JP4968355B2 - Method and apparatus for noise suppression - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and a device for noise suppression capable of obtaining an emphasis voice which is excellent in subjective voice quality without generating noise in an output signal, and achieving high emphasis voice quality by using a suppression coefficient suitable for all background noise. <P>SOLUTION: The device includes: a coefficient calculation section for a silence part which calculates a coefficient for the silence part on the basis of an emphasis voice power spectrum and an estimation noise power spectrum; a coefficient storage section for storing a coefficient for a sonant part; and a post-suppression coefficient calculation section for calculating a post-suppression coefficient on the basis of the obtained coefficient for the silence part and that for the sonant part. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a noise suppression method and apparatus for suppressing noise superimposed on a desired audio signal.

  The noise suppressor is a technology that suppresses noise (noise) superimposed on the desired audio signal. The power spectrum of the noise component is estimated using the input signal converted to the frequency domain, and this estimated power spectrum is input. By subtracting from the signal, it operates to suppress noise mixed in the desired audio signal. By continuously estimating the power spectrum of the noise component, it can also be applied to non-stationary noise suppression. As a noise suppressor, for example, there is a method described in Patent Document 1. FIG. 36 shows the configuration of the noise suppressor described in Patent Document 1.

The input terminal 11 is supplied with a degraded voice signal (a signal in which a desired voice signal and noise are mixed) as a sample value series. The deteriorated speech signal samples are supplied to the frame dividing unit 1 and divided into frames for every K / 2 samples. Here, K is an even number. The degraded speech signal samples divided into frames are supplied to the windowing processing unit 2 and multiplied by the window function w (t). The signal y _n (t) bar windowed with w (t) for the input signal y _n (t) (t = 0, 1,..., K / 2-1) of the nth frame is given by Given.

In addition, it is also widely performed to overlap a part of two consecutive frames. Assuming 50% of the frame length as the overlap length, t = 0, 1,. . . , For K / 2-1,

Y _n (t) bars (t = 0, 1,..., K−1) obtained in the above are output from the windowing processing unit 2. For real signals, a symmetric window function is used. The window function is designed so that the input signal and the output signal when the suppression coefficient is set to 1 match except for calculation errors. This means that w (t) + w (t + K / 2) = 1.

Hereinafter, the description will be continued by taking as an example a case in which 50% of two consecutive frames overlap each other. As w (t), for example, a Hanning window represented by the following equation can be used.

The windowed output y _n (t) bar is supplied to the Fourier transform unit 3 and converted into a degraded speech spectrum Y _n (k). The degraded speech spectrum Y _n (k) is separated into phase and amplitude, the degraded speech phase spectrum arg Y _n (k) is sent to the inverse Fourier transform unit 9, and the degraded speech amplitude spectrum | Y _n (k) | 13 and the multiple multiplier 16.

The multiplex multiplier 13 calculates a deteriorated speech power spectrum using the supplied degraded speech amplitude spectrum | Y _n (k) |, an estimated noise calculator 5, a frequency-specific SNR (signal-to-noise ratio) calculator 6, and This is transmitted to the weighted deteriorated speech calculation unit 14. The weighted deteriorated sound calculation unit 14 calculates a weighted deteriorated sound power spectrum using the deteriorated sound power spectrum supplied from the multiple multiplier 13 and transmits the weighted deteriorated sound power spectrum to the estimated noise calculation unit 5. The estimated noise calculation unit 5 estimates the noise power spectrum using the deteriorated voice power spectrum, the weighted deteriorated voice power spectrum, and the count value supplied from the counter 4, and the SNR calculation unit 6 for each frequency as the estimated noise power spectrum. To communicate. The frequency-specific SNR calculation unit 6 calculates the SNR for each frequency using the input degraded speech power spectrum and the estimated noise power spectrum, and supplies the SNR to the estimated innate SNR calculation unit 7 and the noise suppression coefficient generation unit 8 as an acquired SNR. To do.

The estimated innate SNR calculation unit 7 estimates the innate SNR using the acquired acquired SNR and the corrected suppression coefficient supplied from the suppression coefficient correction unit 15, and a noise suppression coefficient generation unit as the estimated innate SNR 8 is transmitted. The noise suppression coefficient generation unit 8 generates a noise suppression coefficient using the acquired SNR supplied as input, the estimated innate SNR, and the speech non-existence probability supplied from the speech non-existence probability storage unit 21, and serves as a suppression coefficient. This is transmitted to the suppression coefficient correction unit 15. The suppression coefficient correction unit 15 corrects the suppression coefficient using the input estimated innate SNR and the suppression coefficient, and supplies the correction coefficient to the multiple multiplication unit 16 as a corrected suppression coefficient G _n (k) bar. The multiplex multiplier 16 weights the deteriorated speech amplitude spectrum | Y _n (k) | supplied from the Fourier transform unit 3 with the corrected suppression coefficient G _n (k) bar supplied from the suppression coefficient correction unit 15. The emphasized speech amplitude spectrum | X _n (k) | bar is obtained and transmitted to the inverse Fourier transform unit 9. | X _n (k) | bar is given by equation (4).

The inverse Fourier transform unit 9 multiplies the enhanced speech amplitude spectrum | X _n (k) | bar supplied from the multiple multiplication unit 16 and the degraded speech phase spectrum arg Y _n (k) supplied from the Fourier transform unit 3. Then, the emphasized speech X _n (k) bar is obtained. That is,

Execute.

The obtained emphasized speech X _n (k) bar is subjected to inverse Fourier transform, and a time-domain sample value series x _n (t) bar (t = 0, 1,. -1) is transmitted to the frame synthesis unit 10. The frame synthesis unit 10 takes out K / 2 samples from two adjacent frames of the x _n (t) bar and superimposes them,

To obtain the emphasized speech x _n (t) hat. The obtained enhanced speech x _n (t) hat (t = 0, 1,..., K−1) is transmitted to the output terminal 12 as an output of the frame synthesis unit 10.

FIG. 37 is a block diagram showing a configuration of the multiple multiplier 13 included in FIG. The multiple multiplier 13 includes multipliers 1301 _{0 to} 1301 _K−1 , demultiplexers 1302 and 1303, and a multiplexer 1304. The degraded speech amplitude spectrum supplied from the Fourier transform unit 3 in FIG. 36 in the multiplexed state is separated into K samples for each frequency in the separation units 1302 and 1303 and supplied to the multipliers 1301 _{0 to} 1301 _K−1 , respectively. Is done. Multipliers 1301 _{0 to} 1301 _K−1 square the input signals, respectively, and transmit them to multiplexing section 1304. The multiplexing unit 1304 multiplexes the input signal and outputs it as a degraded voice power spectrum.

  FIG. 38 is a block diagram showing the configuration of the weighted deteriorated speech calculation unit 14. The weighted degraded speech calculation unit 14 includes an estimated noise storage unit 1401, a frequency-specific SNR calculation unit 1402, a multiple nonlinear processing unit 1405, and a multiple multiplication unit 1404. The estimated noise storage unit 1401 stores the estimated noise power spectrum supplied from the estimated noise calculation unit 5 of FIG. 36, and outputs the estimated noise power spectrum stored one frame before to the SNR calculation unit 1402 for each frequency. The frequency-specific 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 degraded speech power spectrum supplied from the multiplex multiplier 13 in FIG. The data is output to the processing unit 1405. The multiple nonlinear processing unit 1405 calculates a weight coefficient vector using the SNR supplied from the frequency-specific SNR calculation section 1402, and outputs the weight coefficient vector to the multiple multiplication section 1404.

  The multiplex multiplier 1404 calculates the product of the degraded speech power spectrum supplied from the multiplex multiplier 13 in FIG. 36 and the weight coefficient vector supplied from the multiplex nonlinear processor 1405 for each frequency, and weighted degraded speech power spectrum. Is output to the estimated noise storage unit 5 of FIG. The configuration of the multiple multiplier 1404 is the same as that of the multiple multiplier 13 already described with reference to FIG.

FIG. 39 is a block diagram showing the configuration of the frequency-specific SNR calculator 1402 included in FIG. The frequency-specific SNR calculation unit 1402 includes division units 1421 _{0 to} 1421 _K−1 , separation units 1422 and 1423, and a multiplexing unit 1424. The degraded sound power spectrum supplied from the multiple multiplier 13 in FIG. 36 is transmitted to the separator 1422. The estimated noise power spectrum supplied from the estimated noise storage unit 1401 in FIG. 38 is transmitted to the separation unit 1423. The degraded voice power spectrum is separated into K samples corresponding to the frequency components in the separation unit 1422 and the estimated noise power spectrum is separated in the separation unit 1423, and supplied to the division units 1421 _{0 to} 1421 _K−1 . The division units 1421 _{0 to} 1421 _K−1 divide the supplied degraded speech power spectrum by the estimated noise power spectrum according to the equation (7) to obtain the frequency-specific SNRγ _n (k) hat, and transmit it to the multiplexing unit 1424. To do.

Here, λ _n-1 (k) is an estimated noise power spectrum stored one frame before. The multiplexing unit 1424 multiplexes the transmitted K frequency-specific SNRs, and transmits the multiplexed SNRs to the multiple nonlinear processing unit 1405 in FIG.

Next, the configuration and operation of the multiple nonlinear processing unit 1405 of FIG. 38 will be described in detail with reference to FIG. FIG. 40 is a block diagram illustrating a configuration of the multiple nonlinear processing unit 1405 included in the weighted deteriorated speech calculation unit 14. The multiple nonlinear processing unit 1405 includes a separation unit 1495, nonlinear processing units 1485 _{0 to} 1485 _K−1 , and a multiplexing unit 1475. The separation unit 1495 separates the SNR supplied from the frequency-specific SNR calculation unit 1402 in FIG. 38 into frequency-specific SNRs, and outputs them to the nonlinear processing units 1485 _{0 to} 1485 _K−1 . Each of the nonlinear processing units 1485 _{0 to} 1485 _K−1 has a nonlinear function that outputs a real value corresponding to the input value.

FIG. 41 shows an example of a nonlinear function. When f ₁ is an input value, the output value f ₂ of the nonlinear function shown in FIG.

Given in. However, a and b are arbitrary real numbers.

The non-linear processing units 1485 _{0 to} 1485 _K−1 process the frequency-specific SNR supplied from the demultiplexing unit 1495 by a non-linear function to obtain a weighting coefficient, and output it to the multiplexing unit 1475. That is, the non-linear processing units 1485 _{0 to} 1485 _K−1 output weighting coefficients from 1 to 0 according to the SNR. 1 is output when the SNR is small, and 0 is output when the SNR is large. The multiplexing unit 1475 multiplexes the weighting factors output from the non-linear processing units 1485 _{0 to} 1485 _K−1, and outputs the result to the multiplexing multiplication unit 1404 as a weighting factor vector.

  The weighting coefficient multiplied by the degraded speech power spectrum in the multiple multiplier 1404 in FIG. 38 has a value corresponding to the SNR. The greater the SNR, that is, the greater the speech component included in the degraded speech, The value becomes smaller. In general, a degraded speech power spectrum is used to update the estimated noise. However, a speech component included in the degraded speech power spectrum can be obtained by weighting the degraded speech power spectrum used to update the estimated noise according to the SNR. Can be reduced, and more accurate noise estimation can be performed. In addition, although the example which used the nonlinear function for the calculation of a weighting coefficient was shown, it is also possible to use the function of SNR represented with other forms, such as a linear function and a high-order polynomial, besides a nonlinear function.

FIG. 42 is a block diagram showing a configuration of estimated noise calculation unit 5 included in FIG. The noise estimation calculation unit 5 includes separation units 501 and 502, a multiplexing unit 503, and frequency-specific estimation noise calculation units 504 _{0 to} 504 _K−1 . The separation unit 501 separates the weighted deteriorated sound power spectrum supplied from the weighted deteriorated sound calculation unit 14 of FIG. 36 into the weighted deteriorated sound power spectrum for each frequency, and the frequency-specific estimated noise calculation units 504 _{0 to} 504 _{K. To -1} . Separation section 502 separates the degraded speech power spectrum supplied from multiplex multiplication section 13 in FIG. 36 into degraded speech power spectra for each frequency and outputs them to frequency-specific estimated noise calculation sections 504 _{0 to} 504 _K−1 . 36. The frequency-specific estimated noise calculation units 504 _{0 to} 504 _K−1 are the frequency _- dependent weighted degraded speech power spectrum supplied from the separation unit 501, the frequency-based degraded speech power spectrum supplied from the separation unit 502, and the counter of FIG. 4 calculates an estimated noise power spectrum for each frequency from the count value supplied from 4, and outputs it to the multiplexing unit 503. The multiplexing unit 503 multiplexes the frequency-specific estimated noise power spectrum supplied from the frequency-specific estimated noise calculation units 504 _{0 to} 504 _K−1, and weights the estimated noise power spectrum with the frequency-specific SNR calculation unit 6 of FIG. Output to the deteriorated speech calculator 14. Detailed configuration and operation of the frequency-specific estimated noise calculators 504 _{0 to} 504 _K−1 will be described with reference to FIG.

FIG. 43 is a block diagram showing a configuration of frequency-specific estimated noise calculation units 504 _{0 to} 504 _K−1 included in FIG. The frequency-specific estimated noise calculation units 504 _{0 to} 504 _K−1 include an update determination unit 520, a register length storage unit 5041, an estimated noise storage unit 5042, a switch 5044, a shift register 5045, an adder 5046, a minimum value selection unit 5047, and a division. Part 5048 and counter 5049. The switch 5044 is supplied with the frequency-dependent weighted degraded sound power spectrum from the separation unit 501 in FIG. When the switch 5044 closes the circuit, the frequency-specific weighted degraded sound power spectrum is transmitted to the shift register 5045. The shift register 5045 shifts the stored value of the internal register to the adjacent register in accordance with the control signal supplied from the update determination unit 520. The shift register length is equal to a value stored in a register length storage unit 5041 described later. All register outputs of the shift register 5045 are supplied to the 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 520 is supplied with a count value, a frequency-specific degraded speech power spectrum, and a frequency-specific estimated noise power spectrum. The update determination unit 520 always sets “1” until the count value reaches a preset value, and after reaching the count value, sets “1” when the input deteriorated speech signal is determined to be noise. At other times, “0” is output and transmitted to the counter 5049, the switch 5044, and the shift register 5045. The switch 5044 closes the circuit when the signal supplied from the update determination unit is “1” and opens when the signal is “0”. The counter 5049 increases the count value when the signal supplied from the update determination unit is “1”, and does not change when the signal is “0”. The shift register 5045 captures one sample of the signal sample supplied from the switch 5044 when the signal supplied from the update determination unit is “1”, and simultaneously shifts the stored value of the internal register to the adjacent register. The minimum value selection unit 5047 is supplied with the output of the counter 5049 and the output of the register length storage unit 5041.

The minimum value selection unit 5047 selects the smaller one of the supplied count value and register length and transmits it to the division unit 5048. The division unit 5048 divides the addition value of the degraded speech power spectrum for each frequency supplied from the adder 5046 by the smaller value of the count value or the register length, and sets the quotient as the estimated noise power spectrum for each frequency λ _n (k). Output. Assuming that B _n (k) (n = 0, 1,..., N−1) is a sample value of the degraded speech power spectrum stored in the shift register 5045, λ _n (k) is

Given in.

  However, N is the smaller value of the count value and the register length. Since the count value starts monotonically and increases monotonically, division is first performed by the count value, and thereafter division is performed by the register length. When division is performed by the register length, an average value of values stored in the shift register is obtained. At first, since not enough values are stored in the shift register 5045, the value is divided by the number of registers in which values are actually stored. 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 equal to the register length when the count value is larger than the register length.

  FIG. 44 is a block diagram showing the configuration of the update determination unit 520 included in FIG. The update determination unit 520 includes a logical sum calculation unit 5201, comparison units 5203 and 5205, threshold storage units 5204 and 5206, and a threshold calculation unit 5207. The count value supplied from the counter 4 in FIG. 36 is transmitted to the comparison unit 5203. The threshold value that is the output of the threshold value storage unit 5204 is also transmitted to the comparison unit 5203. The comparison unit 5203 compares the supplied count value with a threshold value, and transmits “1” to the logical sum calculation unit 5201 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 threshold value calculation unit 5207 calculates a value corresponding to the frequency-specific estimated noise power spectrum supplied from the estimated noise storage unit 5042 of FIG. 43 and outputs the value as a threshold value to the threshold value storage unit 5206. The simplest threshold calculation method is a constant multiple of the estimated noise power spectrum for each frequency. In addition, it is possible to calculate the threshold value using a high-order polynomial or a nonlinear function. The threshold storage unit 5206 stores the threshold output from the threshold calculation unit 5207 and outputs the threshold stored one frame before to the comparison unit 5205. The comparison unit 5205 compares the threshold supplied from the threshold storage unit 5206 with the frequency-specific degraded audio power spectrum supplied from the separation unit 502 shown in FIG. 42, and “1” if the frequency-specific degraded audio power spectrum is smaller than the threshold. Is larger, “0” is output to the logical sum calculation unit 5201.

  That is, it is determined whether or not the degraded speech signal is noise based on the magnitude of the estimated noise power spectrum. The logical sum calculation unit 5201 calculates a logical sum of the output value of the comparison unit 5203 and the output value of the comparison unit 5205, and outputs the calculation result to the switch 5044, the shift register 5045, and the counter 5049 in FIG. As described above, the update determination unit 520 outputs “1” when the deteriorated voice power is small not only in the initial state and the silent period but also in the voiced period. That is, the estimated noise is updated. Since the threshold is calculated for each frequency, the estimated noise can be updated for each frequency.

FIG. 45 is a block diagram showing the configuration of the estimated innate SNR calculation unit 7 included in FIG. The estimated innate SNR calculation unit 7 includes a multi-range limitation processing unit 701, an acquired SNR storage unit 702, a suppression coefficient storage unit 703, multiple multiplication units 704 and 705, a weight storage unit 706, a multiple weighted addition unit 707, an adder 708. The acquired SNRγ _n (k) (k = 0, 1,..., K−1) supplied from the frequency-specific SNR calculation unit 6 in FIG. 36 is transmitted to the acquired SNR storage unit 702 and the adder 708. The The acquired SNR storage unit 702 stores the acquired SNRγ _n (k) in the nth frame and transmits the acquired SNRγ _n−1 (k) in the ( _n−1 ) th frame to the multiple multiplier 705.

The corrected suppression coefficient G _n (k) bar (k = 0, 1,..., K−1) supplied from the suppression coefficient correction unit 15 in FIG. 36 is transmitted to the suppression coefficient storage unit 703. The suppression coefficient storage unit 703 stores the corrected suppression coefficient G _n (k) bar in the nth frame and transmits the corrected suppression coefficient G _n−1 (k) bar in the ( _n−1 ) _th frame to the multiple multiplication unit 704. . Multiplex multiplier 704 squares the supplied G _n (k) bar to obtain G ² _n−1 (k) bar, and transmits it to multiple multiplier 705. Multiplex multiplier 705 converts G ² _n−1 (k) bar and γ _n−1 (k) to k = 0, 1,. . . , K−1 is multiplied to obtain G ² _n−1 (k) bar γ _n−1 (k), and the result is transmitted to the multi-weighted addition unit 707 as the past estimated SNR 922. The configuration of the multiple multipliers 704 and 705 is the same as that of the multiple multiplier 13 already described with reference to FIG.

The other terminal of the adder 708 is supplied with −1, and the addition result γ _n (k) −1 is transmitted to the multi-value range limiting processing unit 701. The multi-range limitation processing unit 701 performs an operation using the range limitation operator P [•] on the addition result γ _n (k) −1 supplied from the adder 708, and the result P [γ _n (k) −1. ] As the instantaneous estimated SNR 921. However, P [x] is defined by Formula (10).

A weight 923 is also supplied from the weight storage unit 706 to the multiple weighted addition unit 707. The multiple weighted addition unit 707 obtains an estimated innate SNR 924 using the supplied instantaneous estimated SNR 921, past estimated SNR 922, and weight 923. If the weight 923 is α and ξ _n (k) hat is the estimated innate SNR, ξ _n (k) hat is calculated by the equation (11).

Here, G ² ₋₁ (k) γ ₋₁ (k) bar = 1.

FIG. 46 is a block diagram showing the configuration of the multi-range limitation processing unit 701 included in FIG. The multi-value range limitation processing unit 701 includes a constant storage unit 7011, maximum value selection units 7012 _{0 to} 7012 _K−1 , a separation unit 7013, and a multiplexing unit 7014. Γ _n (k) −1 is supplied to the separation unit 7013 from the adder 708 in FIG. The separation unit 7013 separates the supplied γ _n (k) −1 into K frequency-specific components, and supplies them to the maximum value selection units 7012 _{0 to} 7012 _K−1 . Zeros are supplied from the constant storage unit 7011 to the other inputs of the maximum value selection units 7012 _{0 to} 7012 _K−1 . Maximum value selection sections 7012 _{0 to} 7012 _K−1 compare γ _n (k) −1 with zero and transmit the larger value to multiplexing section 7014. This maximum value selection calculation corresponds to executing Expression (10). The multiplexing unit 7014 multiplexes these values and outputs them.

FIG. 47 is a block diagram showing the configuration of the multiple weighted addition unit 707 included in FIG. The multiple weighted addition unit 707 includes weighted addition units 7071 _{0 to} 7071 _K−1 , separation units 7072 and 7074, and a multiplexing unit 7075. The separation unit 7072 is supplied with P [γ _n (k) −1] as the instantaneous estimated SNR 921 from the multi-value range limitation processing unit 701 in FIG. Separation unit 7072, P a [γ _n (k) -1] is separated into K frequency-components, as frequency-instantaneous estimation SNR921 ₀ ~921 _K-1, weighted adder 7071 ₀ ~7071 _K-1 To communicate. The separation unit 7074 is supplied with G ² _n−1 (k) bar γ _n−1 (k) as the past constant SNR 922 from the multiple multiplication unit 705 in FIG.

The separation unit 7074 separates G ² _n-1 (k) bar γ _n-1 (k) into K frequency-specific components, and sets the weighted addition unit as past frequency-specific estimated SNRs 922 _{0 to} 922 _K−1. 7071 _{0 to} 7071 _K−1 . On the other hand, weights 923 are also supplied to the weighted addition units 7071 _{0 to} 7071 _K−1 . The weighted addition units 7071 _{0 to} 7071 _K−1 perform weighted addition represented by Expression (11), and transmit the frequency-specific estimated innate SNRs 924 _{0 to} 924 _K−1 to the multiplexing unit 7075. Multiplexing section 7075 multiplexes frequency-specific estimated innate SNRs 924 _{0 to} 924 _K−1 and outputs them as estimated innate SNR 924. The operation and configuration of the weighted addition units 7071 _{0 to} 7071 _K−1 will be described next with reference to FIG.

FIG. 48 is a block diagram showing a configuration of the weighted addition unit 7071 included in FIG. The weighted addition unit 7071 includes multipliers 7091 and 7093, a constant multiplier 7095, and adders 7092 and 7094. 47, the instantaneous frequency-specific estimated SNR 921 is supplied from the separation unit 7072, the past frequency SNR 922 is supplied from the separation unit 7074, and the weight 923 is supplied from the weight storage unit 706 of FIG. The weight 923 having the value α is transmitted to the constant multiplier 7095 and the multiplier 7093. The constant multiplier 7095 transmits −α obtained by multiplying the input signal by −1 to the adder 7094. 1 is supplied as the other input of the adder 7094, and the output of the adder 7094 is 1-α which is the sum of both. 1-α is supplied to a multiplier 7091 and is multiplied by the other frequency-specific instantaneous estimated SNR P [γ _n (k) −1], which is the product, (1-α) P [γ _n ( k) −1] is transmitted to the adder 7092. On the other hand, the multiplier 7093 multiplies α supplied as the weight 923 by the past estimated SNR 922 and transmits the product αG ² _n−1 (k) bar γ _n−1 (k) to the adder 7092. . Adder 7092 outputs the sum of (1-α) P [γ _n (k) -1] and αG ² _n-1 (k) bar γ _n-1 (k) as frequency-specific estimated innate SNR 924. .

  FIG. 49 is a block diagram showing the noise suppression coefficient generation unit 8 included in FIG. The noise suppression coefficient generation unit 8 includes an MMSE STSA gain function value calculation unit 811, a generalized likelihood ratio calculation unit 812, and a suppression coefficient calculation unit 814. Hereinafter, based on the calculation formula described in Patent Document 1, a calculation method of the suppression coefficient will be described.

The frame number is n, the frequency number is k, γ _n (k) is the acquired frequency-specific SNR supplied from the frequency-specific SNR calculator 6 in FIG. 36, and ξ _n (k) is the estimated innate SNR in FIG. The frequency-specific estimated innate SNR, q supplied from the calculation unit 7 is defined as the speech non-existence probability supplied from the speech non-existence probability storage unit 21 of FIG. Further, η _n (k) = ξ _n (k) hat / (1-q), v _n (k) = (η _n (k) γ _n (k)) / (1 + η _n (k)). The MMSE STSA gain function value calculator 811 obtains the acquired SNR γ _n (k) supplied from the frequency-specific SNR calculator 6 in FIG. 36 and the estimated innate SNR supplied from the estimated innate SNR calculator 7 in FIG. An MMSE STSA gain function value is calculated for each frequency based on ξ _n (k) hat and the speech absence probability q supplied from the speech absence probability storage unit 21 in FIG. Output. The MMSE STSA gain function value G _n (k) for each frequency is

Given in.

Here, I ₀ (z) is a zero-order modified Bessel function, and I ₁ (z) is a first-order modified Bessel function. Non-Patent Document 1 describes the modified Bessel function.

The generalized likelihood ratio calculator 812 obtains the acquired SNR γ _n (k) supplied from the frequency-specific SNR calculator 6 in FIG. 36 and the estimated innate SNR supplied from the estimated innate SNR calculator 7 in FIG. Based on ξ _n (k) hat and the speech non-existence probability q supplied from the speech non-existence probability storage unit 21 of FIG. 36, a generalized likelihood ratio is calculated for each frequency and output to the suppression coefficient calculation unit 814. To do. The generalized likelihood ratio Λ _n (k) for each frequency is

Given in.

The suppression coefficient calculation unit 814 receives the MMSE STSA gain function value G _n (k) supplied from the MMSE STSA gain function value calculation unit 811 and the generalized likelihood ratio Λ _n (supplied from the generalized likelihood ratio calculation unit 812. The suppression coefficient is calculated for each frequency from k) and output to the suppression coefficient correction unit 15 in FIG. The suppression coefficient G _n (k) bar for each frequency is

Given in. Instead of calculating the SNR for each frequency, it is also possible to obtain and use an SNR common to a band composed of a plurality of frequencies.

50 is a block diagram showing the suppression coefficient correction unit 15 included in FIG. The suppression coefficient correction unit 15 includes frequency-specific suppression coefficient correction units 1501 _{0 to} 1501 _K−1 , separation units 1502 and 1503, and a multiplexing unit 1504.

Separation section 1502 separates the estimated innate SNR supplied from estimated innate SNR calculation section 7 in FIG. 36 into frequency-specific components, and outputs them to frequency-specific suppression coefficient correction sections 1501 _{0 to} 1501 _K−1 . Separation section 1503 separates the suppression coefficient supplied from suppression coefficient generation section 8 of FIG. 36 into frequency-specific components, and outputs them to frequency-specific suppression coefficient correction sections 1501 _{0 to} 1501 _K−1 . Frequency-specific suppression coefficient correction units 1501 _{0 to} 1501 _K−1 calculate frequency-specific correction suppression coefficients from frequency-specific estimated innate SNRs supplied from separation unit 1502 and frequency-specific suppression coefficients supplied from separation unit 1503. And output to the multiplexing unit 1504. 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 uses the multiple multiplication unit 1 of FIG. 36 and the estimated innate SNR calculation unit as the correction suppression coefficients. 7 is output.

Next, the configuration and operation of the frequency-specific suppression coefficient correction units 1501 _{0 to} 1501 _K−1 will be described in detail with reference to FIG.

FIG. 51 is a block diagram showing the configuration of frequency-specific suppression coefficient correction units 1501 _{0 to} 1501 _K−1 included in the suppression coefficient correction unit 15. The frequency-specific suppression coefficient correction unit 1501 includes a maximum value selection unit 1591, a suppression coefficient lower limit value storage unit 1592, a threshold storage unit 1593, a comparison unit 1594, a switch 1595, a correction value storage unit 1596, and a multiplier 1597.

  The comparison unit 1594 compares the threshold supplied from the threshold storage unit 1593 with the frequency-specific estimated innate SNR supplied from the separation unit 1502 in FIG. 50. If the frequency-specific estimated innate SNR is larger than the threshold, “0” is output. "Is supplied to the switch 1595 if it is smaller. The switch 1595 outputs the frequency-specific suppression coefficient supplied from the separation unit 1503 in FIG. 50 to the multiplier 1597 when the output value of the comparison unit 1594 is “1”, and the maximum value selection unit when the output value is “0”. It outputs to 1591. That is, when the frequency-specific estimated innate SNR is smaller than the threshold value, the suppression coefficient is corrected. Multiplier 1597 calculates the product of the output value of switch 1595 and the output value of correction value storage unit 1596 and outputs the product to maximum value selection unit 1591.

  On the other hand, the suppression coefficient lower limit value 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 calculates the frequency-specific suppression coefficient supplied from the separation unit 1503 in FIG. 50 or the product calculated by the multiplier 1597 and the suppression coefficient lower limit value supplied from the suppression coefficient lower limit value storage unit 1592. The larger value is compared and output to multiplexing section 1504 in FIG. In other words, the suppression coefficient is necessarily larger than the lower limit value stored in the suppression coefficient lower limit value storage unit 1592.

Japanese Patent Laid-Open No. 2002-20417

1985, Mathematical Dictionary, Iwanami Shoten, 374. G page December 1979, Proceedings of the IEE, Vol. 67, No. 12 (PROCEEDINGS OF THE IEEE, VOL.67, NO.12, PP.1586-1604, DEC, 1979 ), Pages 1586-1604 April 1979, IEE Transactions on Axetics Speech and Signal Processing, Vol. 27, No. 2 (IEEETRANSACTIONS ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, VOL. 27 , No. 2, PP. 113-120, APR, 1979), pages 113-120.

  In the related art methods described so far, noise suppression is always performed using a suppression coefficient obtained by the same calculation method without distinguishing between a speech section and a noise section. For this reason, there is a problem that voice distortion occurs in the voice section, and suppression in the noise section becomes insufficient.

  It is an object of the present invention to reduce speech distortion in a speech section by distinguishing between a speech section and a noise section and performing noise suppression using a suppression coefficient obtained by a calculation method suitable for each, and sufficient in the noise section. It is an object of the present invention to provide a noise suppression method and apparatus capable of achieving various suppressions.

The noise suppression method and apparatus according to the present invention is characterized by using post-suppression that performs further suppression after noise suppression based on the silence part coefficient and the sound part coefficient.

More specifically, a silent part coefficient calculating unit that calculates a silent part coefficient based on the emphasized speech power spectrum and the estimated noise power spectrum, a voiced part coefficient storage part that stores a voiced part coefficient, A post-suppression coefficient calculator for calculating a post-suppression coefficient based on the obtained silent part coefficient and sound part coefficient is provided.

  In the present invention, since the suppression coefficient is corrected using the coefficient for the silent part calculated based on the emphasized voice power spectrum and the estimated noise power spectrum and the coefficient for the voice part, the voice section is based on the coefficient for the voice part. It is possible to weaken the suppression and perform post-suppression in the noise interval to increase the suppression based on the coefficients for the silenced voice power spectrum and the estimated noise power spectrum in the noise interval. It is possible to obtain emphasized speech with less.

The block diagram which shows the 1st Embodiment of this invention. The block diagram which shows the structure of the emphasis audio | voice spectrum correction | amendment part contained in the 1st Embodiment of this invention. The block diagram which shows the structure of the audio | voice existence probability calculation part contained in FIG. The block diagram which shows the structure of the smoothing part contained in FIG. The block diagram which shows the structure of the post-suppression coefficient calculation part contained in FIG. FIG. 6 is a block diagram illustrating a configuration of a frequency-specific post-frequency suppression coefficient calculation unit included in FIG. 5. The block diagram which shows the structure of the coefficient calculation part for silence parts contained in FIG. The figure which shows an example of the nonlinear function in the coefficient calculation part contained in FIG. The block diagram which shows the 2nd Embodiment of this invention. The block diagram which shows the structure of the emphasis audio | voice spectrum correction | amendment part contained in the 2nd Embodiment of this invention. The block diagram which shows the structure of the post-suppression coefficient calculation part contained in FIG. The block diagram which shows the structure of the post-frequency suppression coefficient calculation part contained in FIG. The block diagram which shows the 3rd Embodiment of this invention. The block diagram which shows the structure of the emphasis audio | voice spectrum correction | amendment part contained in the 3rd Embodiment of this invention. The block diagram which shows the structure of the post-suppression coefficient calculation part contained in FIG. The block diagram which shows the structure of the post-frequency suppression coefficient calculation part contained in FIG. The block diagram which shows the 4th Embodiment of this invention. The block diagram which shows the structure of the emphasis audio | voice spectrum correction | amendment part contained in the 4th Embodiment of this invention. The block diagram which shows the structure of the post-suppression coefficient calculation part contained in FIG. FIG. 20 is a block diagram illustrating a configuration of a frequency-specific post-frequency suppression coefficient calculation unit included in FIG. 19. The block diagram which shows the 5th Embodiment of this invention. The block diagram which shows the structure of the presumed innate SNR calculation part contained in the 5th Embodiment of this invention. The block diagram which shows the structure of the suppression coefficient correction | amendment part contained in the 5th Embodiment of this invention. The block diagram which shows the structure of the suppression coefficient correction | amendment part classified by frequency contained in FIG. The block diagram which shows the 6th Embodiment of this invention. The block diagram which shows the structure of the presumed innate SNR calculation part contained in the 6th Embodiment of this invention. The block diagram which shows the structure of the suppression coefficient correction | amendment part contained in the 6th Embodiment of this invention. The block diagram which shows the structure of the estimation suppression coefficient correction part classified by frequency contained in FIG. The block diagram which shows the 7th Embodiment of this invention. The block diagram which shows the emphasis audio | voice amplitude spectrum correction | amendment part contained in the 7th Embodiment of this invention. The block diagram which shows the 8th Embodiment of this invention. The block diagram which shows the emphasis audio | voice amplitude spectrum correction | amendment part contained in the 8th Embodiment of this invention. The block diagram which shows the speech presence probability calculation part contained in the 8th Embodiment of this invention. The block diagram which shows the 9th Embodiment of this invention. The block diagram which shows the audio | voice presence probability calculation part contained in the 9th Embodiment of this invention. The block diagram which shows the structure of a related art example. The block diagram which shows the structure of the multiple multiplication part contained in the structure of a related art example. The block diagram which shows the structure of the weighted degradation audio | voice calculation part contained in the structure of a related art example. The block diagram which shows the structure of the SNR calculation part classified by frequency contained in FIG. FIG. 39 is a block diagram illustrating a configuration of a multiple nonlinear processing unit included in FIG. 38. The figure which shows an example of the nonlinear function in a nonlinear processing part. The block diagram which shows the structure of the estimation noise calculation part contained in the structure of a related art example. The block diagram which shows the structure of the estimation noise calculation part classified by frequency contained in FIG. The block diagram which shows the structure of the update determination part contained in FIG. The block diagram which shows the structure of the presumed innate SNR calculation part contained in the structure of a related art example. The block diagram which shows the structure of the multiple value range limitation process part contained in FIG. The block diagram which shows the structure of the addition part with multiple weight contained in FIG. The block diagram which shows the structure of the weighted addition part contained in FIG. The block diagram which shows the structure of the noise suppression coefficient production | generation part contained in the structure of a related art example. The block diagram which shows the structure of the suppression coefficient correction | amendment part contained in the structure of a related art example. The block diagram which shows the structure of the suppression coefficient correction part classified by frequency contained in FIG.

  FIG. 1 is a block diagram showing an embodiment of the present invention. FIG. 1 and FIG. 36 which is a related art example are the same except for the emphasized speech amplitude spectrum correction unit 18. Hereinafter, detailed operations will be described focusing on these differences.

  The enhanced speech amplitude spectrum correction unit 18 includes a degraded speech amplitude spectrum from the Fourier transform unit 3, an estimated noise power spectrum from the estimated noise calculation unit 5, an enhanced speech amplitude spectrum from the multiple multiplication unit 16, and a corrected suppression from the suppression coefficient correction unit 15. Each coefficient is supplied. The enhanced speech amplitude spectrum correction unit 18 corrects the enhanced speech amplitude spectrum using the deteriorated speech amplitude spectrum, the estimated noise power spectrum, the enhanced speech amplitude spectrum, and the correction suppression coefficient, and transmits the corrected speech amplitude spectrum to the inverse Fourier transform unit 9. A detailed description of the configuration and operation of the enhanced speech amplitude spectrum correction unit 18 will be given with reference to FIG.

  FIG. 2 is a block diagram showing the configuration of the enhanced speech amplitude spectrum correction unit 18. The enhanced speech amplitude spectrum correction unit 18 includes multiple multiplication units 170 and 173, a speech presence probability calculation unit 171, and a post-suppression coefficient calculation unit 182. Multiplexer 170 calculates an emphasized speech power spectrum using the enhanced speech amplitude spectrum supplied from multiplex multiplier 16 in FIG. 1, and transmits it to speech presence probability calculator 171. The speech presence probability calculation unit 171 calculates a speech presence probability using the enhanced speech power spectrum and the estimated noise power spectrum supplied from the multiple multiplication unit 170 and the estimated noise calculation unit 5 in FIG. 1, and a post-suppression coefficient calculation unit. 182. Both the enhanced speech power spectrum and the estimated noise power spectrum supplied to the speech presence probability calculation unit are calculated from the degraded speech amplitude spectrum. Therefore, it can be said that the speech existence probability is essentially calculated based on the degraded speech power spectrum.

  The post-suppression coefficient calculator 182 supplies the speech presence probability supplied from the speech presence probability calculator 171, the corrected suppression coefficient supplied from the suppression coefficient corrector 15 in FIG. 1, and the estimated noise calculator 5 in FIG. 1. The post-suppression coefficient is calculated using the estimated noise and the correction suppression coefficient supplied from the suppression coefficient correction unit 15 in FIG. 1 and transmitted to the multiple multiplication unit 173. The multiplex multiplication unit 173 obtains a corrected enhanced speech amplitude spectrum by weighting the degraded speech amplitude spectrum supplied from the Fourier transform unit of FIG. 1 with the post-suppression coefficient supplied from the post-suppression coefficient calculation unit 172. 1 to the inverse Fourier transform unit 9. The configuration of the multiple multipliers 170 and 173 is the same as that of the multiple multiplier 13 described with reference to FIG.

  Detailed configuration and operation of the speech existence probability calculation unit 171 and the post-suppression coefficient calculation unit 182 will be described with reference to FIGS.

  FIG. 3 is a block diagram showing the configuration of the speech existence probability calculation unit 171. The speech existence probability calculation unit 171 includes separation units 1700 and 1708, average value calculation units 1701 and 1709, logarithmic calculation units 1702 and 1710, multiplication units 1703 and 1711, smoothing coefficient storage units 1704 and 1706, and smoothing units 1705 and 1707. , Function value calculation units 1712 and 1713, average index calculation unit 1714, instantaneous index calculation unit 1715, and addition unit 1716.

Separating section 1700 separates the emphasized speech power spectrum supplied from multiplex multiplication section 170 in FIG. Average value calculation section 1701 divides the sum of emphasized speech power spectrum | X _n (k) | ² bars from k = 0 to K−1 by K, and transmits the calculation result to logarithmic calculation section 1702. The logarithm calculation unit 1702 calculates the logarithm of the average value input from the average value calculation unit 1701 and transmits the logarithm to the multiplier 1703. The multiplier 1703 multiplies the supplied logarithm value by a constant to obtain the emphasized speech power PE _n and supplies it to the smoothing units 1705 and 1707. That is, the emphasized speech power PE _n of the nth frame is

Given in.

On the other hand, the separation unit 1708 separates the estimated noise power spectrum supplied from the noise estimation calculation unit 5 of FIG. 1 into frequency-specific estimated noise power spectra, and outputs them to the average value calculation unit 1709. Average value calculation section 1709 divides the total sum of frequency-specific estimated noise power spectrum λ _n (k) from k = 0 to K−1 by K, and transmits the calculation result to logarithmic calculation section 1710. The logarithm calculation unit 1710 calculates the logarithm of the average value supplied from the average value calculation unit 1709 and transmits the logarithm to the multiplier 1711. The multiplier 1711 multiplies the supplied logarithmic value by a constant to obtain an estimated noise power PN _n and supplies it to the function value calculation units 1712 and 1713. That is, the estimated noise power PN _n of the nth frame is

Given in.

Index indicating it contains what extent audio input signal is calculated based on the relative relationship between the enhanced speech power PE _n and the estimated noise power PN _n. If the emphasized speech power PE _n is larger than the estimated noise power PN _n , the index indicates that the presence probability of speech is high. In general, the estimated noise power PN _n and the emphasized speech power PE _n are non-stationary signals, and therefore the estimated noise power PN _n may be larger than the enhanced speech power PE _n in the speech interval. On the contrary, the estimated noise power PN _n may be larger than the emphasized speech power PE _n even in the noise interval. Therefore, if the respective powers are used for index calculation without correction, there is a possibility that an erroneous speech existence probability is obtained. For this reason, in order to improve the accuracy of the speech existence probability calculation, it is desirable to appropriately correct the estimated noise power PN _n and the emphasized speech power PE _n . Moreover, if a plurality of correction methods are introduced and the speech existence probability is calculated based on a plurality of indices, the accuracy is further improved.

In this embodiment, the enhanced speech power PE _n is suitable for index calculation using smoothing processing in the smoothing units 1715 and 1716, and the estimated noise power PN _n is appropriate for function calculation in the function value calculation units 1712 and 1713. The value is corrected. Two types of indicators are calculated: an instantaneous indicator and an average indicator with different analysis section lengths.

The smoothing unit 1705 uses the smoothing coefficient supplied from the smoothing coefficient storage unit 1704 to smooth the emphasized speech power PE _n supplied from the multiplier 1703 in the time direction, and obtains the first smoothed enhanced speech power. It supplies to the instantaneous index calculation part 1715. Similarly, the smoothing unit 1707 uses the smoothing coefficient supplied from the smoothing coefficient storage unit 1706 to smooth the emphasized speech power PE _n supplied from the multiplier 1703 in the time direction, and the second smoothed emphasized speech. The power is supplied to the average index calculation unit 1714. Basically, the coefficient stored in the smoothing coefficient storage unit 1704 is set to be smaller than the coefficient in the smoothing coefficient storage unit 1706. This is because the smaller the value of the smoothing coefficient, the smaller the smoothing effect in the time direction of the smoothing unit, which is suitable for the calculation of the instantaneous index.

The function value calculation unit 1713 calculates the first function value from the estimated noise power PN _n supplied from the multiplier 1711 and supplies the first function value to the instantaneous index calculation unit 1715. Similarly, the function value calculating unit 1712, a second function value calculated from the estimated noise power PN _n supplied from the multiplier 1711, and supplies the average index calculation unit 1714. In calculating the function value, a linear or non-linear function is used to compress or expand the dynamic range, and smoothing is used to reduce dispersion. Dynamic range compression or expansion of, the reduction of the dispersion can be reduced degradation of accuracy index calculation due to unsteadiness of the estimated noise power PN _n. Further, in order to reduce the calculation amount, omitting the function value calculations, it is also possible to use the estimated noise power PN _n as they index calculation. In the function value calculation units 1712 and 1713, for example, the following functions are used.

However, PN _n hat is a function value, and a _fc and b _fc are real numbers.

The instantaneous index calculation unit 1715 calculates an instantaneous index using the first smoothing emphasized speech power supplied from the smoothing unit 1705 and the first function value supplied from the function value calculation unit 1713, and the addition unit 1716. The average index calculation unit 1714 calculates an average index using the second smoothing enhancement speech power supplied from the smoothing unit 1707 and the second function value supplied from the function value calculation unit 1712, and adds the addition unit. 1716. For the calculation of the index, a method is used in which the ratio between the emphasized speech power PE _n and the estimated noise power PN _n is calculated and the numerical value is increased according to the ratio. Specific examples include the following calculation methods.

However, IDX _n is an index, PE _n bar is a smooth emphasis voice power, and PN _n hat is a function value. Θ _idx , a _idx and b _idx are real numbers, and a _idx has a value equal to or greater than b _idx .

If a constant is added to the denominator when calculating the ratio, the value of the denominator does not become smaller than the constant, so that divergence can be prevented when calculating the ratio. In addition to this, the difference between the enhanced speech power PE _n and the estimated noise power PN _n and a value obtained by normalizing the difference with the enhanced speech power PE _n can be calculated, but details are omitted.

  The addition unit 1716 calculates the sum of the average index and the instantaneous index supplied from the average index calculation unit 1714 and the instantaneous index calculation unit 1715, and transmits the sum to the post-suppression coefficient calculation unit 172 of FIG. In addition to the addition, weighted addition or multiplication can be used for calculating the speech existence probability. In order to improve the accuracy of the speech existence probability, three or more types of indices having different analysis intervals may be calculated. It is also possible to simplify the calculation using only one type of index.

  A detailed description of the configuration and operation of the smoothing unit 1705 will be given with reference to FIG.

FIG. 4 is a block diagram showing a configuration of the smoothing unit 1705 of FIG. The smoothing unit 1705 includes a constant multiplier 1741, multipliers 1743 and 1744, adders 1742 and 1745, and a delay unit 1746. The emphasized speech power PE _n is supplied from the multiplier 1703 in FIG. 3 and the smoothing coefficient is supplied from the smoothing coefficient storage unit 1704 in FIG. The smoothing coefficient having the value δ is transmitted to the constant multiplier 1741 and the multiplier 1744. The constant multiplier 1741 multiplies the input signal by −1 to obtain −δ, and transmits this to the adder 1742. 1 is supplied as the other input of the adder 1742, and the output of the adder 1742 is 1−δ which is the sum of both. 1-δ is supplied to a multiplier 1743 and multiplied by the emphasized speech power PE _n which is the other input, and (1-δ) PE _n which is a product is transmitted to the adder 1745.

On the other hand, the multiplier 1744 multiplies δ supplied as a smoothing coefficient by the smoothing-enhanced speech power PE _n−1 bar one frame before supplied from the delay unit 1746, and a product δPE _n−1 bar is obtained. It is transmitted to the adder 1745. The adder 1745, to (1-δ) instantaneous index calculator 1715 PE _n and .DELTA.Pe _n-1 bar sum of the delay device 1746 and FIG. 3, as the smoothing enhanced speech power PE _n bars, and outputs. The above calculation can be expressed by equation (19).

  The configuration of the smoothing unit 1707 is the same as that of the smoothing unit 1705. However, the smoothing unit 1707 uses the smoothing coefficient supplied from the smoothing coefficient storage unit 1706 to calculate the smoothed enhanced speech power. Further, the smoothing units 1705 and 1707 can use a moving average in addition to the equation (19).

FIG. 5 is a block diagram showing the configuration of the post-suppression coefficient calculator 182 in FIG. The post-suppression coefficient calculator 182 includes a separator 1722, post-frequency post-suppression coefficients calculators 1821 _{0 to} 1821 _K−1 , and a multiplexer 1723. The separation unit 1722 separates the corrected suppression coefficient supplied from the suppression coefficient correction unit 5 of FIG. 1 into frequency-specific correction suppression coefficients and transmits the frequency-specific post-frequency suppression coefficient calculation units 1821 _{0 to} 1821 _K−1 . The post-frequency suppression coefficient calculation units 1821 _{0 to} 1821 _K−1 are the speech existence probability and the emphasized speech respectively supplied from the speech existence probability calculation unit 171 in FIG. 2, the multiple multiplication unit 16 in FIG. 1, and the estimated noise calculation unit 5. The frequency-specific post-frequency suppression coefficient is calculated using the amplitude spectrum, the estimated noise power spectrum, and the frequency-specific correction suppression coefficient supplied from the separation unit 1722 and transmitted to the multiplexing unit 1723.

A detailed description of the configuration and operation of the frequency-specific post-frequency suppression coefficient calculation units 1821 _{0 to} 1821 _K−1 will be given with reference to FIG.

FIG. 6 is a block diagram illustrating a configuration of the frequency-specific post-frequency suppression coefficient calculation units 1821 _{0 to} 1821 _K−1 in FIG. 5. The post-frequency suppression coefficient calculation unit 1821 includes a sound part coefficient storage unit 1831, a silence part coefficient calculation unit 1832, a coefficient calculation unit 1833, and a multiplier 1834. The frequency-specific post-frequency suppression coefficient calculation unit 1821 calculates the frequency-specific post-frequency suppression coefficient according to the speech existence probability. If the speech existence probability is low, the value of the post-frequency suppression coefficient is reduced using a coefficient with a high contribution ratio of the silence part coefficient. For this reason, the residual noise in the noise section can be further reduced. On the other hand, when the speech existence probability is high, correction is performed so that the post-frequency suppression coefficient becomes equal to the frequency-specific correction suppression coefficient using a coefficient with a high contribution ratio of the sound part coefficient. Further, correction may be performed so that the post-frequency suppression coefficient is slightly larger than the frequency-specific correction suppression coefficient. From the above, when the voice existence probability is high, excessive suppression of voice can be prevented. In this embodiment, the coefficient is calculated for each frequency. However, if a common coefficient is obtained in all bands and the coefficient is applied to the correction suppression coefficient for each frequency, the amount of calculation required for calculating the coefficient can be reduced. .

  The coefficient calculation unit 1833 supplies the sound part coefficient and the soundless part coefficient output from the sound part coefficient storage part 1831 and the soundless part coefficient calculation part 1832 and the sound existence probability calculation part 171 in FIG. The coefficient is calculated based on the voice presence probability.

When the speech existence probability is p, the sound part coefficient is FV, and the silence part coefficient is FU, the coefficient F output from the coefficient calculation unit 1833 is given by Expression (20).

  In the coefficient calculation, if the speech existence probability is large, the contribution ratio of the sound part coefficient to the output value of the coefficient calculation unit 1833 is increased. In the calculation method of Expression (20), the speech existence probability is directly used as a contribution rate.

Further, as shown in the equation (21), the sound existence probability can be used as the contribution rate after correcting the coefficients for the sounded part and the silent part using appropriate functions F _SFC and G _SFC. It is.

In addition to this, when the voice existence probability is equal to or higher than a predetermined value, the coefficient for sound part can be output from the coefficient calculation unit 1833. Then, the multiplier 1834 calculates the product of the frequency-specific correction suppression coefficient supplied from the separation unit 1722 in FIG. 5 and the coefficient supplied from the coefficient calculation unit 1833, and uses the frequency-dependent post-frequency suppression coefficient in FIG. Transmitted to the unit 1723.

  The silent part coefficient calculation unit 1832 includes the speech existence probability, the enhanced speech amplitude spectrum, and the estimated noise power spectrum respectively supplied from the speech existence probability calculation unit 171 in FIG. 2, the multiple multiplication unit 16 in FIG. 1, and the estimated noise calculation unit 1. Is used to obtain the coefficient for silent part and supply it to the coefficient calculation part 1833. In order to reduce residual noise in the noise section, the silent part coefficient calculation unit 1832 is designed so as to output a value smaller than the sound part coefficient.

  A detailed description of the configuration and operation of the silent part coefficient calculation unit 1832 will be given with reference to FIG.

  FIG. 7 is a block diagram showing the configuration of the silent part coefficient calculation unit 1832 of FIG. The silent part coefficient calculation unit 1832 includes separation units 1850 and 1855, average value calculation units 1851 and 1856, audio power mixing unit 1852, smoothing unit 1853, smoothing coefficient storage unit 1854, smooth signal storage unit 1858, and division unit 1857. 1862, a logarithm calculation unit 1859, a constant multiplication unit 1860, a coefficient calculation unit 1861, and an exponent calculation unit 1863.

Separating section 1850 separates the emphasized speech power spectrum supplied from multiplex multiplication section 16 in FIG. The average value calculation unit 1851 divides the sum of the frequency-specific enhanced speech power spectrum | X _n (k) | ² bars from k = 0 to K−1 by K, and transmits the sum to the speech power mixing unit 1852 as the enhanced speech power. . The speech power mixing unit 1852 uses the enhanced speech power supplied from the average value calculation unit 1851 and the smoothed enhanced speech power one frame before supplied from the smoothed signal storage unit 1858 from the speech existence probability calculation unit 171 in FIG. Mixing is performed according to the supplied speech existence probability, and the mixed signal is transmitted to the smoothing unit 1853. At the time of mixing, if the speech existence probability is high, the ratio of the emphasized speech average power supplied from the average power calculator 1851 is increased, and if it is low, the ratio of the smoothed emphasized speech power supplied from the smooth signal storage unit 1858 is increased. To do.

  The smoothing unit 1853 smoothes the mixed signal supplied from the audio power mixing unit 1852 in accordance with the smoothing coefficient supplied from the smoothing coefficient storage unit 1854, and divides the smoothed signal from the smoothed signal storage unit 1858 as the smoothed enhanced audio power. Part 1857. As is clear from the function of the speech power mixing unit, in a section where the speech existence probability is low, the smoothing unit 1853 calculates the smoothed speech power using a signal containing a large amount of the smoothed speech power one frame before. To do. Therefore, the smooth enhancement speech power is hardly updated. For this reason, the smoothing unit 1853 always outputs the enhanced speech power calculated in the speech interval even in the noise interval. On the other hand, in a section where the speech existence probability is high, the smoothing unit 1853 calculates smooth enhanced speech power using a signal that includes a large amount of enhanced speech average power.

  The voice presence probability used in the voice power mixing unit 1852 is supplied from the voice presence probability calculation unit 171 in FIG. 2, and is calculated based on the enhanced voice power spectrum and the estimated noise power spectrum. Since the emphasized speech power spectrum and the estimated noise power spectrum are also input to the silence part coefficient calculation unit 1832, the speech existence probability used in the speech power mixing unit 1852 is also calculated inside the silence part coefficient calculation unit 1832. It is possible.

  Further, as in the case of the speech existence probability calculation unit 171 in FIG. 2, the enhanced speech power spectrum and the estimated noise power spectrum are calculated based on the degraded speech amplitude spectrum, and thus are used in the speech power mixing unit 1852. It can be said that the existing speech existence probability is essentially obtained from the degraded speech amplitude spectrum.

On the other hand, the separation unit 1855 separates the estimated noise power spectrum supplied from the estimated noise calculation unit 5 of FIG. 1 into the estimated noise power spectrum for each frequency, and outputs it to the average value calculation unit 1856. Average value calculating section 1856 divides the sum of k = 1 to K−1 of estimated frequency noise power spectrum λ _n (k) by frequency by K, and transmits the calculation result to dividing section 1857 as estimated noise average power. The division unit 1857 divides the emphasized speech average power supplied from the smoothing unit 1853 by the estimated noise average power supplied from the average value calculation unit 1856 and transmits the division result to the logarithmic calculation unit 1859. The logarithm calculation unit 1859 calculates the logarithm of the division result supplied from the division unit 1857 and transmits the logarithmic value to the constant multiplication unit 1860.

  The constant multiplier 1860 multiplies the logarithmic value supplied from the logarithm calculator 1859 by a constant, and transmits the calculation result to the coefficient calculator 1861. The coefficient calculation unit 1861 obtains a coefficient from the output of the constant multiplication unit 1860 and transmits it to the division unit 1862. Since 10 is supplied as the other input of the division unit 1862, the division unit 1862 divides the coefficient supplied from the coefficient calculation unit 1861 by 10 and transmits the division result to the exponent calculation unit 1863. The exponent part calculation unit 1863 calculates the exponent of the output of the division unit 1862 and transmits the calculation result to the coefficient calculation unit 1833 in FIG. 6 as a silence part coefficient.

  The calculation result of the division unit 1857 corresponds to the ratio between the emphasized speech average power and the estimated noise power, that is, the SNR. Accordingly, the coefficient calculation unit 1861 calculates the suppression degree of the silent part based on the SNR. The purpose of calculating the SNR is to reflect the reliability of the speech existence probability obtained by the speech existence probability calculation unit 171 in the calculation of the coefficient. When the SNR is high, that is, when the reliability of the voice existence probability is high, since the possibility of erroneously suppressing the voice is small, the coefficient is reduced and the degree of suppression is increased. On the other hand, when the reliability of the voice existence probability is low, the coefficient is increased to reduce the degree of suppression in order to prevent the voice from being erroneously suppressed. Since it is important to obtain the coefficient from the SNR, either one or both of the logarithmic calculator 1859 and the exponent calculator 1863 can be omitted to simplify the calculation.

  Further, if division is performed after adding a constant set appropriately in advance to the estimated noise average power, divergence of the division result can be prevented. Divergence can also be prevented by using an approximate operation of division instead of division.

  In this embodiment, when calculating the emphasized speech average power and the estimated noise power, the average value of the power spectrum of the entire band is used. However, the average value of the power spectrum calculated for each subband having an appropriate bandwidth. The method using is also effective. Since the average value is calculated for each band, it is possible to calculate an accurate SNR in each band, compared to the case where the average value of all bands is used.

FIG. 8 shows an example of a nonlinear function used when the coefficient calculation unit 1861 in FIG. 7 calculates the coefficient. When f _cm is an input value, the output value g _cm of the nonlinear function shown in FIG. 8 is given by Expression (22).

However, a _cm , b _cm , c _cm , and d _cm are positive real numbers.

The condition required for the nonlinear function of Equation (22) is that g _cm decreases as f _cm increases. In addition to Expression (22), a linear function that satisfies this condition, a high-order polynomial, and an arbitrary function including weighted addition can be used.

  FIG. 9 is a block diagram showing a second embodiment of the present invention. FIG. 9 and FIG. 1 as the first embodiment are the same except for the enhanced speech amplitude spectrum correction unit 28. A detailed description of the configuration and operation of the enhanced speech amplitude spectrum correction unit 28 will be given with reference to FIG.

  FIG. 10 is a block diagram showing the configuration of the enhanced speech amplitude spectrum correction unit 28. The emphasized speech amplitude spectrum correction unit 18 shown in FIG. 2 is the same except that the post-suppression coefficient calculator 182 is replaced with a post-suppression coefficient calculator 282. Detailed configuration and operation of the post-suppression coefficient calculator 282 will be described with reference to FIG.

FIG. 11 is a block diagram showing the configuration of the post-suppression coefficient calculator 282. The suppression coefficient calculation unit 182 then illustrated in Figure 5, that after the suppression coefficient calculation unit 1821 ₀ ~1821 _K-1 each frequency is replaced after each frequency suppression coefficient calculation unit 2821 ₀ ~2821 _K-1 Except for the same. A detailed description of the configuration and operation of the frequency-specific post-frequency suppression coefficient calculation units 2821 _{0 to} 2821 _K−1 will be given with reference to FIG.

FIG. 12 is a block diagram illustrating the configuration of the frequency-specific post-frequency suppression coefficient calculation units 2821 _{0 to} 2821 _K−1 of FIG. 11. The post-frequency suppression coefficient calculation unit 1821 shown in FIG. 6 is the same except that the sound part coefficient storage unit 1831 is replaced with a sound part coefficient calculation unit 2831. Since not only the silence part coefficient but also the sound part coefficient is calculated, higher sound quality can be achieved in the sound part than the frequency-specific post-frequency suppression coefficient calculation part of FIG.

  The sound part coefficient calculation unit 2831 obtains a sound part coefficient by using the emphasized speech power spectrum and the estimated noise power spectrum respectively supplied from the multiple multiplication unit 16 and the estimated noise calculation unit 5 of FIG. The data is supplied to the calculation unit 1833. When the estimated noise power is greater than the emphasized speech power, or when both powers are equal in magnitude, the sound part coefficient calculation unit 2831 generates 1.0 according to the power ratio between the estimated noise and the enhanced speech. The above values are output. This is performed in order to prevent over-suppression in the speech section because the correction suppression coefficient may be smaller than an appropriate value. On the other hand, when the estimated noise is smaller than the emphasized speech, the possibility of excessive suppression occurring in the speech section is low. Therefore, an appropriate constant value of 1.0 or more is output regardless of the estimated noise and the power ratio of the emphasized speech.

  FIG. 13 is a block diagram showing a third embodiment of the present invention. FIG. 13 and FIG. 1 as the first embodiment are the same except for the emphasized speech amplitude spectrum correction unit 17. As will be described later, the difference between the enhanced speech amplitude spectrum correction units 17 and 18 is that the enhanced speech amplitude spectrum correction unit 17 does not use the estimated noise power spectrum and the enhanced speech power spectrum when calculating the post-suppression coefficient. . A detailed description of the configuration and operation of the enhanced speech amplitude spectrum correction unit 17 will be given with reference to FIG.

  FIG. 14 is a block diagram showing the configuration of the enhanced speech amplitude spectrum correction unit 17. The enhanced speech amplitude spectrum correction unit 18 shown in FIG. 2 is the same as that except that the post-suppression coefficient calculator 182 is replaced with the post-suppression coefficient calculator 172. Hereinafter, detailed operations will be described focusing on this difference.

  The post-suppression coefficient calculator 172 calculates a post-suppression coefficient using the speech presence probability supplied from the speech presence probability calculator 171 and the corrected suppression coefficient supplied from the suppression coefficient corrector 15 in FIG. This is transmitted to the multiplier 173. Detailed configuration and operation of the post-suppression coefficient calculator 172 will be described with reference to FIG.

FIG. 15 is a block diagram illustrating a configuration of the post-suppression coefficient calculation unit 172. The suppression coefficient calculation unit 182 then illustrated in Figure 5, that after the suppression coefficient calculation unit 1821 ₀ ~1821 _K-1 each frequency is replaced after each frequency suppression coefficient calculation unit 1721 ₀ ~1721 _K-1 Except for the same. Hereinafter, detailed operations will be described focusing on this difference.

The frequency-specific post-frequency suppression coefficient calculation units 1721 _{0 to} 1721 _K−1 use the frequency-specific corrected suppression coefficient supplied from the separation unit 1722 and the audio presence probability supplied from the audio existence probability calculation unit 171 in FIG. The frequency-specific post-suppression coefficient is calculated and transmitted to the multiplexing unit 1723. A detailed description of the configuration and operation of the frequency-specific post-frequency suppression coefficient calculation units 1721 _{0 to} 1721 _K−1 will be given with reference to FIG.

FIG. 16 is a block diagram illustrating a configuration of the frequency-specific post-frequency suppression coefficient calculation units 1721 _{0 to} 1721 _K−1 in FIG. 15. The post-frequency suppression coefficient calculation unit 1721 includes a lower limit value storage unit 1691 for a sound part, a lower limit value storage unit 1692 for a silence part, a lower limit value calculation unit 1693, and a maximum value selection unit 1694. The lower limit calculation unit 1693 is based on the lower limit value for the sound part supplied from the lower limit value storage unit 1691 for the sound part and the lower limit value for the silent part supplied from the lower limit value storage part 1692 for the silence part. A lower limit value corresponding to the voice presence probability supplied from the voice presence probability calculation unit 171 in FIG. 14 is calculated and transmitted to the maximum value selection unit 1694. In order to prevent sound distortion, a value larger than the lower limit value for the silent part is set as the lower limit value for the sound part. In the calculation of the lower limit value, if the speech existence probability is large, the contribution ratio of the lower limit value for the sound part to the output value of the lower limit value calculation unit 1693 is increased. For setting the contribution rate, it is possible to similarly use the methods shown in the equations (20) and (21).

  The maximum value selection unit 1694 compares the frequency-specific correction suppression coefficient supplied from the separation unit 1722 in FIG. 15 with the lower limit value supplied from the lower limit value calculation unit 1693, and multiplexes the larger value in FIG. Part 1723. Considering the case where the values are the same, the post-suppression coefficient becomes a value equal to or higher than the lower limit value supplied by the lower limit value calculation unit 1693. Therefore, the suppression coefficient becomes a value equal to or higher than the lower limit value set according to the voice presence probability. If the speech existence probability is high, the lower limit value becomes large, so that speech distortion caused by excessive suppression in the speech section can be prevented. On the other hand, if the speech existence probability is low, the lower limit value becomes small, so that a sufficient degree of suppression can be obtained in the noise interval.

  FIG. 17 is a block diagram showing a fourth embodiment of the present invention. FIG. 17 and FIG. 1 as the first embodiment are the same except for the enhanced speech amplitude spectrum correction unit 29. A detailed description of the configuration and operation of the enhanced speech amplitude spectrum correction unit 29 will be given with reference to FIG.

  FIG. 18 is a block diagram showing the configuration of the enhanced speech amplitude spectrum correction unit 29. The enhanced speech amplitude spectrum correction unit 18 shown in FIG. 2 is the same except that the post-suppression coefficient calculator 182 is replaced with a post-suppression coefficient calculator 292. A detailed description of the configuration and operation of the post-suppression coefficient calculator 292 will be given with reference to FIG.

FIG. 19 is a block diagram showing the configuration of the post-suppression coefficient calculator 292. The suppression coefficient calculation unit 182 then illustrated in Figure 5, that after the suppression coefficient calculation unit 1821 ₀ ~1821 _K-1 each frequency is replaced after each frequency suppression coefficient calculation unit 2921 ₀ ~2921 _K-1 Except for the same. Detailed configuration and operation of the frequency-specific post-frequency suppression coefficient calculation units 2921 _{0 to} 2921 _K−1 will be described with reference to FIG.

FIG. 20 is a block diagram illustrating a configuration of frequency-specific post-frequency suppression coefficient calculation units 2921 _{0 to} 2921 _K−1 in FIG. The frequency-specific post-frequency suppression coefficient calculation unit 1721 shown in FIG. 16 is that the sound part lower limit value storage unit 1691 is replaced with the sound part lower limit value calculation unit 2691, and the soundless part lower limit value storage unit 1692. Are the same except for the silent part lower limit calculator 2692. Since the lower limit value for the voiced part and the silent part is calculated based on the emphasized voice power spectrum and the estimated noise power spectrum, the residual noise in the silent part is Audio distortion can be reduced at the part.

  The voiced lower limit value calculation unit 2691 and the silent part lower limit value calculation unit 2692 use the emphasized speech power spectrum and the estimated noise power spectrum respectively supplied from the multiple multiplication unit 16 and the estimated noise calculation unit 1 in FIG. Then, the lower limit value for the sound part and the lower limit value for the silent part are respectively obtained and supplied to the lower limit value calculation unit 1693. The lower limit value calculation unit for sounded part 2691 and the lower limit value calculation unit for silent part 2692 calculate the respective lower limit values according to the power ratio of the estimated noise and the emphasized speech, and transmit them to the lower limit value calculation unit 1693. Basically, if the estimated noise power is larger than the emphasized voice power, that is, the SNR is lowered, the lower limit value for the sound part is increased for the purpose of preventing voice distortion.

  In order to reduce the amount of residual noise in the silent part and prevent excessive suppression in the voiced part, the lower limit value for the silent part is set to a value equal to or lower than the lower limit value for the voiced part. However, when the SNR is low, control is performed so that the difference between the lower limit value for the sounded part and the lower limit value for the silent part is not increased. If the difference between the lower limit values is too large, the difference in the amount of residual noise between the sound part and the soundless part becomes large, and as a result, it is perceived that sound distortion occurs in the sound section. On the other hand, if the SNR is high, the residual noise in the sound part is masked by the sound component and becomes difficult to perceive. Therefore, as in the case where the SNR is low, the difference in the amount of residual noise between the voiced part and the silent part hardly causes a voice distortion factor in the voice section.

  Therefore, when the SNR is high, the difference between the lower limit value for the silent part and the lower limit value for the voiced part is increased to sufficiently reduce the residual noise in the silent part. From the above, the silent part lower limit value is set to a value depending on the voiced part lower limit value. Therefore, basically, as in the case of the lower limit value of the sound part, the lower limit value for the silent part is increased as the SNR decreases. When comparing the magnitudes of the estimated noise power spectrum and the emphasized speech power spectrum, it is preferable to use the respective average values or the output signal of the division unit 1857 used in the silence part coefficient calculation of FIG.

  FIG. 21 is a block diagram showing a fifth embodiment of the present invention. FIG. 21 and FIG. 36 which is a block diagram of the related art example, the estimated innate SNR calculation unit 7 and the suppression coefficient correction unit 15 are replaced with the estimated innate SNR calculation unit 71 and the suppression coefficient correction unit 19, respectively. It is the same except that. Hereinafter, detailed operations will be described focusing on these differences.

  The estimated innate SNR calculator 71 includes a degraded speech power spectrum from the multiple multiplier 13, an estimated noise power spectrum from the estimated noise calculator 5, an acquired SNR from the frequency-specific SNR calculator 6, and correction suppression from the suppression coefficient corrector 19. A coefficient is supplied. The estimated innate SNR calculation unit 71 obtains the estimated innate SNR and the voice presence probability using the degraded voice power spectrum, the estimated noise power spectrum, the acquired SNR, and the correction suppression coefficient. Then, the speech existence probability is transmitted to the suppression coefficient correction unit 19, and the estimated innate SNR is transmitted to the noise suppression coefficient generation unit 8 and the suppression coefficient correction unit 19. The suppression coefficient correction unit 19 corrects the suppression coefficient supplied from the noise suppression coefficient generation unit 8 using the estimated innate SNR and the speech existence probability supplied from the estimated innate SNR calculation unit 71, and serves as a corrected suppression coefficient. This is transmitted to the multiple multiplier 16 and the estimated innate SNR calculator 71.

  A detailed description of the configuration and operation of the suppression coefficient correction unit 19 and the estimated innate SNR calculation unit 71 will be given with reference to FIGS.

  FIG. 22 is a block diagram showing the configuration of the estimated innate SNR calculation unit 71. The difference between FIG. 22 and FIG. 45 which is a block diagram of the related art example is that the estimated innate SNR calculation unit 71 includes delay units 711 and 712, multiple multiplication unit 713, and speech existence probability calculation unit 714. It is. Hereinafter, detailed operations will be described focusing on these differences.

The delay unit 712 stores the estimated noise power spectrum λ _n (k) of the nth frame supplied from the estimated noise calculation unit 5 of FIG. 21 and at the same time stores the estimated noise power spectrum of the n−1th frame. λ _n−1 (k) is supplied to the speech existence probability calculation unit 714. The delay unit 711 stores the degraded speech power spectrum | Y _n (k) | ² of the nth frame supplied from the multiplex multiplier 13 of FIG. The power spectrum | Y _n−1 (k) | ² is supplied to the multiple multiplier 713. The multiple multiplier 713 converts G ² _n−1 (k) bar supplied from the multiple multiplier 704 and | Y _n−1 (k) | ² supplied from the delay unit 711 into k = 0, 1,. . . , K−1 to obtain G ² _n−1 (k) bar | Y _n−1 (k) | ² , and the calculation result is transmitted to the speech existence probability calculation unit 714 as the estimated enhanced speech power spectrum. To do. The output signal of the multiplex multiplier 713 matches the emphasized speech power spectrum of the (n-1) th frame. In order to treat this as an estimated signal of the enhanced speech power spectrum of the nth frame, the name "estimated enhanced speech power spectrum" is used. Used.

  Since the suppression coefficient supplied from the multiplex multiplier 704 was obtained one frame before, a delay unit 711 was introduced to calculate the enhanced speech power spectrum by combining the suppression coefficient and the frame number of the degraded speech power spectrum. Has been. Further, a delay unit 712 is introduced to match the frame numbers of the emphasized speech power spectrum and the estimated noise power spectrum used for calculating the speech presence probability. However, since the influence of the difference of several frames on the calculation of the speech existence probability is small, it is possible to omit one or both of the delay units 711 and 712.

  The speech existence probability calculation unit 714 calculates the speech presence probability using the estimated emphasized speech power spectrum supplied from the multiple multiplier 713 and the estimated noise power spectrum supplied from the delay unit 712, and corrects the suppression coefficient in FIG. To the unit 19. Since the configuration of the multiple multiplier 713 is the same as that of the multiple multiplier 21 already described with reference to FIG. The configuration of the speech presence probability calculation unit 714 is the same as the speech presence probability calculation unit 171 described with reference to FIG.

FIG. 23 is a block diagram showing a configuration of the suppression coefficient correction unit 19 of FIG. The suppression coefficient correction unit 15 shown in FIG. 50, the same except that the frequency-suppression coefficient correction unit 1501 ₀ ~1501 _K-1 has been replaced with frequency-suppression coefficient correction unit 1901 ₀ ~1901 _K-1 It is. Hereinafter, detailed operations will be described focusing on these differences.

The frequency-specific suppression coefficient correction units 1901 _{0 to} 1901 _K−1 use the frequency-specific estimated innate SNR supplied from the separating unit 1502 and the speech existence probability supplied from the estimated innate SNR calculation unit 71 in FIG. Then, the frequency-specific suppression coefficient supplied from the separation unit 1503 is corrected and transmitted to the multiplexing unit 1504 as the frequency-specific correction suppression coefficient. A detailed description of the configuration and operation of the frequency-specific suppression coefficient correction units 1901 _{0 to} 1901 _K−1 will be given with reference to FIG.

FIG. 24 is a block diagram illustrating the configuration of the frequency-specific suppression coefficient correction units 1901 _{0 to} 1901 _K−1 in FIG. In FIG. 24, instead of the maximum value selection unit 1591 and the suppression coefficient lower limit value storage unit 1592 in the frequency-specific suppression coefficient correction unit 1501 of FIG. 51, a voiced part lower limit value storage unit 1921 and a silent part lower limit value storage unit 1922 are used. , A lower limit calculator 1923 and a maximum value selector 1924 are provided. Hereinafter, detailed operations will be described focusing on these differences.

  The lower limit value calculation unit 1923 is based on the lower limit value for a sound part supplied from the lower limit value storage part 1921 for a sound part and the lower limit value for a soundless part supplied from the lower limit value storage part 1922 for a silence part. A lower limit value corresponding to the speech existence probability supplied from the estimated innate SNR calculation unit 71 in FIG. 21 is calculated and transmitted to the maximum value selection unit 1924. Maximum value selection section 1924 compares the output value of switch 1595 or multiplier 1597 with the lower limit value supplied from lower limit calculation section 1923, and uses the larger value as the correction suppression coefficient, and multiplexing section 1504 in FIG. To communicate. Considering the case where the values are the same, the corrected suppression coefficient is a value greater than the lower limit value supplied by the lower limit value calculation unit 1923. Therefore, since the suppression coefficient becomes a value equal to or higher than the lower limit value set according to the voice presence probability, voice distortion caused by excessive suppression in the voice section can be prevented. Since the configuration of the lower limit calculation unit 1923 is equal to the lower limit calculation unit 1693 already described with reference to FIG. 6, detailed description thereof is omitted.

  FIG. 25 is a block diagram showing a sixth embodiment of the present invention. FIG. 25 and FIG. 36, which is a block diagram of the related art example, show that the estimated innate SNR calculation unit 7 and the suppression coefficient correction unit 15 are replaced with the estimated innate SNR calculation unit 72 and the suppression coefficient correction unit 20, respectively. Is the same except for. Hereinafter, detailed operations will be described focusing on these differences.

  The estimated innate SNR calculation unit 72 includes a degenerate speech power spectrum from the multiple multiplication unit 13, an estimated noise power spectrum from the estimation noise calculation unit 5, an acquired SNR from the frequency-specific SNR calculation unit 6, and correction suppression from the suppression coefficient correction unit 20. A coefficient is supplied. The estimated innate SNR calculator 72 obtains an estimated innate SNR, an audio presence probability, and an estimated enhanced audio power spectrum using the degraded audio power spectrum, the estimated noise power spectrum, the acquired SNR, and the correction suppression coefficient. Then, the estimated innate SNR, the speech existence probability, and the estimated enhanced speech power spectrum are transmitted to the suppression coefficient correction unit 20, and the estimated innate SNR is transmitted to the noise suppression coefficient generation unit 8, respectively. The suppression coefficient correction unit 20 corrects the suppression coefficient supplied from the noise suppression coefficient generation unit 8 using the estimated innate SNR, the speech existence probability, and the estimated enhanced speech power spectrum supplied from the estimated innate SNR calculation unit 72. The corrected suppression coefficient is transmitted to the multiple multiplier 16 and the estimated innate SNR calculator 72. A detailed description of the configuration and operation of the estimated innate SNR calculator 72 and the suppression coefficient correction corrector 20 will be given with reference to FIGS.

  FIG. 26 is a block diagram showing a configuration of the estimated innate SNR calculation unit 72. The estimated innate SNR calculation unit 71 of FIG. 22 is the same except that the multiple multiplier 713 is replaced with a multiple multiplier 715. The multiple multiplier 713 supplies the estimated enhanced speech power spectrum only to the speech presence probability calculator 714, but the multiple multiplier 715 also supplies the suppression coefficient corrector 20 of FIG. The configuration of the multiple multiplier 715 is the same as the multiple multiplier 713 already described with reference to FIG.

FIG. 27 is a block diagram illustrating a configuration of the suppression coefficient correction unit 20. The suppression coefficient correction unit 15 of FIG. 50, are identical except that the frequency-suppression coefficient correction unit 1501 ₀ ~1501 _K-1 has been replaced with frequency-suppression coefficient correction unit 2001 ₀ ~2001 _K-1 . Hereinafter, detailed operations will be described focusing on these differences.

The frequency-specific suppression coefficient correction units 2001 _{0 to} 2001 _K−1 include a frequency-specific estimated innate SNR from the separation unit 1502, an estimated noise power spectrum from the estimated noise calculation unit 5 in FIG. 25, and an estimated innate SNR calculation unit in FIG. 25. The speech existence probability and the estimated enhanced speech power spectrum are supplied from 72. The frequency-specific suppression coefficient supplied from the separation unit 1503 is corrected using the frequency-specific estimated innate SNR, the estimated noise power spectrum, the estimated emphasized voice power spectrum, and the voice presence probability, and is multiplexed as a frequency-specific corrected suppression coefficient. To communicate. A detailed description of the configuration and operation of the frequency-specific suppression coefficient correction units 2001 _{0 to} 2001 _K−1 will be given with reference to FIG.

FIG. 28 is a block diagram illustrating a configuration of the frequency-specific suppression coefficient correction units 2001 _{0 to} 2001 _K−1 in FIG. 28, instead of the maximum value selection unit 1591 and the suppression coefficient lower limit value storage unit 1592 in the frequency-specific suppression coefficient correction unit 1501 of FIG. 51, a sound part correction coefficient storage unit 2011 and a silent part correction coefficient storage unit 2012 are used. , A correction coefficient calculation unit 2013, and a multiplier 2014 are provided. Hereinafter, detailed operations will be described focusing on these differences.

  The silent part correction coefficient calculation unit 2012 includes the speech existence probability and the estimated enhanced speech power spectrum supplied from the estimated innate SNR calculation unit 72 in FIG. 25, and the estimated noise power supplied from the estimated noise calculation unit 5 in FIG. The silent part correction coefficient is calculated using the spectrum and supplied to the correction coefficient calculation part 2013. The correction coefficient calculation unit 2013 is based on the sound part correction coefficient supplied from the sound part correction coefficient storage unit 2011 and the soundless part correction coefficient supplied from the sound part correction coefficient calculation unit 2012. A correction coefficient corresponding to the speech existence probability supplied from the estimated innate SNR calculation unit 72 in FIG. 25 is calculated and transmitted to the multiplier 2014. The multiplier 2014 calculates the product of the correction coefficient supplied from the correction coefficient calculation unit 2013 and the output value of the switch 1595 or the multiplier 1597, and transmits the product as a correction suppression coefficient to the multiplexing unit 1504 in FIG. Since the suppression coefficient is corrected by the correction coefficient calculated according to the speech existence probability, the residual noise can be further suppressed in the noise section. The configuration of the silent part correction coefficient calculation unit 2012 is the same as the silent part correction coefficient calculation unit 1832 already described with reference to FIG. The configuration of the correction coefficient calculation unit 2013 is the same as that of the correction coefficient calculation unit 1833 already described with reference to FIG.

  FIG. 29 is a block diagram showing a seventh embodiment of the present invention. The difference between FIG. 29 and FIG. 13 which is the third embodiment is that a delay unit 23 and an adder 24 are provided in place of the speech non-existence probability storage unit 21, and the enhanced speech amplitude spectrum correction unit 17. Is replaced by the enhanced speech amplitude spectrum correction unit 22. Hereinafter, detailed operations will be described focusing on these differences.

  The speech existence probability output from the enhanced speech amplitude spectrum correction unit 22 is stored in the delay unit 23. The delay unit 23 transmits the voice presence probability of the previous frame to the adder 24. Since the speech existence probability is calculated after the noise suppression coefficient is generated, the speech presence probability of one frame before is used to calculate the speech presence probability necessary for generating the noise suppression coefficient. The adder 24 calculates a value obtained by subtracting the voice presence probability from 1, and transmits the calculation result as a voice non-existence probability to the noise suppression coefficient generation unit. In the third embodiment of FIG. 13, the noise suppression coefficient is always generated using the same speech non-existence probability, but in this embodiment, the speech non-existence is calculated based on the speech existence probability calculated by the enhanced speech amplitude spectrum correction unit. Existence probability is calculated. For this reason, it is possible to use a speech non-existence probability that is more suitable for each input signal than the related art for generating a noise suppression coefficient. A detailed description of the configuration and operation of the enhanced speech amplitude spectrum correction unit 22 will be given with reference to FIG.

  FIG. 30 is a block diagram illustrating a configuration of the enhanced speech amplitude spectrum correction unit 22 of FIG. 14 is the same as the emphasized speech amplitude spectrum correction unit 17 except that the speech presence probability calculation unit 171 is replaced with a speech presence probability calculation unit 221. The speech existence probability calculation unit 171 in FIG. 14 transmits the speech existence probability only to the post-suppression coefficient 172, but the speech existence probability calculation unit 221 in FIG. 30 further transmits the speech existence probability to the delay unit 23 in FIG. Yes.

  FIG. 31 is a block diagram showing an eighth embodiment of the present invention. The difference between FIG. 31 and FIG. 29 which is the seventh embodiment is that a speech existence probability calculation unit 26 is provided in place of the delay unit 23, and the enhanced speech amplitude spectrum correction unit 22 is enhanced speech spectrum. That is, the correction unit 25 is replaced. The speech existence probability calculation unit 26 calculates the speech presence probability using the estimated innate SNR output from the estimated innate SNR calculation unit 7 and transmits the calculated speech existence probability to the adder 24 and the enhanced speech amplitude spectrum correction unit 25. Unlike FIG. 29 which is the seventh embodiment, the noise suppression coefficient generation unit 8 derives based on the voice existence probability calculated one frame before in order to calculate the voice existence probability before generating the noise suppression coefficient. There is no need to use the determined speech non-existence probability. For this reason, the noise suppression coefficient generation unit 8 of the present embodiment can use a more accurate speech non-existence probability than the case of the seventh embodiment. A detailed description of the configuration and operation of the enhanced speech amplitude spectrum correction unit 25 and the speech presence probability calculation unit 26 will be given with reference to FIGS. 32 and 33.

  FIG. 32 is a block diagram showing a configuration of the enhanced speech amplitude spectrum correction unit 25 of FIG. The emphasized speech amplitude spectrum correction unit 22 in FIG. 30 indicates that the speech existence probability calculation unit 221 and the multiple multiplication unit 170 are deleted, and that the post-suppression coefficient calculation unit 172 is replaced with the post-suppression coefficient 252. Except for the same. The post-suppression coefficient calculation unit calculates a post-suppression coefficient from the corrected suppression coefficient output from the suppression coefficient correction unit 15 in FIG. 31 based on the speech existence probability output from the speech existence probability calculation unit 26 in FIG. This is transmitted to the multiple multiplier 173. The difference between the post-suppression coefficient calculator 172 of FIG. 30 and the post-suppression coefficient calculator 252 of FIG. 32 is that the speech non-existence probability is calculated outside the emphasized speech amplitude spectrum correction unit.

  FIG. 33 is a block diagram showing the configuration of the speech existence probability calculation unit 26 of FIG. The speech existence probability calculation unit 171 in FIG. 3 is different from the speech existence probability calculation unit 171 in that the separation unit 1708, the average value calculation unit 1709, the logarithm calculation unit 1710, the multiplier 1711, and the function value calculation units 1712 and 1713 are deleted. 1714 to 2614 are the same except that the instantaneous index calculation unit is replaced from 1715 to 2615 and the input to the separation unit 1700 is replaced from the enhanced speech power spectrum to the estimated innate SNR. The common point of the speech existence probability calculation unit 171 in FIG. 3 and the speech existence probability calculation unit 26 in FIG. 33 is that an index is calculated according to the ratio of speech to noise. The voice presence probability calculator 171 corrects both the emphasized voice power and the estimated noise power to values suitable for index calculation, while the voice presence probability calculator 26 corrects the estimated innate SNR. For this reason, the voice presence probability calculation unit 26 can be realized with a smaller amount of calculation. Hereinafter, detailed operations will be described focusing on these differences.

Separation section 1700 separates the estimated innate SNR supplied from estimated innate SNR calculation section 7 in FIG. 31 into the frequency-specific estimated innate SNR, and outputs the result to average value calculation section 1701. The average value calculation unit 1701 divides the sum of the estimated innate SNR ξ _n (k) by frequency from k = 0 to K−1 by K, and transmits the calculation result to the logarithmic calculation unit 1702. The logarithm calculation unit 1702 calculates the logarithm of the average value input from the average value calculation unit 1701 and transmits the logarithm to the multiplier 1703. The multiplier 1703 multiplies the supplied logarithmic value by a constant to obtain a full-band estimated innate SNR Ξ (n) and supplies it to the smoothing units 1705 and 1707. That is, the full band estimation innate SNR Ξ (n) of the nth frame is given by the following equation.

  The smoothing unit 1705 smoothes the full-band estimated innate SNR Ξ (n) supplied from the multiplier 1703 in the time direction using the smoothing coefficient supplied from the smoothing coefficient storage unit 1704, and performs the first smoothing. This is supplied to the instantaneous index calculation unit 2615 as an innate SNR. Similarly, the smoothing unit 1707 uses the smoothing coefficient supplied from the smoothing coefficient storage unit 1706 to smooth the full-band estimated innate SNR Ξ (n) supplied from the multiplier 1703 in the time direction. Is supplied to the average index calculation unit 2614 as a smooth innate SNR. As described when explaining the speech existence probability calculation unit 171 in FIG. 3, the coefficient stored in the smoothing coefficient storage unit 1704 is set to be smaller than the coefficient in the smoothing coefficient storage unit 1706. The

The instantaneous index calculation unit 2615 calculates an instantaneous index using the first smoothed innate SNR supplied from the smoothing unit 1705 and supplies the calculated instantaneous index to the adding unit 1716. The average index calculation unit 2614 calculates the average index using the second smoothed innate SNR supplied from the smoothing unit 1707 and supplies the average index to the addition unit 1716. For calculating the index, a method of increasing the numerical value according to the smooth innate SNR is used. Specific examples include the following calculation methods.

Where IDX2 _n is an index and Ξ (n) bar is a smooth innate SNR. Also, θ _idx2 , a _idx2 and b _idx2 are real numbers, and a _idx2 has a value equal to or greater than b _idx2 .

  FIG. 34 is a block diagram showing a ninth embodiment of the present invention. The difference between FIG. 34 and FIG. 31 of the eighth embodiment is that the voice presence probability calculation unit 26 is replaced with a voice presence probability calculation unit 27. The speech existence probability calculation unit 27 calculates a speech presence probability using the acquired SNR output from the frequency-specific SNR calculation unit 6 and the estimated innate SNR output from the estimated innate SNR calculation unit 7, and an adder 24 and the enhanced speech amplitude spectrum correction unit 25. A detailed description of the configuration and operation of the speech existence probability calculation unit 27 will be given with reference to FIG.

  FIG. 35 is a block diagram showing a configuration of the speech existence probability calculation unit 27 of FIG. 31 is different from the speech existence probability calculation unit 26 in that the separation unit 1700 is replaced with 2700, the average value calculation unit 1701 is replaced with 2701, and the separation unit 2703, the average value calculation unit 2704, and the SNR mixing unit 2705. Is the same except that is provided. The main difference from the speech existence probability calculation unit 26 in FIG. 31 is that the estimation accuracy of the SNR input to the logarithmic calculation unit 1702 is improved. Hereinafter, detailed operations will be described focusing on these differences.

The separation unit 2700 separates the estimated innate SNR supplied from the estimated innate SNR calculation unit 7 in FIG. 34 into the frequency-specific estimated innate SNR, and outputs it to the average value calculation unit 2701. The average value calculation unit 2701 divides the sum of the estimated innate SNR ξ _n (k) by frequency from k = 0 to K−1 by K, and transmits the calculation result to the SNR mixing unit 2705 as an average innate SNR ξ _n bar. To do. That is, the average innate SNR ξ _n bar of the nth frame is given by

On the other hand, the separation unit 2703 separates the acquired SNR supplied from the frequency-specific SNR calculation unit 6 of FIG. 34 into the frequency-specific acquired SNR, and outputs it to the average value calculation unit 2704. Average value calculation section 2704 divides the sum of frequency-specific acquired SNRγ _n (k) from k = 0 to K−1 by K, and transmits the calculation result as average acquired SNRγ _n bar to SNR mixing section 2705. That is, the average acquired SNRγ _n bar of the nth frame is

Given in.

SNR mixing unit uses the average and congenital SNRkushi _n bars supplied from the average value calculating unit 2701, the average acquired SNRganma _n bars supplied from the average value calculating unit 2703, a mixed SNRΞ _mix (n) calculated To the logarithm calculation unit 1702. In calculating the mixed SNR Ξ _mix (n), a method of increasing the numerical value according to the average innate SNRξ _n bar is used. Specific examples include the following calculation methods.

Where F _mix is a function of the average innate SNRξ _n bar.

F _mix outputs a real number from 0 to 1, and outputs a large value if ξ _n bar is large. That is, when the SNR is high, calculates the average congenital SNRkushi _n estimation accuracy than bar higher average acquired SNRganma _n bar preferentially used mixed SNRΞ _mix (n). Therefore, the estimation accuracy of the mixed SNRＳ _mix (n) obtained using both the innate SNR and the acquired SNR is higher than the full-band estimation innate SNRＳ (n) obtained using only the innate SNR. . Since it is possible to calculate the speech existence probability using the SNR with high estimation accuracy, the speech presence probability calculation unit 27 in FIG. 34 can achieve higher accuracy than the speech presence probability calculation unit 26 in FIG.

  In all the embodiments described so far, the minimum mean square error short-time spectrum amplitude method has been assumed as a noise suppression method, but it can also be applied to other methods. Examples of such methods include the Wiener filter method disclosed in Non-Patent Document 2 and the spectral subtraction method disclosed in Non-Patent Document 3, but the detailed description of these configuration examples is omitted. To do.

DESCRIPTION OF SYMBOLS 1 Frame division part 2 Windowing process part 3 Fourier transform part 4,5049 Counter 5 Estimated noise calculation part 6,1402 Frequency-specific SNR calculation part 7,71,72 Estimated innate SNR calculation part 8 Noise suppression coefficient generation part 9 Inverse Fourier Conversion unit 10 Frame synthesis unit 11 Input terminal 12 Output terminal 14 Weighted degraded speech calculation unit 15 Suppression coefficient correction unit 172, 182, 252, 282, 292 Post suppression coefficient calculation unit 13, 16, 170, 173, 704, 705 713, 715, 1404 Multiplexing units 17, 18, 22, 25, 28, 29 Enhanced speech amplitude spectrum correction unit 21 Speech non-existence probability storage units 171, 221, 26, 27, 714 Speech presence probability calculation units 1742, 1745, 708, 5046, 1716, 7092, 7094, 24 Adders 711, 712, 1746, 23 Extender 1593, 5204, 5206 Threshold storage unit 1594, 5203, 5205 Comparison unit 1702, 1710, 1859 Logarithm calculation unit 1704, 1706, 1854 Smoothing coefficient storage unit 1705, 1707, 1853 Smoothing unit 1712, 1713 Function value calculation unit 1714, 2614 Average index calculation unit 1715, 2615 Instantaneous index calculation unit 2705 SNR mixing unit 1852 Audio power mixing unit 1858 Smooth signal storage unit 1863 Exponential calculation unit 7071 _{0 to} 7071 _K-1 Weighted addition unit 706 Weight storage units 503, 1304 , 1424,1475,1504,1723,7014,7075 multiplexer ₅₀₄ 0 _{~504 K-1} frequency domain estimated noise calculator 520 update determination unit 5207 threshold calculating unit 1595,5044 switches 1857,1862,1421 ₀ 421 _K-1, 5048 divider 501,502,1302,1303,1422,1423,1495,1502,1503,1700,1708,1722,1850,1855,7013,7072,7074,2700,2703 separation unit 1701,1709 , 1851, 1856, 2701, 2704 Average value calculation unit 701 Multiple range restriction processing unit 702 Acquired SNR storage unit 703 Suppression coefficient storage unit 707 Multiple weighted addition units 1401, 5042 Estimated noise storage unit 921 Instantaneous estimation SNR
921 _{0 to} 921 _K-1 instantaneous estimated SNR by frequency
922 Past estimated SNR
922 _{0 to} 922 _K-1 Estimated SNR by frequency in the past
924 Estimated innate SNR
924 _{0 to} 924 _K-1 Estimated Innate SNR by Frequency
1405 Multiple nonlinear processing units 1485 _{0 to} 1485 _K−1 , 5042 Nonlinear processing units 1501 _{0 to} 1501 _K−1 , 1901 _{0 to} 1901 _K−1 , 2001 _{0 to} 2001 _K−1 Frequency-specific suppression coefficient correcting units 1721 _{0 to} 1721 _K-1 , 1821 _{0 to} 1821 _K-1 , 2821 _{0 to} 2821 _K-1 , 2921 _{0 to} 2921 _K-1 frequency-specific post-suppression coefficient calculator 1591, 1694, 1924, 7012 _{0 to} 7012 _K-1 maximum value selection Unit 1592 suppression coefficient lower limit value storage unit 1596 modified value storage unit 1691, 1921 voiced part lower limit value storage unit 1692, 1922 silent part lower limit value storage part 2691 voiced part lower limit value calculation part 2692 silent part lower limit value calculation part 1693, 1923 Lower limit calculation unit 1831 Coefficient storage unit for sound part 2831 Coefficient calculation for sound part 1832 silence coefficient calculating unit 1833,1861 coefficient calculation unit 2011 talkspurt correction coefficient storage unit 2012 silence correction coefficient storage unit 2013 the correction coefficient calculation unit 1301 ₀ ~1301 _K-1, 1597,1703,1711,1743 , 1744, 1834, 2014, 7091, 7093 multipliers 1741, 1860, 7095 constant multiplier 5045 shift register 5047 minimum value selection unit 5201 logical sum calculation unit 5041 register length storage unit 7011 constant storage unit 811 MMSE STSA gain function value calculation unit 812 Generalized likelihood ratio calculator 814 Suppression coefficient calculator

Claims (4)

Convert the input signal to a frequency domain signal,
Determine a suppression coefficient based on the frequency domain signal,
Obtaining the relative relationship between speech and noise based on the frequency domain signal,
A contribution rate is determined based on the relative relationship ,
Obtaining a minimum suppression coefficient based on the contribution rate and a predetermined first temporary minimum suppression coefficient and a second temporary minimum suppression coefficient ;
Comparing the suppression coefficient with the minimum suppression coefficient;
The one with the larger value is the correction suppression coefficient,
A noise suppression method for suppressing noise by weighting the corrected suppression coefficient to the frequency domain signal ,
The minimum suppression coefficient is
Using the contribution rate as a weight,
A noise suppression method characterized by being determined by a weighted sum of the first temporary minimum suppression coefficient and the second temporary minimum suppression coefficient . The first temporary minimum suppression coefficient and the second temporary minimum suppression coefficient, the method of noise suppression according to claim 1, characterized in that it is determined based on the frequency-domain signal. A converter for converting an input signal into a frequency domain signal;
A suppression coefficient calculator that determines a suppression coefficient based on the frequency domain signal;
A relative relationship calculation unit for obtaining a relative relationship between speech and noise based on the frequency domain signal;
Determining a contribution rate based on the relative relationship, a minimum suppression coefficient calculation unit for obtaining a minimum suppression coefficient based on the contribution rate and a predetermined first temporary minimum suppression coefficient and a second temporary minimum suppression coefficient ;
A correction suppression coefficient calculation unit that compares the suppression coefficient with the minimum suppression coefficient and sets a larger value as a correction suppression coefficient;
A weighting operation unit for weighting the corrected suppression coefficient to the frequency domain signal;
Including a noise suppression device comprising :
The minimum suppression coefficient is
Using the contribution rate as a weight,
An apparatus for noise suppression, which is determined by a weighted sum of the first temporary minimum suppression coefficient and the second temporary minimum suppression coefficient . The apparatus for noise suppression according to claim 3, wherein the first temporary minimum suppression coefficient and the second temporary minimum suppression coefficient are obtained based on the frequency domain signal.

Images