speech and Audio Processing, Volume 6, Issue 3 (IEEE TRANS-ACTIONS ON SPEECHAN
D AUDIO PROCESSING, VOL.6, NO.3, PP.287-292, MAY,
1998), pp. 287-292 ”(Reference 4), it is described that the instantaneous values of the estimated noise are averaged and used. This is a transversal type filter (configuration using shift register) instead of the cyclic type.
It suggests the realization of averaging using. Since the functions are the same in both implementations, the method disclosed in Document 4 will be described here.

The frequency-dependent estimated noise calculation unit 514 includes an update determination unit 521, a register length storage unit 5941, and a switch 50.
44, shift register 5045, adder 5046, minimum value selection unit 5047, division unit 5048, counter 5049
Have. The switch 5044 is supplied with the degraded voice power spectrum for each frequency from the separating unit 502 in FIG. When the switch 5044 closes the circuit, the frequency-dependent deteriorated voice power spectrum is transmitted to the shift register 5045. Shift register 5045
Shifts the storage value of the internal register to the adjacent register according to the control signal supplied from the update determination unit 521. The shift register length is the register length storage unit 5 described later.
Equal to the value stored in 941. Shift register 5
All 045 register outputs are provided to adder 5046. The adder 5046 adds all the supplied register outputs and transmits the addition result to the division unit 5048.

On the other hand, the update determination unit 521 is supplied with the count value and the voice detection flag. Update determination unit 521
Is always "1" until the count value reaches a preset value, and when it reaches "1" when the voice detection flag is "0" (silence), otherwise. "0"
Is output and transmitted as a control signal to the counter 5049, the switch 5044, and the shift register 5045. The switch 5044 closes the circuit when the control signal supplied from the update determination unit 521 is “1”, and opens when the control signal is “0”. The counter 5049 increments the count value when the control signal supplied from the update determination unit 521 is “1”,
When it is "0", it is not changed. Shift register 5045
Is a signal sample supplied from the switch 5044 when the signal supplied from the update determination unit 521 is “1”.
Simultaneously with the sampling, the value stored in the internal register is shifted to the adjacent register.

The minimum value selector 5047 has a counter 50.
The output of 49 and the output of the register length storage unit 5941 are supplied. The minimum value selection unit 5047 selects the smaller one of the supplied count value and the register length and transmits it to the division unit 5048. The division unit 5048 uses the adder 504.
The added value of the frequency-dependent deteriorated speech power spectrum supplied from 6 is divided by the count value or the smaller value of the register length, and the quotient is the frequency-dependent estimated noise power spectrum λ _n (k)
Output as. B _n (k) (n = 0, 1, ..., N−
Letting 1) be the sample value of the degraded voice power spectrum stored in the shift register 5045, λ _n (k) is given by equation (10).

[0037]

[Equation 10]

However, N is the smaller value of the count value and the register length. Since the count value starts from zero and monotonically increases, division is first performed by the count value and later division is performed by the register length. On the other hand, the number of registers in which values are actually stored is equal to the count value when the count value is smaller than the register length, and is equal to the register length when the count value is larger than the register length. Therefore, the added value of the frequency-dependent deteriorated voice power spectrum supplied from the adder 5046 is divided by the number of registers in which the values are actually stored. When the count value is larger than the register length, the average value of the values stored in the shift register 5045 is calculated. The result of this calculation is the estimated noise power spectrum for each frequency.

FIG. 55 is a block diagram showing the configuration of the update determination unit 521 included in the frequency-dependent estimated noise calculation unit 514 shown in FIG. The update determination unit 521 includes a logical NOT circuit 5202, a comparison unit 5203, a threshold value storage unit 5204, and a logical sum calculation unit 5211. The count value supplied from the counter 13 in FIG. 48 is transmitted to the comparison unit 5203. The threshold value output from the threshold value storage unit 5204 is also transmitted to the comparison unit 5203. The comparison unit 5203 compares the supplied count value with the threshold value, and transmits “1” to the logical sum calculation unit 5211 when the count value is smaller than the threshold value and “0” when the count value is larger than the threshold value. .

On the other hand, the supplied voice detection flag is transmitted to the logical NOT circuit 5202. Logical NOT circuit 5202
Calculates the logical negation value of the input signal and transmits it to the logical sum calculation unit 5211. That is, the voice-over portion whose voice detection flag is "1" is "0", and the voice-over portion whose voice detection flag is "0" is "1".
It will be transmitted to 1. As a result, the logical sum calculation unit 52
The output of 11 is "1" when the voice detection flag is a silent part where the voice detection flag is "0" or when the count value is smaller than the threshold value, and the switch 5044 in FIG. 54 is closed and the counter 5049 is counted up.

In the frequency-dependent SNR calculation unit 6 in FIG.
explain about. FIG. 56 shows the SNR calculation unit 6 for each frequency.
It is a block diagram which shows a structure. Frequency SNR calculation unit 6
Is K division units 601₀ ~ 601_K-1 , Separation unit 60
2, 603, and a multiplexing unit 604. In FIG. 48
Deteriorated voice power spectrum supplied from multiplex multiplier 17
Are transmitted to the separating unit 602. Estimation in FIG. 48
Estimated noise power spectrum supplied from the noise calculator 51
Are transmitted to the separating unit 603. Degraded voice power spec
In the demultiplexing unit 602, Kuttle calculates the estimated noise power spectrum.
In the separation unit 603, the tor is paired with each frequency component.
The corresponding K samples are separated, and each division unit 601₀ 
~ 601_K-1 Is supplied to. Division unit 601₀ ~ 601
_K-1 Then, according to equation (11),
War spectrum ｜ Y_n(k) |²Estimated noise power spectrum
Tol λ_nDivide by (k) and divide by frequency SNRγ_nFind (k)
Therefore, it is transmitted to the multiplexing unit 604. The multiplexing unit 604 transmits
Reached K SNRγ for each frequency _n(k) is multiplexed,
Output as an acquired SNR.

[0042]

[Equation 11]

The estimated a priori SNR calculator 7 in FIG.
Will be described. FIG. 57 is a block diagram showing the configuration of the estimated a priori SNR calculation unit 7. The estimated a priori SNR calculation unit 7 includes a multiple range limiting processing unit 701, an a posteriori SNR storage unit 702, a suppression coefficient storage unit 703, a multiple multiplication unit 70.
4, 705, weight storage unit 706, multiple weighted addition unit 7
07 and an adder 708. Acquired SNR γ _n (k) supplied from the frequency-dependent SNR calculation unit 6 in FIG.
(K = 0, 1, ..., K−1) is transmitted to one terminal of the adder 708 and the acquired SNR storage unit 702. The acquired SNR storage unit 702 stores the acquired SNR γ _n (k) in the n-th frame and transfers the acquired SNR γ _n-1 (k) in the ( _n-1 ) th frame to the multiplex multiplication unit 705.

The suppression coefficient G _n (k) bar (k = 0, 1, ..., And, supplied from the noise suppression coefficient generator 8 in FIG.
K-1) is transmitted to the suppression coefficient storage unit 703. The suppression coefficient storage unit 703 stores the suppression coefficient G in the nth frame.
_{The n} (k) bar is stored and the suppression coefficient G _n-1 (k) bar in the ( _n-1 ) _th frame is transmitted to the multiplex multiplication unit 704. The multiple multiplication unit 704 squares the supplied G _n-1 (k) bar to obtain a G ² _n-1 (k) bar, and transmits it to the multiple multiplication unit 705. The multiple multiplication unit 705 calculates the G ² _n-1 (k) bar and γ.
_n−1 (k) is multiplied by k = 0, 1, ..., K−1 to obtain G ² _n−1 (k) bar γ _n−1 (k), and the result is multiplied. The past estimated SNR 922 is transmitted to the weighted addition unit 707. The configurations of the multiplex multipliers 704 and 705 are the same as those of the multiplex multiplier 17 already described with reference to FIG. 52, and thus detailed description thereof will be omitted.

-1 is supplied to the other terminal of the adder 708, and the addition result γ _n (k) -1 is transmitted to the multiple range limiting processing unit 701. The multiple range limitation processing unit 701
The addition result γ _n (k) -1 supplied from the adder 708 is calculated by the range limiting operator P [•], and the result P [γ _n (k) -1] is added to the multi-weighted addition unit. It is transmitted to 707 as the instantaneous estimated SNR 921. However, P [x]
Is defined by equation (12).

[0046]

[Equation 12]

A weight 923 is supplied from the weight storage unit 706 to the multiple weighted addition unit 707. The multiple weighted addition unit 707 obtains an estimated a priori SNR 924 by using the supplied instantaneous estimated SNR 921, past estimated SNR 922, and weight 923. If the weight 923 is α and the ξ _n (k) hat is the estimated a priori SNR,
ξ _n (k) hat is calculated by the equation (13). Here, the initial value (n = 0) of the first term on the right side is set to γ ₋₁ (k) G
² ₋₁ (k) bar = 1.

[0048]

[Equation 13]

FIG. 58 shows the estimated innate SN shown in FIG.
6 is a block diagram showing a configuration of a multiple range limiting processing unit 701 included in the R calculation unit 7. FIG. Multiple range limitation processing unit 701
Is a constant storage unit 7011 and K maximum value selection units 7012.
_{0 to} 7012 _K-1 , demultiplexing unit 7013, multiplexing unit 7014
Have. Γ _n (k) −1 is supplied to the separating unit 7013 from the adder 708 in FIG. 57. Separation unit 701
3 separates the supplied γ _n (k) -1 into K frequency components, and the maximum value selection units 7012 _{0 to} 7012 respectively.
Supply to one input of _K-1 . Maximum value selection unit 7012 ₀
~ 7012 _{K-1 has} the other input to the constant storage unit 7011.
Is being supplied by zero. Maximum value selection unit 7012 ₀ ~
7012 _K-1 compares γ _n (k) -1 with zero and transfers the larger value to multiplexing section 7014. This maximum value selection operation is equivalent to executing Expression (12). The multiplexing unit 7014 multiplexes these values and outputs them.

FIG. 59 shows the estimated innate SN shown in FIG.
6 is a block diagram showing the configuration of a multiple weighted addition unit 707 included in the R calculation unit 7. FIG. Multi-weighted addition unit 707
Are _K weighted addition units 7071 _{0 to} 7071 _K-1 ,
It has demultiplexing units 7072 and 7074 and a multiplexing unit 7075.

The demultiplexing unit 7072 is provided with the multiplexing shown in FIG.
From the range limiting processing unit 701, P [γ _n(k) -1] is instantaneous
Supplied as an estimated SNR 921. Separation unit 7072
Is P [γ_n(k) -1] is separated into K frequency components
Frequency-dependent instantaneous estimation SNR 921₀ ~ 921_K-1 age
Respectively, the weighted addition unit 7071₀ ~ 7071_K-1 
Communicate to. The demultiplexing unit 7074 includes the multiplex shown in FIG.
From the multiplication unit 705, G² _n-1(K) bar γ_n-1(k) is the past
Of the estimated SNR 922. Separation unit 7074
Is G² _n-1(K) bar γ_n-1(k) is K frequency components
And separate past estimated SNR922 by frequency₀ ~ 922
_K-1 Respectively, the weighted addition unit 7071₀ ~ 70
71_K-1 Communicate to. On the other hand, the weighted addition unit 7071₀ 
~ 7071_{K- 1} Is also provided with a weight 923. Weight
Adder 7071₀ ~ 7071_K-1 According to equation (13)
The weighted addition represented by
SNR924₀ ~ 924_K-1 Is transmitted to the multiplexing unit 7075.
Reach The multiplexing unit 7075 uses the frequency-specific estimated innate SN.
R924₀ ~ 924_K-1 The estimated innate SNR
Output as 924. Weighted addition unit 7071₀ ~ 7
071_K-1 The configuration and operation of are already explained using FIG.
Since it is equal to the weighted addition unit 407, detailed description will be omitted.
I will omit it. However, the weighted addition calculation is always performed.

The noise suppression coefficient generator 8 in FIG. 48 will be described. FIG. 60 is a block diagram showing the configuration of the noise suppression coefficient generator 8. The noise suppression coefficient generation unit 8 uses K
Individual suppression coefficient search units 801 _{0 to} 801 _K-1 , separation unit 80
2, 803 and a multiplexing unit 804. The separation unit 802 is supplied with the acquired SNR from the frequency-based SNR calculation unit 6 in FIG. The separation unit 802 separates the supplied acquired SNR into K frequency components, and transfers the _K frequency components to the suppression coefficient search units 801 _{0 to} 801 _K-1 . Separation part 8
03 is supplied with the estimated a priori SNR from the estimated a priori SNR calculator 7 in FIG. Separation unit 803, the supplied estimated apriori SNR is separated into K frequency-component, and transmits the respective spectral gain search unit 801 ₀ ~801 _K-1. The suppression coefficient search units 801 _{0 to} 801 _K-1 search for suppression coefficients corresponding to the supplied acquired SNR and estimated a priori SNR, and transmit the search results to the multiplexing unit 804.
The multiplexing unit 804 multiplexes the supplied suppression coefficient and outputs it.

FIG. 61 is a block diagram of the suppression coefficient search units 801 _{0 to} 801 _K-1 included in the noise suppression coefficient generation unit 8 shown in FIG.
3 is a block diagram showing the configuration of FIG. Suppression coefficient search unit 801
Is a suppression coefficient table 8011 and address conversion unit 801.
2,8013. The address conversion unit 8012 has
From the separation unit 802 in FIG.
R is supplied. The address conversion unit 8012 converts the supplied frequency-dependent acquired SNR into a corresponding address,
This is transmitted to the suppression coefficient table 8011. The address conversion unit 8013 is supplied with the estimated a priori SNR for each frequency from the separation unit 803 in FIG. Address conversion unit 8
013 converts the supplied inferred innate SNR for each frequency into a corresponding address, and transfers it to the suppression coefficient table 8011. The suppression coefficient table 8011 outputs the suppression coefficient stored in the area corresponding to the address supplied from the address conversion unit 8012 and the address conversion unit 8013 as the suppression coefficient for each frequency. Here, the suppression coefficient derived by assuming the background noise according to a specific statistical model is used.

[0054]

As described above, in the conventional noise removing apparatus and method, the noise suppression is performed by using the suppression coefficient derived assuming the background noise according to the specific statistical model. The noise that did not comply could not be effectively removed. For this reason, a sufficiently high quality of the emphasized voice cannot be achieved. Further, in the conventional noise removing apparatus and method, the emphasized speech is obtained by superposing and adding the signal samples extracted from two adjacent frames of the time domain signal obtained by the inverse Fourier transform. On the other hand, the window function applied to the time domain signal before the Fourier transform has been designed so that the input is reproduced at the output when the noise suppression processing is not performed. Therefore, if the signal samples that are subject to superposition addition are suppressed with different suppression coefficient values in adjacent frames, discontinuity occurs in the signal samples at frame boundaries, and noise generated in the output signal causes sound quality. Had deteriorated.

As described above, the conventional noise removing apparatus and method have a problem that it is not possible to obtain an emphasized voice having excellent sound quality. The present invention has been made to solve such a problem, and an object of the present invention is to provide a noise removing apparatus and method capable of obtaining an emphasized voice with excellent sound quality.

[0056]

In order to achieve such an object, the noise removing method of the present invention generates pseudo noise based on an input signal and obtains it by injecting this pseudo noise. It is characterized by using the suppressed suppression coefficient. By injecting the above-mentioned pseudo noise when determining the suppression coefficient, the suppression coefficient derived assuming the background noise according to the specific statistical model can be corrected according to the input signal.

More specifically, the noise removal method of the present invention transforms an input signal into a frequency domain signal, calculates pseudo first noise based on the frequency domain signal, and calculates the first noise. Is added to the frequency domain signal, the signal-to-noise ratio is obtained using the frequency-domain signal to which the first noise is added, the suppression coefficient is determined based on this signal-to-noise ratio, and the frequency-domain signal is calculated using this suppression coefficient. Is weighted, and the weighted frequency domain signal is converted into a time domain signal to obtain an output signal from which noise is removed from the input signal.

In this noise removing method, the first noise may be added to the frequency domain signal selectively depending on the nature of the input signal. Thereby, for example, the first noise can be added only when a signal including noise that does not follow the statistical model used for deriving the suppression coefficient is input, and the suppression coefficient can be selectively corrected. Here, the stationarity of the signal may be used as the property of the input signal. In other words, based on how much the characteristics of the signal, such as average power and spectral shape, change with time,
The first noise may be added. As the stationarity of the signal, the number of zero crossings at which the amplitude of the input signal becomes zero may be used, or the high frequency power of the frequency domain signal showing the correlation with the number of zero crossings may be used.

Further, the second noise included in the frequency domain signal is estimated based on the frequency domain signal obtained by converting the input signal, and the power of the first noise is calculated using the second noise and the frequency domain signal. You may decide. Also, the second noise included in the frequency domain signal is estimated based on the frequency domain signal obtained by converting the input signal, the first noise is calculated using the second noise and the frequency domain signal, and the first noise is calculated. The signal-to-noise ratio may be obtained using the sum of the first noise and the frequency domain signal and the sum of the first noise and the second noise. Here, the frequency domain signal obtained by converting the input signal may be weighted, and the second noise may be estimated based on the weighted frequency domain signal.

Further, the noise removing method of the present invention converts an input signal into a frequency domain signal, obtains a signal to noise ratio using this frequency domain signal, and corrects this signal to noise ratio based on the frequency domain signal. Then, the suppression coefficient is determined based on the corrected signal-to-noise ratio, the frequency domain signal is weighted using this suppression coefficient, and the weighted frequency domain signal is converted into the time domain signal to obtain the input signal. It is characterized in that an output signal from which noise is removed is obtained.

In this noise elimination method, the signal-to-noise ratio may be corrected selectively depending on the nature of the input signal. As a result, the signal-to-noise ratio can be corrected and the suppression coefficient can be selectively corrected only when a signal including noise that does not follow the statistical model used for deriving the suppression coefficient is input. Here, as the nature of the input signal,
Signal stationarity may be used. In other words, the signal-to-noise ratio may be corrected based on how much the characteristics of the signal, such as the average power and the spectral shape, change with time. As the stationarity of the signal, the number of zero crossings at which the amplitude of the input signal becomes zero may be used, or the high frequency power of the frequency domain signal showing the correlation with the number of zero crossings may be used.

Further, the noise contained in the frequency domain signal is estimated based on the frequency domain signal obtained by converting the input signal, and the correction amount of the signal-to-noise ratio is determined using this noise and the frequency domain signal. Good. Further, the noise included in the frequency domain signal is estimated based on the frequency domain signal obtained by converting the input signal, the addition signal is obtained using this noise and the signal-to-noise ratio, and the sum of the addition signal and the frequency domain signal, Alternatively, the signal-to-noise ratio may be corrected by recalculating the signal-to-noise ratio using the sum of the added signal and noise. Here, the frequency domain signal obtained by converting the input signal may be weighted, and the noise may be estimated based on the weighted frequency domain signal.

Further, in the above-mentioned noise removing method,
The suppression coefficient may be corrected based on the frequency domain signal, and the frequency domain signal may be weighted using the corrected suppression coefficient. As a result, it is possible to prevent residual noise that occurs due to insufficient suppression when the signal-to-noise ratio is low, and to prevent sound quality deterioration due to audio distortion that occurs due to excessive suppression when the signal-to-noise ratio is high. In the noise removal method described above, windowing processing may be performed on the time domain signal obtained by converting the frequency domain signal.

Further, the noise removing method of the present invention transforms the input signal into a frequency domain signal, estimates the second noise contained in the frequency domain signal based on this frequency domain signal,
On the other hand, a pseudo first noise is calculated based on the frequency domain signal, and a value corresponding to the noise obtained by adding the first noise to the second noise is subtracted from the frequency domain signal to emphasize the frequency domain speech. Is obtained, and the emphasized speech is converted into a time domain signal to obtain an output signal in which noise is removed from the input signal.

In this noise removing method, the first noise may be added to the second noise selectively according to the property of the input signal. Thereby, for example, the first noise can be added only when a signal including noise that does not follow the statistical model used for deriving the suppression coefficient is input, and the enhanced voice can be selectively corrected. Here, the stationarity of the signal may be used as the property of the input signal. In other words, based on how much the characteristics of the signal, such as the average power and the spectral shape, change with time, the first
Noise may be added. The stationarity of the signal is
The number of zero crossings at which the amplitude of the input signal is zero may be used, or the high frequency power of the frequency domain signal that correlates with the number of zero crossings may be used.

The power of the first noise may be determined using the frequency domain signal and the second noise.
Further, the frequency domain signal obtained by converting the input signal may be weighted, and the second noise may be estimated based on the weighted frequency domain signal. Here, the signal-to-noise ratio is obtained using the frequency domain signal obtained by converting the input signal,
Weights may be obtained by using the signal-to-noise ratio, and the frequency domain signals may be weighted by using the weights. As a result, the influence of the voice component included in the frequency domain signal can be reduced and the estimation of the second noise can be performed with higher accuracy. For example, the signal-to-noise ratio is obtained by using the frequency-domain signal obtained by converting the input signal, the signal-to-noise ratio is processed by the non-linear processing function to obtain the weight, and the frequency-domain signal is weighted by using the weight. You may In the noise removal method described above, windowing processing may be performed on the time domain signal obtained by converting the emphasized speech in the frequency domain.

Further, the noise removing method of the present invention is characterized in that windowing processing is applied to the time domain signal obtained by converting the emphasized speech in the frequency domain. When two adjacent frames of the time domain signal obtained by converting the emphasized speech in the frequency domain are superposed and added, even if the signal sample to be superposed and added is suppressed by different suppression coefficient values in each frame, The continuity of the signal samples at the frame boundaries can be improved by windowing each frame to reduce the amplitude of the signal samples at the frame boundaries.

More specifically, the noise removal method of the present invention converts an input signal into a frequency domain signal, obtains a signal to noise ratio using this frequency domain signal, and suppresses the signal based on this signal to noise ratio. By defining a coefficient, weighting the frequency domain signal using this suppression coefficient, converting the weighted frequency domain signal into a time domain signal, and applying a windowing process to this time domain signal. , The output signal is obtained by removing noise from the input signal.

Further, the noise removing method of the present invention transforms the input signal into a frequency domain signal, estimates the second noise contained in the frequency domain signal based on this frequency domain signal,
A value corresponding to the second noise is subtracted from the frequency domain signal to obtain an emphasized voice in the frequency domain, the emphasized voice is converted into a time domain signal, and the time domain signal is subjected to windowing processing to obtain the emphasized voice from the input signal. It is characterized in that an output signal from which noise is removed is obtained.

Further, the noise removing apparatus of the present invention includes a first windowing processing section which performs windowing processing on an input signal and outputs the windowed signal, and an input signal which is windowed by the first windowing processing section. A conversion unit that converts to a frequency domain signal and separates and outputs an amplitude component and a phase component, and an estimated noise calculation that estimates and outputs the second noise included in the frequency domain signal based on the amplitude component of the frequency domain signal. Section, an injection noise calculation section for calculating and outputting pseudo first noise using the second noise and the amplitude component of the frequency domain signal, and adding the amplitude components of the first noise and the frequency domain signal. A first adder for outputting
A second adder for adding and outputting the first noise and the second noise, and a first signal-to-noise ratio for receiving the output signal of the first adder and the output signal of the second adder To output and output
Signal-to-noise ratio calculation unit, a suppression coefficient generation unit that determines and outputs a suppression coefficient based on the first signal-to-noise ratio, and outputs by weighting the amplitude component of the frequency domain signal using the suppression coefficient. A first multiplication unit; an inverse transformation unit that transforms the amplitude component of the frequency domain signal and the phase component of the frequency domain signal weighted by the first multiplication unit into a time domain signal and outputs the time domain signal; At least a second windowing processing unit that performs windowing processing and outputs is provided.

Here, the injection noise calculator calculates the number of zero crossings at which the input signal is input and the amplitude of the input signal is zero, and outputs a control signal according to the calculation result. , And a switch for selectively setting the first noise to zero by the control signal input from the zero crossing calculation unit. The injection noise calculation unit calculates the high frequency power of the amplitude component of the frequency domain signal input from the conversion unit, and outputs the control signal according to the calculation result, and the high frequency power calculation unit. A switch for selectively setting the first noise to zero according to a control signal input from the calculator may be included.

Further, the above-mentioned noise removing apparatus weights the amplitude component of the frequency domain signal, outputs the obtained weighted amplitude component to the estimated noise calculation unit, and the estimated noise calculation unit is based on the weighted amplitude component. It may further include a weighted deteriorated speech calculation unit that estimates the second noise. Here, the weighted deteriorated speech calculation unit calculates a second signal-to-noise ratio using the amplitude component of the frequency domain signal and outputs the second signal-to-noise ratio calculation unit, and the second signal-to-noise ratio calculation unit. A non-linear processing unit that processes the second signal-to-noise ratio input from the noise ratio calculation unit by a non-linear function to obtain and output weights, and an amplitude component of the frequency domain signal using the weights input from the non-linear processing unit. May be weighted and output to the estimated noise calculation section and a second multiplication section may be included.

Further, the above-described noise removing apparatus corrects the suppression coefficient input from the suppression coefficient generating section based on the frequency domain signal, outputs the correction coefficient to the first multiplying section, and corrects it to the first multiplying section. A suppression coefficient correction unit that weights the amplitude component of the frequency domain signal using the suppression coefficient may be further included.

The noise removing apparatus of the present invention further includes a first windowing processing section for performing windowing processing on the input signal and outputting the input signal, and an input signal subjected to the windowing processing by the first windowing processing section. A conversion unit that converts the signal into a frequency domain signal and outputs the signal after separating it into an amplitude component and a phase component, and a first signal-to-noise ratio that calculates and outputs a first signal-to-noise ratio using the amplitude component of the frequency domain signal. A calculation unit, an estimation noise calculation unit that estimates and outputs noise included in the frequency domain signal based on the amplitude component of the frequency domain signal, and a first signal-to-noise ratio using the noise and the amplitude component of the frequency domain signal. A signal-to-noise ratio correction unit that corrects and outputs a corrected signal-to-noise ratio, a suppression coefficient generation unit that determines and outputs a suppression coefficient based on the corrected signal-to-noise ratio, and a frequency-domain signal Output weighted amplitude components 1 multiplication unit, an inverse transformation unit that transforms the amplitude component of the frequency domain signal and the phase component of the frequency domain signal weighted by the first multiplication unit into a time domain signal and outputs the time domain signal, and a window for the time domain signal It is characterized by comprising at least a second window cliff processing unit for carrying out cliff processing.

Here, the signal-to-noise ratio correction unit calculates the number of zero crossings at which the input signal is input and the amplitude of the input signal becomes zero, and outputs a control signal according to the calculation result. A switch for selectively setting the correction signal-to-noise ratio to the same value as the pre-correction first signal-to-noise ratio by the control signal input from the determination unit may be included. Further, the signal-to-noise ratio correction unit calculates the high frequency power of the amplitude component of the frequency domain signal input from the conversion unit and outputs a control signal according to the calculation result, and a determination unit input from this determination unit. A switch for selectively setting the correction signal-to-noise ratio to the same value as the first signal-to-noise ratio before correction by the generated control signal may be included.

Further, the above-mentioned noise removing apparatus weights the amplitude component of the frequency domain signal, outputs the obtained weighted amplitude component to the estimated noise calculation unit, and the estimated noise calculation unit based on the weighted amplitude component. It may further include a weighted deteriorated speech calculation unit for estimating noise. Here, the weighted deteriorated speech calculation unit calculates a second signal-to-noise ratio using the amplitude component of the frequency domain signal and outputs the second signal-to-noise ratio calculation unit, and the second signal-to-noise ratio calculation unit. A non-linear processing unit that processes the second signal-to-noise ratio input from the noise ratio calculation unit by a non-linear function to obtain and output weights, and an amplitude component of the frequency domain signal using the weights input from the non-linear processing unit. May be weighted and output to the estimated noise calculation section and a second multiplication section may be included.

Further, the above-mentioned noise removing apparatus corrects the suppression coefficient input from the suppression coefficient generating section based on the frequency domain signal, outputs it to the first multiplying section, and corrects it to the first multiplying section. A suppression coefficient correction unit that weights the amplitude component of the frequency domain signal using the suppression coefficient may be further included.

[0078]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings.

(First Embodiment) FIG. 1 is a block diagram showing the overall configuration of a first embodiment of a noise eliminator of the present invention. This noise removing device and the conventional noise removing device shown in FIG. 48 are the same except for the windowing processing unit 22, injection noise calculation unit 55, and adders 56 and 57.
The same parts are designated by the same reference numerals. Less than,
The difference will be mainly described in detail.

The windowing processing section 22 includes an inverse Fourier transform section 9
The time domain sample value series x _n (t) bar supplied from the above is multiplied by the window function h (t), and the product h (t) x _n (t) bar is transmitted to the frame synthesis unit 10. Frame synthesizer 1
For 0, K / 2 samples are taken out from two adjacent frames of the h (t) × _n (t) bar and overlapped, and the expression (1
4), the emphasized speech x _n (t) hat (t = 0,
, ..., K / 2-1). Obtained emphasized voice x
_{The n} (t) hat is transmitted to the output terminal 12 as the output of the frame synthesis unit 10.

[0081]

[Equation 14]

If the overlap is not 50% but M samples and the frame length is L samples (M <L),
The emphasized speech x _n (t) hat is obtained by the equation (15).
In accordance with this, the frame division unit is also modified.

[0083]

[Equation 15]

As described above, the symmetric window function is used for real signals. Further, the window function is designed so that the input signal and the output signal when the suppression coefficient is set to 1 match each other except for a calculation error. For any window function that satisfies these conditions, w (t), h
It can be used as (t). As an example,
A function (root Hanning window) obtained by square rooting the Hanning window can be mentioned. There are other window functions that satisfy these conditions, but the details are omitted. Even if the x _n-1 (t) bar and the x _n (t) bar that form two adjacent frames are suppressed with different suppression coefficient values in each frame, x x
Each of the _n-1 (t) bar and the x _n (t) bar is multiplied by the window function h (t) described above to obtain x _n-1 (t) at the frame boundary.
By reducing the amplitude of the bar and x _n (t) bar,
It is possible to improve continuity at frame boundaries and reduce noise generation. Therefore, it is possible to suppress the sound quality deterioration due to noise and obtain an emphasized sound with excellent sound quality.

The injection noise calculation unit 55 uses the deteriorated voice power spectrum and the estimated noise power spectrum supplied from the multiplex multiplication unit 17 and the estimated noise calculation unit 51, respectively, to generate pseudo noise (first noise). ),
It transmits to the adders 56 and 57. The adder 56 adds the injection noise obtained by the injection noise calculation unit 55 to the estimated noise power spectrum supplied from the estimation noise calculation unit 51, and transmits the sum to the frequency-specific SNR calculation unit 6. Adder 57
Adds the injection noise obtained by the injection noise calculation unit 55 to the degraded speech power spectrum supplied from the multiplex multiplication unit 17, and transmits the sum to the frequency-dependent SNR calculation unit 6.

FIG. 2 is a block diagram showing the configuration of the injection noise calculator 55. The injection noise calculator 55 includes an SNR calculator 551, a threshold calculator 552, and an injection level calculator 5.
With 53. The deteriorated voice power spectrum supplied from the multiplex multiplication unit 17 in FIG. 1 is the SNR calculation unit 551.
Be transmitted to. The estimated noise power spectrum supplied from the estimated noise calculation unit 51 in FIG.
51 and the threshold value calculation unit 552. The SNR obtained by the SNR calculator 551 and the threshold obtained by the threshold calculator 552 are supplied to the injection level calculator 553. The injection level calculation unit 553 supplies the supplied SNR.
And calculate the noise level to be injected according to the threshold,
The signal corresponding to that level is output as injection noise.

If the noise to be injected is W _n (k), then W
_n (k) is set to take a smaller value as the SNR increases. As a relation between such SNR and W _n (k), S
When NR is larger than the first threshold TH ₁ , it takes the first value W ₁ , and SNR is the second threshold TH ₂ (<TH
When it is smaller than ₁ ), the second value W ₂ (> W ₁ ) is taken and the SNR is the first threshold TH ₁ and the second threshold T.
When taking an intermediate value of H ₂ , W corresponding to SNR
It is possible to consider a function such that _n (k) becomes small.
The simplest example is, as shown in FIG. 3, when the SNR takes an intermediate value between the _first threshold value TH ₁ and the second threshold value TH ₂ , the first value W ₁ It is a function that changes linearly up to the value W ₂ .

First and second threshold values TH₁, TH₂Is German
Can be determined as the second threshold TH₂To
First threshold TH₁Set to a constant multiple of to simplify the calculation
You can also measure. Similarly, independent decisions can be made
W that can_nThe first and second values W of (k)₁, W₂Is also the second value
W₂Is the first value W₁It can be set to a constant multiple of.
Also, W_nThe first and second values W of (k)₁, W₂Is an estimated miscellaneous
It can be determined according to the sound level. Estimated noise
W when the level is high_nThe first and second values W of (k) ₁, W₂
Decrease, increase when low. Like this W_n(k)
The first and second values W of₁, W₂To set the same S
Easier when the estimated noise level is higher than the NR value
Small W_n(k) can be set. In this case, the injection level meter
Configuration for supplying the estimated noise power spectrum to the calculation unit 553
Needless to say,

Further, the thresholds TH ₁ and TH ₂ can also be determined corresponding to the estimated noise level. The thresholds TH ₁ and TH ₂ are reduced when the estimated noise level is high, and increased when the estimated noise level is low. Thus, the threshold value TH ₁ ,
By setting TH ₂ , a smaller W _n (k) can be easily set for the same SNR value when the estimated noise level is higher. The reason why W _n (k) is made smaller as the estimated noise level is higher is that the conventional suppression coefficient is almost appropriate and the amount of correction of the suppression coefficient by noise injection is smaller when the estimated noise level is higher. As a result, when the original amount of suppression is small and the residual noise is likely to be perceived, it is possible to suppress the component having a medium amplitude relatively greatly.
An improvement in subjective sound quality can be achieved.

In the above description, the noise to be injected is W
_n (k), and an example of injecting different noises into each frequency component has been described. Actually, in the deteriorated voice power spectrum and the estimated noise power spectrum supplied to the injection noise calculation unit 55, values corresponding to all frequency components are multiplexed. Therefore, the SNR obtained by the SNR calculation unit 551 and the number of threshold values obtained by the threshold value calculation unit 552 correspond to the number of frequency components. However, these SNR and threshold value may be set in common for all frequency components.

As an example, the deteriorated speech power spectrum and the estimated noise power spectrum are added to all frequency components and summed, the ratio thereof is set as a common SNR, and the average value of the estimated noise power spectrum is used. The threshold can be calculated. In that case, the SNR calculation unit 5
51 and the threshold calculation unit 552, instead of separating the values corresponding to the respective frequency components and then calculating the SNR and the threshold using the individual values, the sum and average values are used,
A common SNR and threshold will be calculated for all frequency components. These values are the frequency-dependent SNR calculation unit 6
Be transmitted to.

In the frequency-dependent SNR calculation unit 6, equation (11)
Instead of
Calculate _n (k).

[0093]

[Equation 16]

Referring to the equation (16), since | Y _n (k) | ² > λ _n (k) in the region of SNR> 0, the SNR γ _n (k) at the time of noise injection is smaller than the original value. Will be modified to On the other hand, referring to Reference 1, the characteristic of the suppression coefficient with respect to the SNR gradually increases corresponding to the SNR, then rapidly increases at a certain SNR value, and then gradually increases and reaches saturation again, as shown in FIG. Therefore, noise injection causes γ
_As the value of _n (k) becomes smaller, the effect of reducing the suppression coefficient becomes larger relative to the SNR in the vicinity where the suppression coefficient value suddenly changes. Therefore, a frequency component corresponding to such an SNR, specifically, a component having a medium amplitude is relatively suppressed. For this reason, a part of the background noise, which has a smaller amplitude than the voice but cannot be ignored, is more strongly suppressed, and is less likely to be perceived as noise in the emphasized voice. Therefore, it is possible to obtain emphasized speech of sufficiently high quality with respect to actual background noise.

(Second Embodiment) FIG. 5 is a block diagram showing the overall configuration of a second embodiment of the noise removing apparatus of the present invention. This noise removing device includes an injection noise calculating unit 55 and an adder 5 included in the noise removing device shown in FIG.
Instead of 6, 57, an SNR correction unit 65 is provided. Hereinafter, these differences will be mainly described in detail.

The SNR correction unit 65 includes a multiple multiplication unit 17,
The deteriorated speech power spectrum, the estimated noise power spectrum, and the acquired SNR are supplied from the estimated noise calculation unit 51 and the frequency-based SNR calculation unit 6, respectively. S
The corrected a priori SNR is supplied from the NR corrector 65 to the estimated a priori SNR calculator 7 and the noise suppression coefficient generator 8. That is, in the noise eliminator shown in FIG. 1, the acquired SNR is calculated using the deteriorated speech power spectrum into which noise is injected and the estimated noise power spectrum into which noise is injected, whereas in FIG. In the noise removing device, the calculated acquired SNR is corrected using the injection noise calculated using the deteriorated speech power spectrum and the estimated noise power spectrum.

Regarding the SNR correction unit 65 in FIG.
Further description will be made. FIG. 6 is a block diagram showing a configuration example of the SNR correction unit 65. The SNR correction unit 65 includes K correction SNR calculation units 654 _{0 to} 654 _K−1 and a separation unit 65.
1, 652, 653 and a multiplexing unit 655. The acquired SNR is supplied from the frequency-based SNR calculation unit 6 in FIG. 5 to the separation unit 651. The separation unit 651 separates the acquired acquired SNR into K frequency components, and transmits the _K frequency components to the corrected SNR calculation units 654 _{0 to} 654 _K−1 .
The demultiplexing unit 652 is supplied with the degraded voice power spectrum from the multiplex multiplication unit 17 in FIG. Separation unit 652
Separates the supplied degraded speech power spectrum into K frequency components, and corrects the SNR calculation unit 654 _0.
~ 654 _K-1 . The estimated noise power spectrum is supplied from the estimated noise calculation unit 51 in FIG. 5 to the separation unit 653. The separation unit 653 separates the supplied estimated noise power spectrum into K frequency components, and transfers them to the corrected SNR calculation units 654 _{0 to} 654 _K-1 .
The corrected SNR calculation units 654 _{0 to} 654 _K-1 add the corrections corresponding to the supplied deteriorated speech power spectrum and estimated noise power spectrum to the acquired SNR and add the corrected acquired SN.
The R is transmitted to the multiplexing unit 655. The multiplexing unit 655 multiplexes and outputs the corrected post-correction SNR.

FIG. 7 is a block diagram showing the configuration of the corrected SNR calculation units 654 _{0 to} 654 _K-1 included in the SNR correction unit 65 shown in FIG. The corrected SNR calculation unit 654
It has a threshold value calculation unit 6541, an injection noise calculation unit 6542, adders 6543 and 6544, and a division unit 6545.

The estimated noise power spectrum is supplied from the separation unit 653 in FIG. 6 to the threshold value calculation unit 6541, and the threshold value is calculated by the same operation as the threshold value calculation unit 552 in FIG. Injection noise calculator 654
Communicate to 2. The injection noise calculation unit 6542 is also supplied with the acquired SNR from the separation unit 651 in FIG.
Pseudo noise (first noise, addition signal) to be injected by the same operation as the injection level calculation unit 553 in FIG.
Is calculated and transmitted to the adders 6543 and 6544. The estimated noise power spectrum is also supplied from the separation unit 653 in FIG. 6 to the adder 6543, and the addition result with the noise supplied from the injection noise calculation unit 6542 is divided by the division unit 6
To 545. The adder 6544 is also supplied with the degraded voice power spectrum from the separation unit 652 in FIG. 6, and transfers the addition result with the noise supplied from the injection noise calculation unit 6542 to the division unit 6545. Division unit 654
5 outputs the quotient obtained from the output of the adder 6543 and the output of the adder 6544 as the corrected SNR.

FIG. 8 is a block diagram showing another configuration example of the SNR correction unit 65. In this configuration example, the SNR and the threshold value are commonly set for all frequency components. Therefore, in comparison with the configuration example shown in FIG. 6, new average value calculation units 661 and 663 and injection noise calculation unit 662 are newly added.
Correcting SNR calculator 664 to have, also in the form of replacing a correction SNR calculator _{654 0 ~654 K-1 0 ~664} K-1
have.

The average value calculation unit 661 is provided by the separation unit 651.
Acquired SNRγ_nFind the average of (k) for k
Therefore, it is transmitted to the injection noise calculation unit 662. Therefore, injection miscellaneous
The value transmitted to the sound calculation unit 662 becomes one. on the other hand,
The average value calculation unit 663 uses the estimation value supplied from the separation unit 653.
Constant noise power spectrum λ_nFind the average of (k) for k
Therefore, it is transmitted to the threshold calculation unit 6541. Threshold calculation
Part 6541 sets the threshold value by the operation already described.
Obtained and transmitted to the injection noise calculation unit 662. Injection noise calculation
The unit 662 is the same as the injection noise calculation unit 6542 in FIG.
Pseudo noise to be injected in the same procedure (first noise, summing signal
No.) and the corrected SNR calculation unit 664 ₀ ~ 664_K-1 
Communicate to. Unlike the configuration example shown in FIG. 6, the corrected SNR
Calculation unit 664₀ ~ 664_K-1 The injection noise transmitted to
All have the same value.

FIG. 9 is a block diagram showing the configuration of the corrected SNR calculation units 664 _{0 to} 664 _K-1 included in the SNR correction unit 66 shown in FIG. The corrected SNR calculation unit 664 is
The injection noise supplied from the injection noise calculation unit 662 is added to the estimated noise power spectrum and the deteriorated speech power spectrum, the quotient of the two is calculated, and the corrected noise SNR is output as the corrected SNR. More specifically, it is as follows. That is, the injection noise calculated by the injection noise calculation unit 662 is
It is transmitted to adders 6543 and 6544. Adder 65
The estimated noise power spectrum is also supplied to the 43 from the separating unit 653 in FIG.
The addition result with the noise supplied from is transmitted to the division unit 6545. The adder 6544 includes a separating unit 65 in FIG.
The degraded voice power spectrum is also supplied from No. 2, and the addition result with the noise supplied from the injection noise calculation unit 6542 is transmitted to the division unit 6545. The division unit 6545 outputs the quotient obtained from the output of the adder 6543 and the output of the adder 6544 as the corrected SNR.

In the configuration example shown in FIGS. 8 and 9, the correction SN is
By sharing the injection noise calculation unit 662 and a threshold calculating unit 6541 for R calculation unit 664 ₀ ~664 _K-1, and injected noise calculation unit to all the correction SNR calculator 664 ₀ ~664 _K-1 Since it is not necessary to provide the threshold value calculation unit, the configuration can be simplified.

As described above, the SNR correction units 65 and 66
1 is used to correct the acquired SNR, and the suppression coefficient is determined using the corrected acquired acquired SNR.
Similar to the noise eliminator shown in, it is possible to obtain emphasized speech of sufficiently high quality against actual background noise.

(Third Embodiment) FIG. 10 is a block diagram showing the overall configuration of a third embodiment of the noise eliminator of the present invention. This noise removing device has a configuration in which the injection noise calculating unit 55 is replaced with an injection noise calculating unit 58 in the noise removing device shown in FIG. Hereinafter, this difference will be mainly described in detail. The noise eliminator shown in FIG. 10 selectively applies noise injection according to the nature of the input signal. Therefore, in order to evaluate the characteristics of the input signal, the time-domain degraded speech signal output from the frame division unit 1 is supplied to the injection noise calculation unit 58.

FIG. 11 is a block diagram showing the structure of the injection noise calculation unit 58 in FIG. The injection noise calculation unit 55 shown in FIG. 2 is different in that a zero crossing calculation unit 581 and a switch 582 are further provided. The time domain deteriorated speech signal output from the frame division unit 1 is supplied to the zero crossing calculation unit 581. Zero crossing calculator 58
1, the SNR is supplied from the SNR calculation unit 551 and the threshold value is supplied from the threshold value calculation unit 552.
The zero crossing calculator 581 counts zero crossings at which the amplitude of the supplied deteriorated voice signal becomes zero. At the same time, SNR
Then, it is evaluated whether or not the SNR is smaller than the second threshold value TH _{2 from} the threshold value. Only when the SNR is less than the second threshold TH ₂ , the number of zero crossings is averaged over the past several frames. That is, the average value is obtained only when it is determined that the deteriorated voice is silent. The average value thus obtained is compared with the third threshold value, and when the average value is larger, "1" is transmitted, and otherwise "0" is transmitted to the switch 582 as a control signal. To do.
The third threshold value can be set in advance or can be changed during the operation.

The switch 582 has an injection level calculation unit 5
Injection noise is supplied together with 0 from 53. The switch 582 supplies the injection noise supplied from the injection level calculation unit 553 when “1” is supplied as the control signal from the zero crossing calculation unit 581 and 0 when “0” is supplied.
Is selected and output as injection noise. Therefore, only if the average number of zero crossings is greater than the third threshold,
The injection noise from the injection level calculator 553 is supplied to the adders 56 and 57 in FIG. It is known that the number of zero crossings increases in a non-stationary signal. Therefore, it is possible to perform noise injection and correct the suppression coefficient only for a signal whose non-stationarity is a certain value or more.

(Fourth Embodiment) FIG. 12 is a block diagram showing the overall structure of a fourth embodiment of the noise eliminator of the present invention. This noise removing device has a configuration in which the injection noise calculating unit 58 in the noise removing device shown in FIG. 10 is replaced with an injection noise calculating unit 59. Hereinafter, this difference will be mainly described in detail.

The noise eliminator shown in FIG. 12 is the same as the noise eliminator shown in FIG. 10 in that noise injection is selectively applied according to the characteristics of the input signal. However, the degraded speech signal in the time domain, which is the output of the frame division unit 1, is not supplied to the injection noise calculation unit 59. The reason is that, unlike the noise removing apparatus shown in FIG. 10, the deteriorated speech signal in the time domain is not used in order to evaluate the property of the input signal. Instead, the degraded speech power spectrum is used. In the noise eliminator shown in FIG. 10, the non-stationarity of the signal is evaluated using the number of zero crossings per frame, but the number of zero crossings and the power spectrum in the high frequency region (high range) have no correlation. As is known, the degraded speech power spectrum can be used instead of the number of zero crossings.

FIG. 13 is a block diagram showing the structure of the injection noise calculator 59 shown in FIG. The difference from the injection noise calculation unit 58 shown in FIG. 11 is that the zero crossing calculation unit 581 is replaced with a high frequency power calculation unit 591. The high frequency power calculation unit 591, together with the SNR calculation unit 551,
Degraded voice power spectrum is provided. The high frequency power calculation unit 591 uses the deteriorated voice power spectrum | Y _n (k)
Of ² |, those for which k is greater than the reference value k _TH are summed. The reference value k _TH is set according to the deteriorated sound signal and other conditions so that the high frequency power corresponding to the number of zero crossings of the deteriorated sound signal described above can be obtained by taking the sum. As a result, a high frequency power corresponding to the number of zero crossings is obtained, and therefore the injection noise calculation unit 5 shown in FIG.
The switch 582 can be controlled in the same manner as 8 above. That is, the injection level calculation unit 55 is determined by the value of the high frequency power.
The injection noise supplied from 3 and 0 are selected and output as injection noise.

Note that the degraded voice power spectrum | Y
_{Of n} (k) | ^2, the one in which k is larger than the reference value k _TH may be weighted and summed to obtain the high frequency power. Further, the fourth threshold value can be set in advance or can be changed during the operation.

(Fifth Embodiment) FIG. 14 is a block diagram showing the overall configuration of a fifth embodiment of the noise removing apparatus of the present invention. This noise removing device has a configuration in which the SNR correcting unit 65 is replaced with an SNR correcting unit 67 in the noise removing device shown in FIG. Hereinafter, this difference will be mainly described in detail. The noise eliminator shown in FIG. 14 selectively applies noise injection according to the nature of the input signal, as in the noise eliminator shown in FIG. Therefore, in order to evaluate the property of the input signal, the degraded speech signal in the time domain, which is the output of the frame division unit 1, is
It is supplied to the correction unit 67.

FIG. 15 shows the SNR correction unit 6 in FIG.
7 is a block diagram showing a configuration example of 7. SN shown in FIG.
The difference from the configuration example of the R correction unit 65 is that the injection noise calculation unit 662 is replaced with the injection noise calculation unit 672. Unlike the injection noise calculation unit 662, the injection noise calculation unit 672 is supplied with the time domain deteriorated speech signal output from the frame division unit 1 in order to evaluate the property of the input signal.

FIG. 16 is a block diagram showing a configuration example of the injection noise calculation unit 672. The injection noise calculation unit 672 includes an injection level calculation unit 6721, a switch 6722, and a determination unit 6.
723. Injection level calculation unit 6721 and determination unit 6
In 723, the acquired SNR from the average value calculation unit 661 in FIG. 15 and the threshold value calculation unit 65 in FIG.
The threshold value is supplied from 41. Judgment unit 6723
Is further supplied with a degraded audio signal. The injection level calculation unit 6721 is the injection level calculation unit 5 in FIG.
By the same operation as 53, the injection level is obtained and transmitted to the switch 6722. The determination unit 6723 receives the deteriorated voice signal, the acquired SNR, and the threshold value, and generates a control signal for the switch 6722 according to the property of the input signal.

Here, the determination unit 6723 further determines that there is no sound.
Section detection unit 67231, zero crossing calculation unit 67232, ratio
It is composed of a comparison unit 67233. Silent section detection unit 672
31 receives the acquired SNR and the threshold, and
R is the second threshold TH ₂"1" when smaller
Otherwise, "0" otherwise, the zero crossing calculation unit 672
32. That is, the degraded voice is determined to be silent.
Then, "1" is calculated, otherwise "0" is calculated as zero crossing
It will be transmitted to the part 67232. Zero crossing calculator 6
7232, the amplitude of the supplied degraded audio signal is zero.
Counting zero crossings from the silent interval detection unit 67231
Only when "1" is received, the number of zero crossings is the past number.
Average over frames. Thus obtained
The average value is transmitted to the comparison unit 67233. Comparison unit 67
233, the average value of the zero crossings supplied is the third value.
"1" when the average value is larger than the threshold value of
, Otherwise switch to “0” as the control signal.
C 6722.

The switch 6722 selects the injection noise supplied from the injection level calculation unit 6721 when "1" is supplied from the comparison unit 67233 of the determination unit 6723, and selects 0 when "0" is supplied. , Output as injection noise. That is, the operation of the switch 6722 is the same as the operation of the switch 582 in FIG. 11, and the noise injection can be executed and the suppression coefficient can be corrected only for the signal whose non-stationarity is constant or more.

(Sixth Embodiment) FIG. 17 is a block diagram showing the overall structure of a sixth embodiment of the noise eliminator of the present invention. This noise removing apparatus is the same as the noise removing apparatus shown in FIG.
The configuration is replaced by the correction unit 68. Hereinafter, this difference will be mainly described in detail.

In the noise eliminator shown in FIG. 17, noise injection is selectively applied according to the property of the input signal. At that time, unlike the noise eliminator shown in FIG. 14, the quality of the input signal is evaluated by using the degraded voice power spectrum instead of the degraded voice signal in the time domain. That is, unlike the fifth embodiment in which the non-stationarity of a signal is evaluated by the number of zero crossings per frame, the non-stationarity of a signal is evaluated using a deteriorated voice power spectrum in a high frequency region (high range). . For this reason, the time domain degraded audio signal output from the frame division unit 1 is not supplied to the SNR correction unit 68. FIG. 18 shows the SNR correction unit 6 in FIG.
8 is a block diagram showing a configuration example of No. 8. S shown in FIG.
The difference from the NR correction unit 67 is that the injection noise calculation unit 672 is replaced with the injection noise calculation unit 682.

FIG. 19 is a block diagram showing a configuration example of the injection noise calculation unit 682. The difference from the injection noise calculation unit 672 shown in FIG. 16 is that the zero crossing calculation unit 67232 is replaced with a high frequency power calculation unit 68232. The high frequency power calculator 68232 includes a silent section calculator 6723.
A degraded audio power spectrum is provided along with the 1 output signal. The high frequency power calculation unit 68232 performs the same operation as the high frequency power calculation unit 591 in FIG. 13 to calculate the sum of the deteriorated voice power spectrum | Y _n (k) | ² whose k is larger than the reference value k _TH. And obtain the high frequency power. This high frequency power is transmitted to the comparison unit 67233. The comparing unit 67233 generates a control signal for the switch 6722 using the result of comparing the high frequency power with the fourth threshold value. That is, depending on the value of high frequency power,
The injection noise supplied from the injection level calculator 6721 and 0
Is selected and output as injection noise.

(Seventh Embodiment) FIG. 20 is a block diagram showing the overall configuration of a seventh embodiment of the noise eliminator of the present invention. The noise removing apparatus and the noise removing apparatus shown in FIG. 1 are the same except for the estimated noise calculation unit 5, the weighted deteriorated speech calculation unit 14, and the suppression coefficient correction unit 15. The configuration of the noise eliminator shown in FIG. 20 is, except for the windowing processor 22 and the injection noise calculator 58, “April 2000, IEICE Technical Research Report, DSP, 53-”.
Page 60 ”(reference 5). Unlike the conventional method disclosed in Literature 1, the method disclosed in Literature 5 uses a weighted degraded speech spectrum to
By estimating the power spectrum of noise, an accurate estimated noise can be obtained. Hereinafter, these differences will be mainly described in detail.

First, the weighted deteriorated speech calculation unit 14 in FIG. 20 will be described. FIG. 21 is a block diagram showing the configuration of the weighted deteriorated speech calculation unit 14. The weighted deteriorated speech calculation unit 14 includes an estimated noise storage unit 1401, a frequency-based SNR calculation unit 1402, and a multiple nonlinear processing unit 140.
5, and multiplex multiplication section 1404. The estimated noise storage unit 1401 stores the estimated noise power spectrum supplied from the estimated noise calculation unit 5 in FIG. 20, and calculates the estimated noise power spectrum stored one frame before by S for each frequency.
Output to the NR calculation unit 1402. The frequency-based SNR calculation unit 1402 obtains an SNR for each frequency using the estimated noise power spectrum supplied from the estimated noise storage unit 1401 and the deteriorated voice power spectrum supplied from the multiplex multiplication unit 17 in FIG. Multiple Nonlinear Processing Unit 14
Output to 05. The multiplex nonlinear processing unit 1405 calculates a weighting coefficient vector using the SNR supplied from the frequency-based SNR calculating unit 1402, and outputs the weighting coefficient vector to the multiplex multiplication unit 1404. The multiplying unit 1404 is shown in FIG.
20 calculates the product of the deteriorated voice power spectrum supplied from the multiplex multiplication unit 17 and the weighting coefficient vector supplied from the multiplex nonlinear processing unit 1405 for each frequency, and calculates the weighted deteriorated voice power spectrum in FIG. Output to the unit 5.

The configuration of the frequency-dependent SNR calculation section 1402 is as follows.
Since it is the same as the frequency-based SNR calculation unit 6 already described with reference to FIG. 56, detailed description thereof will be omitted. Further, since the configuration of the multiplex multiplication unit 1404 is the same as that of the multiplex multiplication unit 17 already described with reference to FIG. 52, detailed description thereof will be omitted. Therefore, next, the configuration and operation of the multiple nonlinear processing unit 1405 in FIG. 21 will be described in detail.

FIG. 22 shows the weighted deteriorated speech calculation unit 14.
A block diagram showing the configuration of the included multiple nonlinear processing unit 1405.
It is a diagram. The multiple nonlinear processing unit 1405 includes a separating unit 14
95, K non-linear processing units 1485₀ ~ 1485_K-1 ,
And a multiplexing unit 1475. Separation unit 1495
21 is supplied from the frequency-dependent SNR calculation unit 1402.
The non-linear processing unit
1485₀ ~ 1485_{K- 1} Output to. Non-linear processing unit 1
485₀ ~ 1485_K-1 Are the actual values corresponding to the input values.
It has a non-linear function that outputs a numerical value. In FIG. 23, the nonlinear
Here is an example of a function: f₁ When the input value is
Output value f of the nonlinear function₂ Is given by equation (17)
Be done.

[0124]

[Equation 17]

Non-linear processing section 1485 _{0 to} 1485 _K-1
Is the frequency-dependent SNR supplied from the separation unit 1495,
Calculate the weighting coefficient by processing with the above-mentioned nonlinear function,
Output to the multiplexing unit 1475. That is, the non-linear processing units 1485 _{0 to} 1485 _K−1 have 1 to 0 depending on the SNR.
The weighting factors up to are output. If the SNR is small, set 1
When it is larger, 0 is output. The multiplexing unit 1475 multiplexes the weighting factors output from the non-linear processing units 1485 _{0 to} 1485 _K-1, and outputs the resulting weighting factor vector to the multiplexing multiplication unit 1404 in FIG.

As described above, the multiple multiplication unit 1 in FIG.
The weighting coefficient to be multiplied by the deteriorated speech power spectrum in 404 has a value according to the SNR. The larger the SNR, that is, the larger the speech component included in the deteriorated speech,
The value of the weighting factor becomes smaller. Although the deteriorated speech power spectrum is generally used for updating the estimated noise, SN is used for the deteriorated speech power spectrum used for updating the estimated noise.
By weighting according to R, it is possible to reduce the influence of the voice component included in the deteriorated voice power spectrum, and it is possible to perform noise estimation with higher accuracy. Although an example in which a non-linear function is used for the calculation of the weighting coefficient is shown, it is also possible to use a SNR function represented in another form such as a linear function or a high-order polynomial in addition to the non-linear function.

Next, the estimated noise calculator 5 in FIG. 20 will be described. FIG. 24 is a block diagram showing the configuration of the estimated noise calculation unit 5. This estimated noise calculation unit 5 and FIG.
The estimated noise calculator 51 shown, and the separation unit 505 is present, the frequency domain estimated noise calculator 514 _0-514
It is identical except _{that K-1} has been replaced with a frequency domain estimated noise calculator 504 ₀ ~504 _K-1. Hereinafter, these differences will be mainly described in detail.

The separating section 505 separates the weighted deteriorated speech power spectrum supplied from the weighted deteriorated speech calculation section 14 in FIG. 20 into frequency-dependent weighted deteriorated speech power spectrums, and each frequency-dependent estimated noise calculation section. Output to 504 _{0 to} 504 _K-1 . The frequency _- dependent estimated noise calculation units 504 _{0 to} 504 _K−1 are the frequency _- dependent deteriorated speech power spectrum supplied from the separation unit 502, and the separation unit 505.
20 to calculate the frequency-dependent estimated noise power spectrum from the frequency-dependent weighted deteriorated voice power spectrum, the voice detection flag supplied from the voice detection unit 4 in FIG. 20, and the count value supplied from the counter 13 in FIG. , To the multiplexing unit 503. Multiplexer 50
3 multiplexes the frequency _- dependent estimated noise power spectrum supplied from the frequency _- dependent estimated noise calculation units 504 _{0 to} 504 _K-1, and the estimated noise power spectrum obtained as a result is added to the adder 56 and injection noise calculation in FIG. The data is output to the unit 58 and the weighted deteriorated speech calculation unit 14. A detailed description of the configuration and operation of the frequency _- dependent estimated noise calculation units 504 _{0 to} 504 _K-1 will be given with reference to FIGS. 25 to 27.

FIG. 25 shows the estimated noise calculation section shown in FIG.
Frequency-dependent estimated noise calculation unit 504 included in No. 5₀ ~ 504
_K-1 3 is a block diagram showing a first configuration example of FIG. In Figure 54
The difference from the frequency-dependent estimated noise calculation unit 514 shown is that
Estimated noise calculation unit 504 by wave number ₀ ~ 504_K-1 Is the estimated noise
The storage unit 5942 is included, and the update determination unit 521 updates
The determination unit 520 is replaced, and the switch 50
The input to 44 is from the degraded speech power spectrum by frequency
Replaced by weighted degraded speech power spectrum by frequency
It is that. Frequency-dependent estimated noise calculation unit 504₀ ~
504_K-1 Degrades the speech power spectrum in the estimated noise calculation.
Using weighted degraded speech power spectrum instead of toll
In addition, the estimated noise and deterioration are
These differences are due to the use of the voice power spectrum.
Dots occur. The estimated noise storage unit 5942 has a division unit 50.
Estimated noise power spectrum by frequency supplied from 48
And the estimated noise for each frequency stored one frame before
The power spectrum is output to the update determination unit 520. update
See FIG. 26 for a detailed description of the configuration and operation of the determination unit 520.
Do it while shining.

FIG. 26 is a diagram showing an update determining unit 5 included in the frequency _- dependent estimated noise calculating units 504 _{0 to} 504 _K-1 shown in FIG.
It is a block diagram which shows the structure of 20. 55 is different from the update determination unit 521 shown in FIG. 55 in that the logical sum calculation unit 5211 is replaced with the logical sum calculation unit 5201, and the update determination unit 520 includes a comparison unit 5205, a threshold value storage unit 5206, and a threshold value calculation unit. That is to have a part 5207. Hereinafter, detailed operations will be described focusing on these differences. Threshold calculator 5
207 calculates a value according to the frequency-dependent estimated noise power spectrum supplied from the estimated noise storage unit 5942 in FIG. 25, and outputs it to the threshold storage unit 5206 as a threshold value. The simplest threshold calculation method is a constant multiple of the estimated noise power spectrum for each frequency. Besides, it is also possible to calculate the threshold value using a high-order polynomial or a non-linear function.

The threshold storage unit 5206 has a threshold calculation unit 520.
The threshold value output from No. 7 is stored, and the threshold value stored one frame before is output to the comparison unit 5205. Comparison unit 5205
Compares the threshold supplied from the threshold storage unit 5206 with the frequency-dependent deteriorated voice power spectrum supplied from the separation unit 502 in FIG. 24. If the frequency-dependent deteriorated voice power spectrum is smaller than the threshold, increase “1”. For example, “0” is output to the logical sum calculation unit 5201. That is,
Based on the size of the estimated noise power spectrum, it is determined whether the deteriorated speech signal is noise. The logical sum calculation unit 5201 outputs the output value of the comparison unit 5203 and the logical negation circuit 5
The logical sum of the output value of 202 and the output value of the comparison unit 5205 is calculated, and the calculation result is the switch 504 in FIG.
4, output to the shift register 5045 and the counter 5049.

Therefore, in addition to the initial state and the silent section,
If the deteriorated voice power is small even in the voiced section, the update determination unit 520 outputs “1”. That is, the estimated noise is updated. Since the threshold value is calculated for each frequency, the estimated noise can be updated for each frequency.

In FIG. 25, the CNT counter 504
The count value of 9 and N are the register length of the shift register 5045. Then, B _n (k) (n = 0, 1, ..., N
-1) is the weighted deteriorated speech power spectrum for each frequency stored in the shift register 5045. At this time, the frequency-dependent estimated noise power spectrum λ _n (k) output from the division unit 5048 is given by Expression (18).

[0134]

[Equation 18]

That is, λ _n (k) is the shift register 50
The average value of the weighted deteriorated speech power spectrum for each frequency stored in 45. The average value may be calculated using a weighted addition unit (cyclic filter). Next, with reference to FIG. 27, a configuration example in which a weighted addition unit is used for calculating λ _n (k) will be described.

FIG. 27 shows the frequency-dependent estimated noise calculators 504 _{0 to} 504 included in the estimated noise calculator 5 shown in FIG.
It is a block diagram showing the 2nd example of composition of _K-1 . The shift register 5045, the adder 5046, the minimum value selection unit 504 in the frequency-dependent estimated noise calculation unit 504 shown in FIG.
7, the division unit 5048, the counter 5049, the register length storage unit 5941, and the minimum value selection unit 5047 are replaced by the frequency-dependent estimated noise calculation unit 507, and the weighted addition unit 507.
1 has a weight storage unit 5072.

The weighted addition unit 5071 outputs the estimated noise power spectrum for each frequency before one frame supplied from the estimated noise storage unit 5942, the weighted deteriorated speech power spectrum for each frequency supplied from the switch 5044, and the weight storage unit 5072. The estimated noise for each frequency is calculated using the output weight, and is output to the multiplexing unit 503 in FIG. That is, δ is the weight stored in the weight storage unit 5072, and |
Y _n (k) | ² bar, weighted addition unit 5071
Estimated noise power spectrum for each frequency output from λ
_n (k) is given by Expression (19).

[0138]

[Formula 19]

The structure of the weighted addition unit 5071 is the same as that of the weighted addition unit 407 already described with reference to FIG. 51, and therefore detailed description thereof will be omitted. However, the weighted addition calculation is always performed.

Next, the suppression coefficient correction unit 15 in FIG.
Will be described. FIG. 28 is a block diagram showing the configuration of the suppression coefficient correction unit 15 in FIG. In order to prevent the residual noise generated due to insufficient suppression when the SNR is low and the sound quality deterioration due to the distortion of the voice that occurs due to the excessive suppression when the SNR is high, the suppression coefficient correction unit 15 corrects the suppression coefficient according to the SNR. Do. As an example of correction, SNR
When is low, a correction value is added to the suppression coefficient to suppress residual noise, and when the SNR is high, a lower limit value can be set to the suppression coefficient to prevent voice distortion. The suppression coefficient correction unit 15 includes K frequency-based suppression coefficient correction units 1501 ₀ to 1501.
501 _K-1 , demultiplexing units 1502 and 1503, and multiplexing unit 1
504.

The separation unit 1502 separates the estimated a priori SNR supplied from the estimated a priori SNR calculation unit 7 in FIG. 20 into frequency components, and outputs them to the frequency suppression coefficient correction units 1501 _{0 to} 1501 _K-1 . To do. Separation part 150
3 separates the suppression coefficient supplied from the suppression coefficient generation unit 8 in FIG. 20 into frequency components, and outputs them to the frequency suppression coefficient correction units 1501 _{0 to} 1501 _K-1 . The frequency _- dependent suppression coefficient correction units 1501 _{0 to} 1501 _K-1 are frequency _- dependent estimated a priori SNRs supplied from the separation unit 1502.
And the correction suppression coefficient for each frequency is calculated from the suppression coefficient for each frequency supplied from the separating section 1503, and the multiplexing section 1504
Output to. The multiplexing unit 1504 multiplexes the frequency-specific correction suppression coefficients supplied from the frequency-specific suppression coefficient correction units 1501 _{0 to} 1501 _K-1 and, as the correction suppression coefficient, the multiplex multiplication unit 16 and the estimated a priori SNR calculation unit. Output to 7.

FIG. 29 shows frequency-dependent suppression coefficient correction units 1501 ₀ to 1501 included in the suppression coefficient correction unit 15 shown in FIG.
It is a block diagram which shows the structure of 501 _K-1 . The frequency-dependent suppression coefficient correction unit 1501 includes a maximum value selection unit 1591, a suppression coefficient lower limit value storage unit 1592, a threshold value storage unit 1593, a comparison unit 1594, a switch 1595, and a correction value storage unit 1596.
And a multiplier 1597. The comparison unit 1594 compares the threshold value supplied from the threshold value storage unit 1593 with the frequency-based estimated a priori SN supplied from the separation unit 1502 in FIG.
R is compared, and if the inferred innate SNR for each frequency is larger than the threshold value, “0” is set.
Supply to 5.

The switch 1595 outputs the suppression coefficient for each frequency supplied from the separation unit 1503 in FIG. 28 to the multiplier 1597 when the output value of the comparison unit 1594 is "1", and the output value of the comparison unit 1594 is " When it is 0 ”, it is directly supplied to the maximum value selection unit 1591. The multiplier 1579 calculates the product of the output value of the switch 1595 and the output value of the correction value storage unit 1596, and supplies the calculation result to the maximum value selection unit 1591. The correction value is usually smaller than 1 in order to reduce the suppression coefficient value, but it is not limited to this depending on the purpose. In this way, the suppression coefficient is corrected when the estimated a priori SNR for each frequency is smaller than the threshold value. By correcting the suppression coefficient when the SNR is small, it is possible to reduce the residual noise amount without excessively suppressing the voice component.

The suppression coefficient lower limit storage unit 1592 supplies the stored lower limit value of the suppression coefficient to the maximum value selection unit 1591. The maximum value selection unit 1591 compares the signal supplied from the switch 1595 or the multiplier 1597 with the suppression coefficient lower limit value supplied from the suppression coefficient lower limit value storage unit 1592, and sets the larger value as the frequency-dependent correction suppression coefficient. , Fig. 2
8 to the multiplexing unit 1504. This allows
The suppression coefficient always has a value larger than the lower limit value stored in the suppression coefficient lower limit storage unit 1592. Therefore, it is possible to prevent the distortion of the voice generated by the excessive suppression. In the noise removal apparatus shown in FIGS. 1, 5, 10, 12, 14, and 17, the suppression coefficient is supplied to the multiplex multiplication unit 16 and the estimated a priori SNR calculation unit 7. In the noise eliminator shown in FIG. 20, the corrected suppression coefficient is supplied instead of the suppression coefficient.

Next, the noise suppression coefficient generator 8 in FIG. 20 will be described. As described using FIG. 60,
The suppression coefficient is the estimated innate SNR and acquired SN provided.
The value can be obtained by searching from R, but can also be obtained by calculation. Hereinafter, based on the calculation formula described in Document 1, together with the calculation method of the suppression coefficient, the noise suppression coefficient generation unit 8
Another configuration example will be described. FIG. 30 is a block diagram showing another configuration example of the noise suppression coefficient generation unit 8 in FIG. The noise suppression coefficient generation unit 81 uses the MMSE ST
SA gain function value calculation unit 811, generalized likelihood ratio calculation unit 8
12, a voice existence probability storage unit 813, and a suppression coefficient calculation unit 814.

Let the frame number be n and the frequency number be k,
γ _n (k) is the frequency-dependent a posteriori SNR supplied from the frequency-based SNR calculating unit 6 in FIG. 20, and ξ _n (k) hat is the frequency-based estimated a priori supplied from the estimated a priori SNR calculating unit 7 in FIG. Target SNR. Also, η _n (k) = ξ
_n (k) hat / q, v _n (k) = (η _n (k) γ _n (k)) /
_Let (1 + η _n (k)). The MMSE STSA gain function value calculation unit 811 receives the acquired SNR γ _n (k) supplied from the frequency-based SNR calculation unit 6 in FIG. 20, and the estimated a priori SNR ξ _n supplied from the estimated a priori SNR calculation unit 7 in FIG. (k) Hat and voice existence probability storage unit 81
The MMMESTSA gain function value is calculated for each frequency based on the voice existence probability q supplied from No. 3, and is output to the suppression coefficient calculation unit 814. MMSE STS for each frequency
The A gain function value G _n (k) is given by the equation (20).

[0147]

[Equation 20]

Here, I ₀ (z) is the 0th-order modified Bessel function, and I ₁ (z) is the 1st-order modified Bessel function. The modified Bessel function is described in "1985, Mathematics Dictionary, Iwanami Shoten, 374.G page" (Reference 6).
The generalized likelihood ratio calculation unit 812 receives the acquired SNR γ _n (k) supplied from the frequency-based SNR calculation unit 6 in FIG.
Based on the estimated a priori SNR ξ _n (k) hat supplied from the estimated a priori SNR calculation unit 7 and the voice existence probability q supplied from the voice existence probability storage unit 813 in FIG. 20, the generalized likelihood for each frequency. The ratio is calculated, and the suppression coefficient calculation unit 814
Output to. The generalized likelihood ratio Λ _n (k) for each frequency is given by equation (21).

[0149]

[Equation 21]

The suppression coefficient calculation unit 814 uses the MMSE ST
MMSE supplied from the SA gain function value calculation unit 811
A suppression coefficient is calculated for each frequency from the STSA gain function value G _n (k) and the generalized likelihood ratio Λ _n (k) supplied from the generalized likelihood ratio calculation unit 812, and the suppression coefficient correction unit 15 in FIG. Output to. The suppression coefficient G _n (k) bar for each frequency is
It is given by equation (22).

[0151]

[Equation 22]

Instead of calculating the SNR for each frequency, it is also possible to obtain the SNR common to the band composed of a plurality of frequencies and use it. Therefore, next, in FIG.
As another configuration example of the frequency-based SNR calculation unit 6 for 0, an example of calculating the SNR for each band will be described. FIG. 31 is a block diagram showing another configuration example of the frequency-based SNR calculation unit 6. The difference from the frequency-based SNR calculation unit 6 shown in FIG. 56 is that the band-based SNR calculation unit 61 includes band-based power calculation units 611 and 612. The band-specific power calculation unit 611 calculates the band-specific power based on the frequency-specific deteriorated voice power spectrum supplied from the separation unit 602, and outputs the calculated power to the division units 601 _{0 to} 601 _K-1 . Further, the band-specific power calculation unit 612 calculates the band-specific power based on the frequency-specific estimated noise power spectrum supplied from the separation unit 603, and outputs the calculated power to the division units 601 _{0 to} 601 _K-1 .

FIG. 32 is included in the band-based SNR calculation unit 61.
Block diagram showing the configuration of a band-specific power calculation unit 611
Is. Here, it is equally divided into M bands having a bandwidth L.
An example will be described. Where L and M are K = LM
Let it be a natural number that satisfies. Band-wise SNR calculator 61
Is the M adders 6110₀~ 6110_M-1Have. Figure
Degradation by frequency supplied from the separation unit 602 in No. 31
Audio power spectrum 910₀ ~ 910_K-1 (910₀ 
~ 910_ML-1) Is an adder 6110 corresponding to each frequency
₀ ~ 6110_M-1 Respectively transmitted to. For example, the band
Frequency numbers corresponding to number 0 are 0 to L-1, so
Deteriorated voice power spectrum 910 by wave number ₀~ 910_L-1 
Is the adder 6110₀Transmitted to. Also, for band number 1
Corresponding frequency numbers are from L to 2L-1.
Degraded speech power spectrum 910 _L~ 910_2L-1Is addition
Bowl 6110₁Transmitted to.

Adder 6110₀ ~ 6110_M-1 Supply
The sum of the degraded speech power spectrum by frequency
Each is calculated, and the degraded voice power spectrum 911 for each band₀ 
~ 911_ML-1(911₀ ~ 911_K-1 ) In Figure 31
Division unit 601₀ ~ 601_{K- 1} Output to. Of each adder
The calculation result is calculated for each frequency according to each band number.
It is output as the degraded voice power spectrum for each region. example
For example, adder 6110₀The calculation result of
War spectrum 911₀ ~ 911_L-1 Is output as
It Also, the adder 6110₁ The calculation result of is degraded by band
Audio power spectrum 911_L ~ 911_2L-1Output as
To be done. The configuration and operation of the power calculation unit 612 for each band is
Since it is equivalent to the separate power calculation unit 611, its description is omitted.
I will omit it.

Although an example of equally dividing into a plurality of bands is shown here, the critical band described in “Hearing and Speech, The Institute of Electronics, Information and Communication Engineers, pp. 115 to 118, 1980” (Reference 7) is used. The method of division, "1983, Multirate Digital Signal Processing (Multirat
e Digital Signal Processing), 1983, Prentice-
It is also possible to use another band division method such as the method of dividing into octave bands described in "Hall Inc., USA" (Reference 8).

(Eighth Embodiment) FIG. 33 is a block diagram showing the overall structure of an eighth embodiment of the noise removing apparatus of the present invention. The difference from the noise removal apparatus shown in FIG. 20 is that the injection noise calculation unit 58 and the adders 56 and 57 are
That is, it is replaced with the NR correction unit 67. The relationship between FIGS. 20 and 33 is the relationship between FIGS. 1 and 5 and the relationship between FIGS.
Since the SNR correction unit 67 has been described with reference to FIGS. 15 and 14, the detailed description of the noise eliminator shown in FIG. 33 will be omitted.

(Ninth Embodiment) FIG. 34 is a block diagram showing the entire structure of a ninth embodiment of the noise removing apparatus of the present invention. The difference from the noise removal apparatus shown in FIG. 20 is that the estimated noise calculation unit 5 is replaced by the estimated noise calculation unit 52 and that the weighted deteriorated speech calculation unit 14 does not exist. Hereinafter, these differences will be mainly described in detail.

FIG. 35 shows the estimated noise calculating section in FIG.
It is a block diagram which shows the structure of 52. The estimation shown in FIG.
The difference from the constant noise calculation unit 5 is that the frequency-dependent estimated noise calculation unit
504₀ ~ 504_K-1 Is a frequency-dependent estimated noise calculation unit 506
₀ ~ 506_K-1 And the estimated noise calculation
The part 52 weights the input signal with the deteriorated speech power spectrum
Is not to have. This is the estimated noise calculation for each frequency
Part 504₀ ~ 504_{K- 1} Is weighted by frequency to the input signal
It requires a degraded speech power spectrum, whereas
Constant noise calculation unit 506₀ ~ 506_K-1 Is the frequency of the input signal
No weighted degraded speech power spectrum is required
This is because. Hereinafter, with reference to FIG.
Frequency-dependent estimated noise calculation unit 506₀ ~ 506_K-1 Configuration of
And the operation will be described in detail.

FIG. 36 shows the frequency-dependent estimated noise calculators 506 ₀ to 50 ₀ included in the estimated noise calculator 52 shown in FIG.
It is a block diagram which shows the structure of 6 _K-1 . The difference from the frequency-specific estimated noise calculation unit 504 shown in FIG. 25 is that the frequency-specific estimated noise calculation unit 506 does not have a frequency-specific weighted deteriorated speech power spectrum in the input signal, and a division unit 5041, Non-linear processing unit 5042 and multiplier 504
3 is to have. Hereinafter, these differences will be mainly described in detail.

The division unit 5041 divides the frequency-dependent deteriorated speech power spectrum supplied from the separation unit 502 in FIG. 35 by the estimated noise power spectrum one frame before supplied from the estimated noise storage unit 5942, and obtains the division result. It is output to the non-linear processing unit 5042. The non-linear processing unit 5042 having the same configuration and function as the non-linear processing unit 1485 shown in FIG. 22 calculates a weighting coefficient according to the output value of the division unit 5041, and outputs it to the multiplier 5043. Multiplier 50
Reference numeral 43 denotes a frequency-dependent deteriorated speech power spectrum supplied from the separation unit 502 in FIG.
Calculate the product of the weighting factors supplied from 2 and switch 50
Output to 44.

The output signal of the multiplier 5043 is equivalent to the weighted deteriorated speech power spectrum by frequency in the estimated noise by frequency calculation unit 504 shown in FIG. That is, the weighted deteriorated speech power spectrum for each frequency can be calculated in the estimated noise for each frequency calculation unit 506. Therefore, in the noise eliminator shown in FIG. 34, the weighted deteriorated speech calculation unit 14 can be omitted.

(Tenth Embodiment) FIG. 37 is a block diagram showing the overall structure of a tenth embodiment of the noise removing apparatus of the present invention. The difference from the noise removing apparatus shown in FIG. 34 is that the injection noise calculating unit 58 and the adders 56 and 57 are provided.
Is replaced by the SNR correction unit 67. The relationship between FIGS. 34 and 37 is equal to the relationship between FIGS. 1 and 5, the relationship between FIGS. 10 and 14, and the relationship between FIGS.
Since the R correction unit 67 has been described with reference to FIGS. 15 and 14, detailed description of the noise removal device shown in FIG. 37 will be omitted.

(Eleventh Embodiment) FIG. 38 is a block diagram showing the overall structure of an eleventh embodiment of the noise removing apparatus of the present invention. The noise eliminator shown in FIG. 20 is the same as the noise eliminator except for the estimated a priori SNR calculator 71. Therefore, the difference will be mainly described below. FIG. 39
FIG. 39 is a block diagram showing a configuration of an estimated a priori SNR calculation unit 71 in FIG. 38. Estimated congenital S shown in FIG.
The NR calculation unit 7 has an acquired SNR storage unit 702, a suppression coefficient storage unit 703, and multiplex multiplication units 705 and 704, whereas the estimated a priori SNR calculation unit 71 replaces them.
Estimated noise storage unit 712, emphasized speech power spectrum storage unit 713, frequency-dependent SNR calculation unit 715, multiplex multiplication unit 7
Have 16. Further, the estimated a priori SNR calculator 7 has a suppression coefficient in the input signal, but the estimated a priori SNR calculator 71 has an emphasized speech amplitude spectrum and an estimated noise power spectrum in the input signal instead of the suppression coefficient. Less than,
The difference between the estimated a priori SNR calculators 7 and 71 will be mainly described below.

The multiplex multiplication section 716 receives the emphasized speech amplitude spectrum | X supplied from the multiplex multiplication section 16 in FIG.
_n (k) | bar = G _n (k) bar · | Y _n (k) | is squared for each frequency to obtain the emphasized voice power spectrum, and outputs the emphasized voice power spectrum storage unit 713. The configuration of the multiplex multiplying unit 716 is the same as that of the multiplex multiplying unit 17 already described with reference to FIG. 52, and thus detailed description thereof will be omitted. The emphasized sound power spectrum storage unit 713 stores the emphasized sound power spectrum supplied from the multiplex multiplication unit 716, and outputs the emphasized sound power spectrum supplied one frame before to the frequency-based SNR calculation unit 715. Estimated noise storage unit 7
The storage unit 12 stores the estimated noise power spectrum λ _n (k) supplied from the estimated noise calculation unit 5 in FIG. 38, and outputs the estimated speech power spectrum supplied one frame before to the frequency-based SNR calculation unit 715.

The frequency-based SNR calculation unit 715
Enhanced speech supplied from the power spectrum storage unit 713
Power spectrum G_n-1 ²(K) bar ・ | Y_n-1(k) |² 
And the estimated noise power supplied from the estimated noise storage unit 712.
ー spectrum λ_n-1Calculate SNR of (k) for each frequency
And outputs it to the multiple weighted addition unit 707. Frequency S
The configuration of the NR calculation unit 715 has already been described using FIG.
Since it is the same as the frequency-based SNR calculator 6, a detailed description will be given.
Omit it. G which is the output of the frequency-dependent SNR calculation unit 715
_n-1 ²(K) bar ・ | Y_n-1(k) | ² / Λ_n-1(k) is the formula
From the relationship of (11), the multiple multiplication unit 705 in FIG.
Is the output of γ_n-1(k) G_n-1 ²(K) is equivalent to bar
It Therefore, it is included in the noise eliminator shown in FIG.
The estimated a priori SNR calculator 7 is replaced by the estimated a priori SNR calculator 7
It becomes possible to replace with 1.

(Twelfth Embodiment) FIG. 40 is a block diagram showing the overall structure of a twelfth embodiment of the noise removing apparatus of the present invention. The difference from the noise removal apparatus shown in FIG. 38 is that the injection noise calculation unit 58 and the adders 56 and 57 are provided.
Is replaced by the SNR correction unit 67. The relationship between FIGS. 38 and 40 is the relationship between FIGS. 1 and 5, the relationship between FIGS. 10 and 14, the relationship between FIGS. 20 and 33, and the relationship between FIGS.
7 and the SNR correction unit 67 is shown in FIG.
Since the description has been made with reference to FIGS. 14 and 14, the detailed description of the noise eliminator shown in FIG.

(Thirteenth Embodiment) FIG. 41 is a block diagram showing the overall structure of a thirteenth embodiment of the noise removing apparatus of the present invention. The difference from the noise removal apparatus shown in FIG. 20 is that the estimated noise calculation unit 5 is an estimated noise unit 52 and the estimated a priori SNR calculation unit 7 is an estimated a priori SNR calculation unit 71.
Are replaced, and the weighted deteriorated speech calculation unit 14 does not exist. The configuration and operation of the estimated noise unit 52 are the same as those described with reference to FIGS. 35 and 36. Further, the configuration and operation of the estimated a priori SNR calculation unit 71 are the same as those described with reference to FIG. Therefore, the noise removing apparatus shown in FIG.
A function equivalent to that of the noise eliminator shown in is realized.

(Fourteenth Embodiment) FIG. 42 is a block diagram showing the overall structure of a fourteenth embodiment of the noise eliminator of the present invention. The difference from the noise removal apparatus shown in FIG. 41 is that the injection noise calculation unit 58 and the adders 56 and 57 are
Is replaced by the SNR correction unit 67. The relationship between FIGS. 41 and 42 is the relationship between FIGS. 1 and 5, the relationship between FIGS. 10 and 14, the relationship between FIGS. 20 and 33, the relationship between FIGS. 34 and 37, and the relationship between FIGS. 38 and 40. Similarly, since the SNR correction unit 67 has been described with reference to FIGS. 15 and 14, a detailed description of the noise eliminator shown in FIG. 42 will be omitted.

(Fifteenth Embodiment) FIG. 43 shows the present invention.
15 shows an overall configuration of a fifteenth embodiment of the noise eliminator
It is a block diagram. With the noise eliminator shown in FIG.
The difference is that the estimated noise calculation unit 5
It has been replaced and that the voice detector 4 does not exist
Is. That is, a voice detector is required for noise estimation.
It is not configured. Below, focusing on these differences
The details will be described. FIG. 44 shows the estimated noise meter in FIG.
It is a block diagram showing a configuration of a calculation unit 53. Shown in Figure 24
The difference from the estimated noise calculator 5 is that
Arithmetic part 504₀ ~ 504_K-1 Is the estimated noise calculation section 5 for each frequency
08₀ ~ 508_K-1 Replaced with the estimated noise
The calculator 53 does not have a voice detection flag in the input signal
That is. Estimation noise by frequency with reference to FIG.
Calculation unit 508₀ ~ 508_{K- 1} Explains the configuration and operation of
To do.

FIG. 45 shows the frequency-dependent estimated noise calculators 508 ₀ to 50 included in the estimated noise calculator 53 shown in FIG.
It is a block diagram which shows the structure of _8K-1 . The difference from the frequency-dependent estimated noise calculation unit 504 shown in FIG. 25 is that the update determination unit 520 is replaced with the update determination unit 522.
That is, 508 _{0 to} 508 _K-1 do not have a voice detection flag in the input signal. FIG. 46 shows an update determination unit 522 included in the frequency-dependent estimated noise calculation unit 508 shown in FIG.
3 is a block diagram showing the configuration of FIG. The difference from the update determination unit 520 shown in FIG. 26 is that the logical sum calculation unit 5201 is replaced by the logical sum calculation unit 5221.
22 does not have the logical negation circuit 5202, and does not have a voice detection flag in the input signal. That is, the update determination unit 522 does not use the voice detection flag for updating the estimated noise. This point is different from the update determination unit 520 shown in FIG.

The logical sum calculation unit 5221 includes a comparison unit 5205.
And the output value of the comparison unit 5203 is calculated, and
The calculation result is output to the switch 5044, the shift register 5045, and the counter 5049 in FIG. That is, the update determination unit 522 always outputs "1" until the count value reaches the preset value, and after reaching the count value, outputs "1" when the deteriorated voice power is smaller than the threshold value. . As described using FIG. 26, the comparison unit 520
5 determines whether or not the deteriorated voice signal is noise. That is, it can be said that the comparison unit 5205 performs voice detection for each frequency. Therefore, it is possible to realize the update determination unit and the estimated noise calculation unit that do not have the voice detection flag in the input signal.

(Sixteenth Embodiment) FIG. 47 is a block diagram showing the overall structure of a sixteenth embodiment of the noise removing apparatus of the present invention. The difference from the noise removing apparatus shown in FIG. 43 is that the injection noise calculating section 58 and the adders 56 and 57 are
Is replaced by the SNR correction unit 67. The relationship between FIGS. 43 and 47 is the relationship between FIGS. 1 and 5, the relationship between FIGS. 10 and 14, the relationship between FIGS. 20 and 33, the relationship between FIGS. 34 and 37, the relationship between FIGS. The relationship is the same as that of FIG. 41 and FIG.
Since it has been described with reference to FIG. 4, the detailed description of the noise elimination device shown in FIG. 47 will be omitted.

20, FIG. 33, FIG. 34, FIG. 37, FIG.
8, FIG. 40 to FIG. 43, and FIG. 47, FIG. 10 and FIG.
2 and selective noise injection using the degraded voice power spectrum instead of the degraded voice signal as in the relationship between FIG. 14 and FIG. 17 is possible, but the configuration is clear,
Details are omitted.

In all the embodiments described so far, the minimum mean square error short-time spectrum amplitude method has been assumed as the noise removal method, but it can be applied to other methods. As an example of such a method,
"December 1979, Proceedings of the
I E E E Vol. 67, No. 12 (PROCEE
DINGS OF THE IEEE, VOL.67, NO.12, PP.1586-1604, DE
C, 1979), pp. 1586-1604 "(Reference 9) and the Wiener filter method and" 1979 4
Mon, I E E Transactions On Auctions Speech And Signal
Processing, Volume 27, Issue 2 (IEEE TRANSACTIONS
ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, VOL.2
7, NO.2, PP.113-120, APR, 1979), pages 113 to 120 ”(Reference 10), and the like, but a detailed configuration example of these will be described. Omit it.

Regarding the schematic operation of the spectral subtraction method disclosed in Document 10, for example, FIGS. 43 and 47 can be referred to. 43 and 47, if the multiplex multiplication unit 16 is replaced with a multiple subtraction unit, the noise suppression coefficient generation unit 8 is replaced with a noise suppression amount calculation unit, and the suppression coefficient correction unit 15 is replaced with a suppression amount correction unit, the operation by the spectrum subtraction method is performed. Can be realized. In the multiple subtraction unit, the corrected noise suppression amount is subtracted from the deteriorated voice amplitude spectrum, and the obtained result is subjected to the inverse Fourier transform to obtain the emphasized voice. Here, after calculating the SNR,
Although the example of calculating the noise suppression amount based on the SNR has been described, the estimated noise obtained by the estimated noise calculation unit 53 can be directly subtracted from the deteriorated speech amplitude spectrum.

[0176]

As described above, in the present invention, pseudo noise is generated based on an input signal, and the suppression coefficient obtained by injecting this pseudo noise is used. By injecting the above-mentioned pseudo noise when determining the suppression coefficient, the suppression coefficient derived assuming the background noise according to a specific statistical model is corrected according to the input signal, and noise that does not follow the statistical model is corrected. It can be effectively removed. Therefore, it is possible to obtain the emphasized speech of sufficiently high quality for all background noises.

Further, according to the present invention, windowing processing is applied to the time domain signal obtained by converting the emphasized speech in the frequency domain. When two adjacent frames of the time domain signal obtained by converting the emphasized speech in the frequency domain are superposed and added, even if the signal sample to be superposed and added is suppressed by different suppression coefficient values in each frame, The continuity of the signal samples at the frame boundaries can be improved by windowing each frame to reduce the amplitude of the signal samples at the frame boundaries. Thereby, it is possible to prevent the generation of noise and reduce the deterioration of the sound quality due to the noise.

[Brief description of drawings]

FIG. 1 is a block diagram showing an overall configuration of a first embodiment of a noise removing device of the present invention.

FIG. 2 is a block diagram showing a first configuration of an injection noise calculation unit included in the noise removal device shown in FIG.

FIG. 3 is a diagram showing an example of a relationship between SNR and injection noise.

FIG. 4 is a diagram showing an example of characteristics of suppression coefficients with respect to SNR.

FIG. 5 is a block diagram showing an overall configuration of a second embodiment of a noise removing device of the present invention.

6 is an SN included in the noise removing device shown in FIG.
It is a block diagram which shows the 1st structure of an R correction part.

7 is a correction S included in the SNR correction unit shown in FIG.
It is a block diagram which shows the structure of an NR calculation part.

FIG. 8 is a block diagram showing a second configuration of an SNR correction unit.

9 is a correction S included in the SNR correction unit shown in FIG.
It is a block diagram which shows the structure of an NR calculation part.

FIG. 10 is a block diagram showing an overall configuration of a third embodiment of a noise removing device of the present invention.

FIG. 11 is a block diagram showing a second configuration of the injection noise calculation unit.

FIG. 12 is a block diagram showing an overall configuration of a fourth embodiment of a noise removing device of the invention.

FIG. 13 is a block diagram showing a third configuration of the injection noise calculation unit.

FIG. 14 is a block diagram showing an overall configuration of a fifth embodiment of a noise removing device of the invention.

FIG. 15 is a block diagram showing a third configuration of the SNR correction unit.

FIG. 16 is a block diagram showing a fourth configuration of the injection noise calculation unit.

FIG. 17 is a block diagram showing an overall configuration of a sixth embodiment of a noise removing device of the present invention.

FIG. 18 is a block diagram showing a fourth configuration of the SNR correction unit.

FIG. 19 is a block diagram showing a fifth configuration of the injection noise calculation unit.

FIG. 20 is a block diagram showing the overall configuration of a seventh embodiment of a noise removal device of the present invention.

21 is a block diagram showing a configuration of a weighted deteriorated speech calculation unit included in the noise removal apparatus shown in FIG.

22 is a block diagram showing a configuration of a multiplex nonlinear processor included in the weighted deteriorated speech calculator shown in FIG. 21.

FIG. 23 is a diagram showing an example of a non-linear function in a non-linear processing unit.

FIG. 24 is a block diagram showing a first configuration of an estimated noise calculator included in the noise eliminator shown in FIG.

25 is a block diagram showing a first configuration of a frequency-specific estimated noise calculation unit included in the estimated noise calculation unit shown in FIG. 24.

26 is a block diagram showing a configuration of an update determination unit included in the frequency-dependent estimated noise calculation unit shown in FIG. 25.

FIG. 27 is a block diagram showing a second configuration of the frequency-dependent estimated noise calculation unit.

28 is a block diagram showing a configuration of a suppression coefficient correction unit included in the noise removal device shown in FIG.

29 is a block diagram showing the configuration of a frequency-specific suppression coefficient correction unit included in the suppression coefficient correction unit shown in FIG. 28.

FIG. 30 is a block diagram showing a second configuration of the noise suppression coefficient generation unit.

FIG. 31 is a block diagram showing a second configuration of a frequency-specific SNR calculation unit.

32 is a block diagram illustrating a configuration of a band-specific power calculation unit included in the frequency-based SNR calculation unit illustrated in FIG. 31.

FIG. 33 is a block diagram showing the overall configuration of an eighth embodiment of a noise removal device of the present invention.

FIG. 34 is a block diagram showing an overall configuration of a ninth embodiment of a noise removal device of the present invention.

FIG. 35 is a block diagram showing a second configuration of the estimated noise calculation unit.

36 is a block diagram showing a configuration of a frequency-specific estimated noise calculation unit included in the estimated noise calculation unit shown in FIG. 35.

FIG. 37 is a block diagram showing the overall structure of a tenth embodiment of a noise eliminator of the present invention.

FIG. 38 is a block diagram showing the overall configuration of an eleventh embodiment of a noise removal device of the present invention.

39 is a block diagram showing a configuration of an estimated a priori SNR calculation unit included in the noise eliminator shown in FIG.

FIG. 40 is a block diagram showing an overall configuration of a twelfth embodiment of a noise removal device of the present invention.

FIG. 41 is a block diagram showing the overall configuration of a thirteenth embodiment of a noise removal device of the present invention.

FIG. 42 is a block diagram showing the overall configuration of a fourteenth embodiment of a noise removal device of the present invention.

FIG. 43 is a block diagram showing the overall configuration of a fifteenth embodiment of a noise removal device of the present invention.

FIG. 44 is a block diagram showing a third configuration of the estimated noise calculation unit.

45 is a block diagram showing a configuration of a frequency-specific estimated noise calculation unit included in the estimated noise calculation unit shown in FIG. 44.

FIG. 46 is a block diagram showing the configuration of an update determination unit included in the frequency-dependent estimated noise calculation unit shown in FIG. 45.

FIG. 47 is a block diagram showing the overall configuration of a sixteenth embodiment of a noise removal device of the present invention.

FIG. 48 is a block diagram showing an overall configuration of a conventional noise removal device.

FIG. 49 is a block diagram showing a configuration of a voice detection unit included in a conventional noise removal device.

50 is a block diagram showing a configuration of a power calculation unit included in the voice detection unit shown in FIG. 49.

51 is a block diagram showing a configuration of a weighted addition unit included in the voice detection unit shown in FIG. 49.

FIG. 52 is a block diagram showing a configuration of a multiple multiplication unit included in a conventional noise removal device.

FIG. 53 is a block diagram showing a configuration of an estimated noise calculation unit included in a conventional noise removal device.

54 is a block diagram showing a configuration of a frequency-specific estimated noise calculation unit included in the estimated noise calculation unit shown in FIG. 53.

FIG. 55 is a block diagram showing a configuration of an update determination unit included in the frequency-dependent estimated noise calculation unit shown in FIG. 54.

FIG. 56 is a block diagram showing a configuration of a frequency-specific SNR calculation unit included in a conventional noise removal device.

FIG. 57 is a block diagram showing a configuration of an estimated a priori SNR calculation unit included in a conventional noise removal device.

58 is a block diagram showing a configuration of a multiple range limitation processing unit included in the estimated a priori SNR calculation unit shown in FIG. 57.

59 is a block diagram showing a configuration of a multiple weighted addition unit included in the estimated a priori SNR calculation unit shown in FIG. 57.

FIG. 60 is a block diagram showing a configuration of a noise suppression coefficient generation unit included in a conventional noise removal device.

61 is a block diagram showing a configuration of a suppression coefficient search unit included in the noise suppression coefficient generation unit shown in FIG. 60.

[Explanation of symbols]

1 ... Frame division unit, 2, 22 ... Windowing processing unit, 3 ... Frame
Fourier transforming unit, 4 ... Voice detecting unit, 5, 51, 52, 53
... Estimated noise calculator, 6, 61, 715, 1402 ... Frequency
SNR calculator by number, 7, 71 ... Estimated innate SNR calculation
, 8, 81 ... Noise suppression coefficient generator, 9 ... Inverse Fourier transform
Conversion unit, 10 ... Frame synthesizing unit, 11 ... Input terminal, 12 ...
Output terminal, 13, 5049 ... Counter, 14 ... Weighted
Degraded speech calculation unit, 15 ... Suppression coefficient correction unit, 16, 17,
704, 705, 716, 1404 ... Multiplexing unit, 5
5, 58, 59, 662, 672, 682, 6542 ...
Injection noise calculator, 56, 57, 708, 4063, 40
72,4074,5046,6110₀ ~ 6110
_M-1 , 6543, 6544 ... Adder, 65, 66, 6
7, 68 ... SNR correction unit, 401, 1593, 520
4, 5206 ... Threshold storage unit, 402, 1594, 520
3, 5205, 67233 ... Comparison section, 404, 4075
... Constant multiplier, 405 ... Logarithmic calculator, 406 ... Power meter
Arithmetic unit, 407, 5071, 7071₀ ~ 7071_K-1 …
Weighted addition unit, 408, 706, 5072 ... Weight storage
Section, 409, 5202 ... Logical NOT circuit, 502, 50
5,602,603,802,803,1495,15
02,1503,1702,1703,4061,50
3,604,655,804,1475,1504,1
704, 6115, 7014, 7075 ... Multiplexing unit, 5
04₀ ~ 504_K-1 , 506₀ ~ 506 _K-1 , 507,
508₀ ~ 508_K-1 , 514₀ ~ 514_K-1 …frequency
Another estimated noise calculation unit, 520, 521, 522 ... Update determination
Section, 551 ... SNR calculation section, 552, 6541 ... Threshold
Value calculator, 553, 6721 ... Injection level calculator, 58
1,67232 ... Zero crossing calculation unit, 582, 1595,
5044, 6722 ... Switch, 591, 68232 ...
High frequency power calculator, 601₀ ~ 601_K-1 , 5041, 5
048, 6545 ... Division unit, 611, 612 ... By frequency
Power calculator, 651, 652, 653, 6111, 7
013, 7072, 7074 ... Separation part, 654₀ ~ 65
Four_K-1 , 664₀ ~ 664_K-1 ... corrected SNR calculation unit, 6
61, 663 ... Average value calculation unit, 701 ... Multiple range limiting process
Science section, 702 ... Acquired SNR storage section, 703 ... Suppression coefficient
Storage unit, 707 ... Multiple weighted addition unit, 712, 140
1, 5942 ... Estimated noise storage unit, 713 ... Enhanced speech power
-Spectrum storage unit, 801₀~ 801_K-1 … Suppression coefficient
Search unit, 811 ... MMSE STSA gain function value calculation
Part, 812 ... Generalized likelihood ratio calculation part, 813 ...
Rate storage unit, 814 ... Suppression coefficient calculation unit, 901 ... Degraded voice
Power, 902 ... Threshold value, 903, 923 ... Weight, 904
... update threshold value, 905 ... weighted addition unit control signal, 910
₀ ~ 910_K-1 , 910₀ ~ 910_ML-1… Deterioration by frequency
Voice power spectrum, 911₀ ~ 911 _K-1 , 911
₀ ~ 911_ML-1... Deteriorated voice power spectrum by band, 9
21 ... Instantaneous estimated SNR, 921₀ ~ 921_K-1 …frequency
Another instantaneous estimated SNR, 922 ... Past estimated SNR, 922
₀ ~ 922_K-1 ... Estimated SNR by frequency in the past, 924 ...
Estimated innate SNR, 924₀ ~ 924_K-1 … Inference by frequency
Constant a priori SNR, 1405 ... Multiple nonlinear processing unit, 148
5₀ ~ 1485_K-1 , 5042 ... Non-linear processing unit, 150
1₀ ~ 1501_K-1 ... Suppression coefficient correction unit for each frequency, 159
1,7012₀ ~ 7012_K-1 ... Maximum value selection section, 159
2 ... Suppression coefficient lower limit storage unit, 1596 ... Correction amount storage unit,
1597, 1701₀ ~ 1701_K-1 , 4062₀ ~ 4
062_K-1 , 4071, 4073, 5043 ... Multipliers,
5045 ... Shift register, 5047 ... Minimum value selection unit,
5201, 5211, 5221 ... Logical sum calculator 520
7 ... Threshold value calculation unit, 5941 ... Register length storage unit, 672
3, 6823 ... Judgment unit, 7011 ... Constant storage unit, 801
1 ... Suppression coefficient table, 8012, 8013 ... Address
Conversion unit, 67231 ... Silent section detection unit.

Claims (42)

[Claims] 1. An input signal is converted into a frequency domain signal, a signal to noise ratio is obtained using this frequency domain signal, a suppression coefficient is determined based on this signal to noise ratio, and the frequency is calculated using this suppression coefficient. In a noise removal method for obtaining an output signal from which noise is removed from the input signal by weighting a domain signal and converting the weighted frequency domain signal into a time domain signal, the frequency domain signal obtained by converting the input signal A method for removing noise, characterized in that the pseudo first noise is calculated based on the above, and the signal-to-noise ratio is obtained after adding the first noise to the frequency domain signal. 2. The noise removing method according to claim 1, wherein the addition of the first noise is selectively performed according to the property of the input signal. 3. The noise removing method according to claim 2, wherein the continuity of the signal is used as the property of the input signal. 4. The noise removing method according to claim 3, wherein a number of zero crossings at which the amplitude of the input signal becomes zero is used as the stationarity of the signal. 5. The noise removing method according to claim 3, wherein the high frequency power of the frequency domain signal obtained by converting the input signal is used as the stationarity of the signal. 6. The noise removing method according to claim 1, wherein second noise included in the frequency domain signal is estimated based on the frequency domain signal obtained by converting the input signal, and the second noise is estimated. Using the second noise and the frequency domain signal,
Noise removal method characterized by determining the power of the noise of the. 7. The noise removing method according to claim 1, wherein second noise included in the frequency domain signal is estimated based on the frequency domain signal obtained by converting the input signal, and the second noise is estimated. A first noise is calculated using the second noise and the frequency domain signal, and the sum of the first noise and the frequency domain signal and the sum of the first noise and the second noise are calculated. A noise removal method characterized by obtaining a signal-to-noise ratio using. 8. The noise removing method according to claim 6, wherein the frequency domain signal obtained by converting the input signal is weighted, and the second noise is estimated based on the weighted frequency domain signal. A noise removal method characterized by the above. 9. An input signal is converted into a frequency domain signal, a signal to noise ratio is obtained using this frequency domain signal, a suppression coefficient is determined based on this signal to noise ratio, and the frequency is calculated using this suppression coefficient. In a noise removal method for obtaining an output signal from which noise is removed from the input signal by weighting a domain signal and converting the weighted frequency domain signal into a time domain signal, the frequency domain signal obtained by converting the input signal A noise removal method characterized in that the signal-to-noise ratio is corrected based on the above, and the suppression coefficient is determined based on the corrected signal-to-noise ratio. 10. The noise removing method according to claim 9, wherein the correction of the signal-to-noise ratio is selectively performed according to the property of the input signal. 11. The noise removing method according to claim 10, wherein signal steadiness is used as the property of the input signal. 12. The noise removing method according to claim 11, wherein the continuity of the signal is a number of zero crossings at which the amplitude of the input signal becomes zero. 13. The noise removing method according to claim 11, wherein high frequency power of the frequency domain signal obtained by converting the input signal is used as stationarity of the signal. 14. The noise removing method according to claim 9, wherein noise included in the frequency domain signal is estimated based on the frequency domain signal obtained by converting the input signal, and the noise and the noise are estimated. A method of removing noise, wherein a correction amount of the signal-to-noise ratio is determined using a frequency domain signal. 15. The noise removing method according to claim 9, wherein noise included in the frequency domain signal is estimated based on the frequency domain signal obtained by converting the input signal, and the noise and the noise are included. A signal-to-noise ratio is used to obtain an addition signal, and the signal-to-noise ratio is recalculated using the sum of the addition signal and the frequency domain signal and the sum of the addition signal and the noise. A noise removing method characterized by correcting a noise ratio. 16. The noise removing method according to claim 14, wherein the frequency domain signal obtained by converting the input signal is weighted, and noise is estimated based on the weighted frequency domain signal. Noise removal method. 17. The noise removing method according to claim 1, wherein the suppression coefficient is corrected based on the frequency domain signal,
A method of removing noise, characterized in that the frequency domain signal is weighted using the corrected suppression coefficient. 18. The noise removing method according to claim 1, wherein the time domain signal obtained by converting the frequency domain signal is subjected to windowing processing. 19. Converting an input signal into a frequency domain signal,
The second noise included in the frequency domain signal is estimated based on the frequency domain signal, a value corresponding to the second noise is subtracted from the frequency domain signal to obtain a frequency domain enhanced voice, and the enhanced voice is obtained. A noise removal method for obtaining an output signal from which noise has been removed from the input signal by converting the input signal into a time domain signal, wherein a pseudo first noise is calculated based on the frequency domain signal obtained by converting the input signal, A noise removing method, characterized in that a value corresponding to the noise obtained by adding the first noise to the second noise is subtracted from the frequency domain signal to obtain a emphasized voice in the frequency domain. 20. The noise removing method according to claim 19, wherein the addition of the first noise is selectively performed according to the property of the input signal. 21. The noise removing method according to claim 20, wherein the stationarity of the signal is used as the property of the input signal. 22. The noise removing method according to claim 21, wherein the stationarity of the signal is a number of zero crossings at which the amplitude of the input signal becomes zero. 23. The noise removing method according to claim 21, wherein the high frequency power of the frequency domain signal obtained by converting the input signal is used as the stationarity of the signal. 24. The noise removing method according to claim 19, wherein the power of the first noise is determined by using the frequency domain signal and the second noise. Noise removal method. 25. The noise removing method according to claim 19, wherein the frequency domain signal obtained by transforming the input signal is weighted, and the second noise is based on the weighted frequency domain signal. A noise removal method characterized by estimating. 26. The noise removing method according to claim 25, wherein a signal-to-noise ratio is obtained by using the frequency domain signal obtained by converting the input signal, a weight is obtained by using the signal-to-noise ratio, and the weight is obtained. A method for removing noise, wherein the frequency domain signal is weighted using 27. The noise removing method according to claim 25, wherein a signal-to-noise ratio is obtained by using the frequency domain signal obtained by converting the input signal, and the signal-to-noise ratio is processed by a non-linear processing function to weight the signal. Is obtained, and the frequency domain signal is weighted using this weight. 28. The noise removing method according to claim 19, wherein windowing processing is performed on the time domain signal obtained by converting the emphasized speech in the frequency domain. 29. Converting an input signal into a frequency domain signal,
A signal-to-noise ratio is obtained using this frequency-domain signal, a suppression coefficient is determined based on this signal-to-noise ratio, the frequency-domain signal is weighted using this suppression coefficient, and this weighted frequency-domain signal is A noise removal method for obtaining an output signal from which noise has been removed from the input signal by converting into a time domain signal, wherein the time domain signal obtained by converting the frequency domain signal is subjected to windowing processing. . 30. Converting an input signal into a frequency domain signal,
The second noise included in the frequency domain signal is estimated based on the frequency domain signal, a value corresponding to the second noise is subtracted from the frequency domain signal to obtain a frequency domain enhanced voice, and the enhanced voice is obtained. To a time-domain signal to obtain an output signal from which noise has been removed from the input signal, wherein the time-domain signal obtained by converting the emphasized speech in the frequency domain is subjected to windowing processing. Noise removal method. 31. A first windowing processing section for applying windowing processing to an input signal and outputting the input signal, and an input signal windowed by the first windowing processing section is converted into a frequency domain signal, and an amplitude is obtained. A converter that separates and outputs a component and a phase component; an estimated noise calculator that estimates and outputs the second noise included in the frequency domain signal based on the amplitude component of the frequency domain signal; Noise and the amplitude component of the frequency domain signal to calculate and output a pseudo first noise, and an injection noise calculation unit that adds the amplitude components of the first noise and the frequency domain signal and outputs the result. A first adder, a second adder that adds and outputs the first noise and the second noise, an output signal of the first adder, and an output of the second adder A first signal that receives a signal and obtains and outputs a first signal-to-noise ratio A signal-to-noise ratio calculation unit, a suppression coefficient generation unit that determines and outputs a suppression coefficient based on the first signal-to-noise ratio, and weights the amplitude component of the frequency domain signal using the suppression coefficient. A first multiplying unit for outputting; an inverse transforming unit for transforming and outputting the amplitude component of the frequency domain signal and the phase component of the frequency domain signal weighted by the first multiplying unit into a time domain signal; A noise removing device comprising at least a second windowing processing unit that performs windowing processing on a time domain signal and outputs the windowed signal. 32. The noise removing device according to claim 31, wherein the injection noise calculation unit calculates the number of zero crossings at which the input signal is input and the amplitude of the input signal is zero, and the calculation result is obtained. A zero crossing calculation section for outputting a control signal according to the above, and a switch for selectively setting the first noise to zero according to the control signal input from the zero crossing calculation section. Removal device. 33. The noise eliminator according to claim 31, wherein the injection noise calculation unit calculates the high frequency power of the amplitude component of the frequency domain signal input from the conversion unit, and according to the calculation result. And a switch for selectively setting the first noise to zero according to the control signal input from the high band power calculation unit. Removal device. 34. The noise removing device according to claim 31, wherein the amplitude component of the frequency domain signal is weighted, and the obtained weighted amplitude component is output to the estimated noise calculation unit, The noise removing apparatus further comprising a weighted deteriorated speech calculation unit that causes the estimated noise calculation unit to estimate the second noise based on the weighted amplitude component. 35. The noise removing apparatus according to claim 34, wherein the weighted deteriorated speech calculation unit calculates and outputs a second signal-to-noise ratio using an amplitude component of the frequency domain signal. And a second signal-to-noise ratio calculation section that is input from the second signal-to-noise ratio calculation section.
A non-linear processing unit that processes the signal-to-noise ratio of the signal by a non-linear function to obtain a weight and outputs the weight, and an amplitude component of the frequency domain signal is weighted using the weight input from the non-linear processing unit, and the estimated noise A noise removing device, comprising: a second multiplying unit that outputs the calculating unit. 36. The noise removing device according to claim 31, wherein the suppression coefficient input from the suppression coefficient generation unit is corrected based on the frequency domain signal, and the first multiplication is performed. The noise removal apparatus further comprising a suppression coefficient correction unit that outputs the amplitude component of the frequency domain signal using the suppression coefficient corrected by the first multiplication unit. 37. A first windowing processing section for subjecting an input signal to windowing processing and outputting the same, and an input signal subjected to windowing processing by the first windowing processing section is converted into a frequency domain signal to obtain an amplitude. A converter for separating and outputting a component and a phase component; a first signal-to-noise ratio calculator for calculating and outputting a first signal-to-noise ratio using the amplitude component of the frequency-domain signal; An estimated noise calculator that estimates and outputs noise included in the frequency domain signal based on the amplitude component of the signal, and corrects the first signal-to-noise ratio using the noise and the amplitude component of the frequency domain signal. Then, a signal-to-noise ratio correction unit that outputs as a correction signal-to-noise ratio, a suppression coefficient generation unit that determines and outputs a suppression coefficient based on the corrected signal-to-noise ratio, and the frequency-domain signal using the suppression coefficient. Weight the amplitude components of A first multiplication unit that outputs the frequency domain signal and an inverse transformation unit that converts the amplitude component of the frequency domain signal weighted by the first multiplication unit and the phase component of the frequency domain signal into a time domain signal and outputs the time domain signal. A noise removing device comprising at least a second windowing processing unit that performs windowing processing on the time domain signal. 38. The noise removing device according to claim 37, wherein the signal-to-noise ratio correction unit calculates the number of zero crossings at which the input signal is input and the amplitude of the input signal is zero. A determination unit that outputs a control signal according to a calculation result, and the correction signal-to-noise ratio is selectively set to the same value as the first signal-to-noise ratio before correction by the control signal input from the determination unit. A noise removing device comprising a setting switch. 39. The noise eliminator according to claim 37, wherein the signal-to-noise ratio correction unit calculates the high frequency power of the amplitude component of the frequency domain signal input from the conversion unit, and the calculation result thereof. And a control unit that outputs a control signal according to the control signal, and the control signal input from the determination unit selectively sets the correction signal-to-noise ratio to the same value as the first signal-to-noise ratio before correction. A noise eliminator including a switch. 40. The noise eliminator according to claim 37, wherein the amplitude component of the frequency domain signal is weighted, and the obtained weighted amplitude component is output to the estimated noise calculator. The noise removing apparatus further comprising a weighted deteriorated speech calculation unit that causes the estimated noise calculation unit to estimate the noise based on the weighted amplitude component. 41. The noise removal apparatus according to claim 40, wherein the weighted deteriorated speech calculation unit calculates and outputs a second signal-to-noise ratio using an amplitude component of the frequency domain signal. And a second signal-to-noise ratio calculation section that is input from the second signal-to-noise ratio calculation section.
A non-linear processing unit that processes the signal-to-noise ratio of the signal by a non-linear function to obtain a weight and outputs the weight, and an amplitude component of the frequency domain signal is weighted using the weight input from the non-linear processing unit, and the estimated noise A noise removing device, comprising: a second multiplying unit that outputs the calculating unit. 42. The noise removing device according to claim 37, wherein the suppression coefficient input from the suppression coefficient generation unit is corrected based on the frequency domain signal, and the first multiplication is performed. The noise removal apparatus further comprising a suppression coefficient correction unit that outputs the amplitude component of the frequency domain signal using the suppression coefficient corrected by the first multiplication unit.