JP2000082999A - Noise reduction processing method/device and program storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce a processing delay without being affected by S/N and to reduce deterioration in sound quality in terms of listening. SOLUTION: An input signal is divided into a plurality of frequency bands (22) and power is calculated at every band (24). A noise rate against an instantaneous objective sound signal is estimated by using respective band input signals (201). The gain factors of the respective bands are decided based on instantaneous SN (27). Noise is reduced by using the gain factors against the respective band input signals (28), the long time average value of instantaneous S/N is calculated (52), an addition rate is decided based on the long time average value (53), the respective band input signals and the noise reduction signal are added at the addition rate (54) and the addition signal is converted into a time area signal for the whole frequency band (29).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention
In a voice / sound device such as a TV conference device, an input signal in which a signal of interest and a signal of unnecessary noise and the like are mixed,
The present invention relates to a noise reduction processing method for outputting a signal with reduced noise, a device therefor, and a program storage medium.

[0002]

2. Description of the Related Art In a voice communication system such as a voice conference or a TV conference, if a surrounding signal other than a target voice is mixed in a transmission signal received by a microphone and transmitted to a partner side, the clarity of the voice is increased. And the call quality is significantly degraded. For this reason, there is a strong demand for reducing ambient noise other than the target voice included in the transmission signal.

[0003] The noise reduction method is a technique of outputting a noise-reduced signal from an input signal in which a target voice signal and a signal such as unnecessary ambient noise are mixed. FIG. 4 shows a sound pick-up system, which is used for a conventional noise reduction method.
The device will be described. In this specification, the time representation of a signal uses an integer value n representing a discrete time, for example, X
(N).

[0004] A target voice signal 13 uttered by a speaker 12 located at a position distant from the microphone 11 is represented by S
(N), N (n) represents unnecessary ambient noise 14 such as air-conditioning, and X (n) represents the input signal 15 which is received by the microphone 11 and input to the noise reduction device 16. , The output signal 17 of the noise reduction device 16 is represented by Y (n).
And The input signal X (n) to the noise reduction device 16 includes:
Ambient noise N (n) is mixed in addition to the target audio signal S (n). That is, X (n) = S (n) + N (n) (1) At this time, the noise N in the input signal X (n)
A device that reduces (n) and extracts a signal close to the target audio signal S (n) as an output signal Y (n) is called a noise reduction device.

FIG. 5 shows Spectrum Subtraction (SFBol
l, IEEE Trans. on ASSP, vol.27, no.2, pp.113-120,
Apr (1979)). Wiener Filer (JSLim. & AVOppenheim,
in Proc.IEEE, vol.67, no.12, pp.1586-1604, Dec (1
979)). Maximum Likelihood Envelop (RJMcAulay &
MLMalpass, IEEE Trans.on ASSP, vol.28, no.2, p
p.137-145, Apr (1980)). minimum mean squared error
method (MMSE) (Y.Ephraim & D.Malah, IEEE Trans.on AS
SP, vol.32, no.6, pp.1109-1121, Dec (1984)) etc. (Short Time S)
This is a noise reduction method based on Spectral Amplitude (STSA) Estimation, and shows the functional configuration of a conventionally used method. Using this, a conventional noise reduction method will be described. Common symbols are used for the same elements as in FIG.

First, an A / D converter 21 digitizes an input signal 15 mixed with a target signal and unnecessary noise received by a microphone 11, and transfers the digitized signal to a frequency band division unit 22 and a noise discrimination unit 23. . In the frequency band dividing unit 22, the transferred signal is divided into M frequency bands. Division into frequency bands is performed using, for example, a discrete Fourier transform or the like. Here, the band-divided signal is generally a complex number, but may be a real number depending on the division method. Here, in general, discussion will be made assuming a complex number, but the same argument can be made for a real number. X _k (n) = X _{k, r} (n) + jX _{k, i} (n) (2) where X _{k, r} , X _{k, i} are X _k (n), the real part and the imaginary part of X (n)), X _k (n) is
4. The signal is transferred to the input signal phase calculator 25 and the noise power calculator 26. The input signal power calculator 24 calculates the input signal power level P _{X, k} (n) = X _{k, r} (n) ² + X _{k, i} (n) ² (3) for each band. The unit 25 calculates the phase Φ _k (n) = tan ^-1 [X _{k, i} (n) / X _{k, r} (n)] (4) for each band. Then, P _{X, k} (n) becomes instantaneous S / N
Φ _k (n) is transferred to the ratio estimating unit 201 and the gain factor inserting unit 28, and Φ _k (n) is transferred to the time domain transforming unit 29.
On the other hand, in the noise discrimination unit 23, first, for X (n) transferred from the A / D conversion unit 21, the power level P _X (n) = {X (nk) ² (5)} is k = From 0 to L-1 are calculated. Here, L represents the integration time. Next, for example, a determination of P _X (n) <P _th (6) is performed for a predetermined threshold value P _th , and if this conditional expression is satisfied, it is determined that the noise is present. The noise power calculator 26 sets the power level of each band of noise to P _{N, k} (n) = X _{k, r} only when the noise discriminator 23 determines that the input signal X (n) is noise. (n) ² + X _{k, i} (n) ² (7), and the time average Pav _{N, k} (n) is calculated as the instantaneous S / N
Transfer to ratio estimating section 201. Time average, for example _{Pav N, k (n) =} (1 / A) Σ m γ m P N, is calculated as k (n-m) (8 ). Here, γ _m is, for example, an exponentially weighted coefficient expressed as γ _m = (γ) ^m (9) (γ <1),
A is a constant for normalization to be _{(1 / A) Σ m γ} m = 1 (10).

In the instantaneous S / N ratio estimator 201, P _{X, k} (n) calculated by the input signal power calculator 24 for each band.
And the noise power Pav calculated by the noise power calculator 26
_The S / N ratio, which is the ratio of the target speech signal to the noise signal, is estimated using _{N, k} (n). FIG. 6 shows a flowchart of the processing for the instantaneous S / N ratio estimating unit 201. Detailed description of the processing will be made using this figure.

[0008] First, in step 31, calculates the _{SNR k (n) '= P} X, k (n) / P N, k (n) SNR k defined in (11) (n)'. Next, in step 32, for the determined SNR _k (n) ′, (1
SNR _k by averaging using the estimated value (noise reduced power) P _{Y, k} (n-1) one time ago represented by the expression (3)
(n). That is, SNR _k (n) = (1−β) P [SNR _k (n) ′ − 1] + β [P _{Y, k} (n−1) / P _{N, k} (n−1)] (12) . P [*] takes * if * is positive and 0 if * is negative. This SNR _k (n) and, if necessary, SNR _k (n) ′
Is transferred to the gain factor calculator 27.

In the gain factor calculation unit 27, the instantaneous S
SNR _k (n) transferred from the / N ratio estimation unit 201,
In some cases, using this and SNR _k (n) ′, a gain factor G (SNR _k (n)) at each frequency defined in each noise reduction scheme is calculated. The gain factor G (SNR _k (n)) is transferred to the gain factor insertion unit 28. FIG. 7 shows a gain factor according to each method. SNR _k is i.e. three methods above in FIG. 5
(n) is used to determine the gain factor, but when the lowest MMSE method is used, SNR _k (n) and SNR _k
(n) 'is used.

The gain factor insertion unit 28 performs noise reduction for each band using the gain factor calculated by the gain factor calculation unit 27. That is, the band signal P transferred from the input signal power calculator 24
_With respect to _{X, k} (n), a band output P _Y, obtained by calculating P _{Y, k} (n) = G (SNR _k (n)) × P _{X, k} (n) (13) to reduce noise Output _k (n). P _{Y, k} (n) is transferred to the time domain transformation unit 29, and Y _k (n) = Y _{k, r} (n) using Φ _k (n) sent from the input signal phase calculation unit 25. + JY _{k, i} (n) where Y _{k, r} (n) = √P _{Y, k} (n) cos [Φ _k (n)] Y _{k, i} (n) = √P _{Y, k} (n) sin [Φ _k (n)] (14), synthesized into a full-band signal, and further converted into a time-domain signal by, for example, an inverse discrete Fourier transform. The D / A converter 30 converts the result into an analog signal, and outputs a signal 17, Y (n) in which noise is reduced.

In these conventional systems, 256 to 102
In many cases, processing is performed by dividing into four frequency bands,
Under this condition, S defined by the following equation (15)
/ N _{ratio, S / N = 10 log 10} (P S / P N) , however, P _S: Average object signal power P _N: when the average power of the noise signal (15) has more than about 15dB, the noise reduction Improves the sound quality, but if it is less than about 10 dB,
Although the noise is reduced, the sound signal is distorted in accordance with the noise, and the remaining noise is temporally changed, so that the auditory sense is deteriorated and the quality is deteriorated. Also 256-1
If the process of dividing into frequency bands of 024 is used as it is in a conference system, a large delay occurs, and the call performance deteriorates when a conference is held. If the number of frequency divisions is reduced and the delay is reduced to prevent this, the distortion due to the non-linear processing increases, so that even if the S / N ratio is about 15 dB, the voice quality deteriorates.

That is, the conventional method has a problem that noise cannot be effectively reduced when the S / N ratio is poor or when it is desired to reduce the delay. For this reason, the audio conference device T
This method cannot be applied as it is to a V-conference device or the like for sound collection for which the sound quality is important for the purpose of listening and real-time performance is required, or for calls under high noise conditions.

[0013]

As a method for reducing the noise of an input signal mixed with ambient noise, a conventional method of subtracting the amplitude of the noise from a speech signal has a problem that the S / N ratio is poor or the delay is small. In such a case, there is a problem in that the sound signal is distorted, and the noise left behind is changed over time, so that an unpleasant sound is heard. An object of the present invention is to provide a noise reduction processing method, an apparatus thereof, and a program recording medium which have a small processing delay and a small deterioration in audible sound quality regardless of the S / N ratio.

[0014]

According to the present invention, a sound signal received by a microphone in which a target voice signal and ambient noise are mixed is divided into a plurality of bands, and noise for each band signal is divided. Estimating the power, comparing the estimated noise power with the input signal power that has actually been input, and estimating the ratio of the instantaneous and long-term average speech signal and noise signal,
Noise is suppressed by multiplying the gain factor for noise suppression calculated based on the ratio of the instantaneous speech signal and noise signal to the input signal for each band, and the long-term average speech signal and noise signal Distortion caused by adding the optimal amount of band-specific signals determined according to the ratio is masked, and as a result, effective noise reduction with less distortion is enabled.

The present invention has the following features. Measures the ratio of the long-term average audio signal to noise signal of each band, and adaptively adds the optimal amount of each band signal to prevent deterioration in sound quality and maximize the noise reduction effect at the same time It is possible to

[0016]

FIG. 1 is a block diagram showing an embodiment of the present invention. The processing procedure of the present invention will be described with reference to FIG. 4 and 5, the same symbols are used. A noise power estimation unit 51, a long-time S / N ratio calculation unit 52, an optimum input signal addition rate determination unit 53, and an input signal addition unit 54 are provided. The noise power estimation may be the one shown by the conventional method, but here, here, Japanese Patent Application No. Hei 8-685.
The following description will be made by estimating using the method shown at 48.

First, an A / D converter 21 digitizes an input signal 15 mixed with a target signal and unnecessary noise received by a microphone 11 and transfers the digitized signal to a frequency band divider 22. In the frequency band dividing section 22, the transferred signal is divided into frequency bands. Each of the divided band signals is transferred to the input signal power calculator 24, the noise power estimator 51, the gain factor inserter 28, and the input signal adder 54. Hereinafter, the k-th band signal is _represented by X _k (n)
The process flow for X _k (n) will be described.

The input signal power calculator 24 transfers the
X_k(n) power level P_{X, k}(n) = X_{k, r}(n)^Two+ X_{k, i}(n)^Two (16) is calculated and transferred to the instantaneous S / N ratio estimating unit 201. Miscellaneous
In the sound power estimating unit 51, the transmitted X_kusing (n)
Noise power Pav _{N, k}(n) is estimated. Noise
FIG. 2 shows the processing flow of the word estimation unit. Details
New processing will be described with reference to FIG.

First, at step 61, the power level P _{X, k} (n) of the transferred X _k (n) is calculated by using the equation (16). Next, P _{X, k} (n) is averaged over a predetermined time using the following equation, and Pav _{X, k} (n) is averaged.
And _{Pav X, k (n) =} (1 / A) Σ m γ m P X, is calculated as _k (nm) (17). Here, γ _m is, for example, an exponentially weighted coefficient expressed as γ _m = (γ) ^m (18) (γ <1),
A is a constant for normalization to be _{(1 / A) Σ m γ} m = 1 (19). The average time is, for example, 5 to 6 msec.

Next, at step 62, a histogram of the level distribution of Pav _{X, k} (n) is _obtained . That is, Pav _{X, k} (n)
Is added to the number h _k (int Pav _{X, k} (n)) of the power section to which it belongs, that is, h _k (int Pav _{X, k} (n)) = h _k (int Pav _{X, k} (n)) + 1 (20) is calculated. int (*) indicates that the decimal part is truncated to an integer. Further, in step 63, the histogram h
The peak section of _k (i) is detected and stored. That is, i that satisfies h _k (i ′) ≦ h _k (i) (21) with respect to values before and after is obtained. The smallest i among the i is set as the noise power Pav _{N, k} (n). Pav _{N, k} (n) is the instantaneous S
/ N ratio estimating section 201.

The instantaneous S / N ratio estimator 201 calculates P _{X, k} (n) calculated by the input signal power calculator 24 and the noise power Pav _{N, k} (n) estimated by the noise power estimator 51. SNR _k (n) ′ and SNR in the manner shown in FIG.
_k (n) is estimated. The SNR _k (n) ′ estimated by the instantaneous S / N ratio estimator 201 is calculated by the gain factor calculator 27
The SNR _k (n) is transferred to the gain factor calculator 27 and the long-time S / N ratio calculator 52.

In the gain factor calculation section 27, the instantaneous S
SNR _k (n) ′ transferred from the / N ratio estimation unit 201
And SNR _k (n), the gain factor G (SN
R _k (n)) are determined. Here, the gain factor G (SNR _k (n)) in the gain factor calculation unit 27
The calculation method of Wiener Fi shown in FIG.
litering method, envelope maximum likelihood estimation method (Maxi
mum likeelihood envelope e
), spectrum subtraction method (Spectrum subtraction method)
n), minimum mean square error method (MMSE; minimum)
mean squared errormetho
d) and the like are typical. The gain factor G (SNR _k (n)) estimated by the gain factor calculation unit 27 is transferred to the gain factor insertion unit 28. Note that instantaneous S
The method of the / N ratio estimation unit 201 and the method of the gain factor calculation unit 27 may be methods other than those described here.

The gain factor insertion unit 28 performs noise reduction using the gain factor calculated by the gain factor calculation unit 27. That is, the frequency band dividing unit 2
Relative band signal has been transferred from the _{2 X k (n), Y} k '(n) = G (SNR k (n)) × X k (n) (22) is calculated and a signal with reduced noise Y _k ′ (n) is transferred to the input signal adding section 54.

The long-time S / N ratio calculation section 52 calculates the transferred SNR _k (n) for a predetermined time.
Averaging is performed using the following equation to obtain SNRav _k (n), which is transferred to the optimum input signal addition rate determination unit 53. It is calculated as _{SNRav k (n) = (1} / A) Σ m γ m SNR k (nm) (23). Here, γ _m is an exponentially weighted coefficient expressed as γ _m = (γ) ^m (24) (γ <1),
A is a constant for normalization to be _{(1 / A) Σ m γ} m = 1 (25). The average time is, for example, 1 to 3 seconds.

In the optimum input signal addition rate determination section 53, the SNRav _k transferred from the long-time S / N ratio calculation section 52 is used.
Based on (n), the optimum input signal addition rate α of the input signal is determined. For example, when the number of frequency band divisions is 32, 15 ≦ 10 log ₁₀ [SNRav _k (n)] → α = 0.05 10 log ₁₀ [SNRav _k (n)] <15 → α = 0.5 (26) To decide. The optimum input signal addition rate α is transferred to the input signal addition unit 54.

In the input signal adding section 54, the band signal X transferred from the frequency band dividing section 22, the gain factor inserting section 28, and the optimum input signal adding rate determining section 53, respectively.
_The band output signal Y _k (n) is output using _k (n), the noise-reduced signal Y _k ′ (n), and the optimum input signal addition rate α.
That is, _Yk (n) = α × _Xk (n) + (1−α) × _Yk ′ (n) (27) Y _k (n) is transferred to the time domain conversion unit 29 and is converted to the time domain. The result is converted into an analog signal by the D / A converter 30 to output a signal 17, Y (n) in which noise is reduced.

The effectiveness of the present invention was evaluated using an opinion evaluation method assuming the situation of an actual TV conference system. The evaluation experiment was performed with a volume of 87 m ³ and a reverberation time of 300 mse.
c, The test was performed in a variable reverberation room having a background noise level of 30 dB (A) or less. The evaluation was performed on 20 subjects. The created stimulus sound was played from the speaker, and the quality was evaluated by the subject as the received voice in the public conference. The number of frequency divisions was examined for two cases: 256, which is general, and 32, which has low processing delay.

FIGS. 3 (a) and 3 (b) show the MO of a signal subjected to noise reduction processing with the input signal addition rate fixed (fixed value 0 to 0.5).
This is a comparison of S values (5 levels), and was performed on a signal having an S / N ratio of 5 to 25 dB when the number of frequency divisions was 32 and 256, respectively. Here, the original sound addition rate = 1 corresponds to the signal before the noise reduction processing, and the original sound addition rate = 0 corresponds to the processed signal according to the conventional method in which the noise reduction processing is performed and the original sound is not added. FIG. 3 (c) shows reference audio data (original audio and S / N ratio −5 with noise added thereto).
５０50 dB) and the MOS value.

3 (a) and 3 (b), the optimum input signal addition rate for maximizing the MOS value depends on the S / N ratio of the input signal. It can be seen that the ratio is about 0.05 when the ratio is good and 0.5 or more when the ratio is bad. Regarding this result, when the S / N ratio is good, the processing distortion is small, so that the amount of original sound necessary for masking is small and sufficient. On the other hand, when the S / N ratio is bad, the processing distortion is large. Therefore, it can be considered that the amount of the original sound also increases. Further, the MOS value improvement effect from the conventional method at the optimum input signal addition rate is about 0.2 when the S / N ratio is good, and about 0.6 to 1.0 when the S / N ratio is bad. This is approximately 5 to 10 dB from the comparison with FIG.
Can be estimated to be equivalent to the improvement in the S / N ratio.

From the above results, it is considered that the present invention is effective in that the input signal addition rate is not fixed but is adaptively changed according to the S / N ratio of the input signal. In the conventional method, S /
When the N ratio is poor, the MOS value is lower than before the noise reduction processing, or when the number of frequency divisions is reduced, the MOS value is lower than when the number of divisions is large. According to the present invention, an input signal is adaptively added at an optimum input signal addition rate according to an S / N ratio of an input signal.
Even when the / N ratio is poor, a subjective S / N of about 5 to 10 dB
The effect of improving the ratio can be obtained, and the same MOS value can be realized regardless of the number of frequency divisions.

Each unit in FIG. 1 can be configured so that a computer reads out a program, executes the program, and makes it function.

[0032]

According to the present invention, a signal in which a target voice and noise are mixed is divided into frequency bands, a ratio of a noise signal to an instantaneous target signal in each band is estimated, and a gain for each band is estimated based on the estimation result. Determine the factor to reduce noise. According to the present invention, the original signal is added to the noise-reduced sound in accordance with the ratio of long-time noise to the sound in order to reduce distortion of the target sound which occurs when the S / N ratio is poor or the number of frequency band divisions is small. By doing so, a noise-reduced voice with little distortion can be obtained.

The noise reduction for the purpose of listening by the method and the apparatus makes it possible to obtain a target signal that is easy to hear even under a low delay or low S / N. As a result, in a voice communication system such as a voice conference or a TV conference, even when ambient noise other than the target voice is mixed in a transmission signal received by a microphone and transmitted to the other party, this method and apparatus can be used. As a result, the clarity of the voice can be maintained and the communication quality can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a functional configuration of an embodiment of the present invention.

FIG. 2 is a flowchart showing a processing procedure in a noise power estimation unit 51 in FIG. 1;

3A and 3B are diagrams showing the effect of the present invention, and FIG. 3A is a diagram showing the subjective evaluation result (the number of divisions: 32) of the noise reduction effect according to the present invention, and FIG. FIG. 7C is a diagram showing the number of divisions 256), and FIG. 7C is a diagram showing the subjective evaluation result of the reference signal.

FIG. 4 is a diagram illustrating the concept of a noise reduction device.

FIG. 5 is a block diagram showing a functional configuration of a conventional noise reduction device.

FIG. 6 is a flowchart showing a processing procedure in an instantaneous S / N ratio estimating unit 201 in FIG. 5;

FIG. 7 is a diagram showing a representative gain factor.

Claims (9)

[Claims] 1. A noise reduction processing method for reducing a noise signal from an input signal in which a target voice signal and a noise signal are mixed, wherein: a frequency band dividing step of dividing the input signal into a plurality of frequency bands; An input signal power calculation step of calculating power for each frequency band for the input signal divided into each frequency band by the frequency band division step, and an input signal divided into each frequency band by the frequency band division step. Estimating the ratio of noise to the instantaneous target audio signal in each of the frequency bands using the gain factor calculation for determining the gain factor for each frequency band based on the estimated instantaneous target audio signal to noise ratio. The input signal divided into each frequency band by the frequency band dividing step. A noise reduction process of reducing noise based on the input signal, a process of calculating a long-term average value of the estimated instantaneous target audio signal-to-noise ratio, and an input signal divided into each frequency band by the frequency band division process. A step of determining, for each of the frequency bands, an addition ratio with the signal whose noise has been reduced by the noise reduction step, based on the long-term average value; and an input signal divided into each frequency band by the frequency band division step. And a step of adding a signal whose noise has been reduced by the noise reduction step based on the determined addition ratio to obtain a band output signal, and a step of synthesizing a signal over the entire frequency band from the band output signal. A noise reduction processing method comprising: 2. The noise reduction processing method according to claim 1, wherein the addition ratio is determined according to the long-term average value and the number of frequency bands. 3. The noise reduction processing method according to claim 2, wherein the long-term average value is 1 when the number of frequency bands is 256.
A noise reduction processing method, wherein the addition ratio is set to 0.5 when it is worse than 0 dB, and the addition ratio is set to 0.05 when the long-term average value is better than 10 dB. 4. The noise reduction processing method according to claim 2, wherein the long-term average value is 15 when the number of frequency bands is 32.
A noise reduction processing method, wherein the addition ratio is set to 0.5 when the average value is worse than 15 dB, and the addition ratio is set to 0.05 when the long-term average value is better than 15 dB. 5. A noise reduction processing apparatus for reducing a noise signal from an input signal in which a target voice signal and a noise signal are mixed, wherein: a frequency band dividing means for dividing the input signal into a plurality of frequency bands; Input signal power calculation means for calculating power for each of the frequency bands for the input signal divided into each frequency band by the frequency band division means, and an input signal divided into each frequency band by the frequency band division means. An instantaneous target audio signal-to-noise ratio estimating means for estimating a ratio of noise to an instantaneous target audio signal in each of the frequency bands, and a gain factor for each frequency band based on the instantaneous target audio signal-to-noise ratio. Gain factor calculating means for determining, and an input signal divided into each frequency band by the frequency band dividing means. Noise reducing means for reducing noise based on the gain factor, a long-term average value calculating means for calculating a long-term average value of the instantaneous target audio signal-to-noise ratio, and a frequency band dividing means for each frequency band. An adding ratio determining unit that determines an adding ratio of the divided input signal and the signal whose noise is reduced by the noise reducing unit for each of the frequency bands based on the long-term average value, A band output signal synthesizing unit for obtaining a band output signal by adding the input signal divided into each frequency band and the signal whose noise has been reduced by the noise reduction unit based on the addition ratio; A noise reduction processing device comprising: an entire frequency band signal synthesizing means for synthesizing a signal over a frequency band. 6. The noise reduction processing device according to claim 5, further comprising: means for determining the optimum addition ratio according to the long-term average value and the number of frequency bands. 7. The noise reduction processing apparatus according to claim 6, wherein when the number of frequency bands is 256, the long-term average value is 1
A noise reduction processing device, wherein the addition ratio is selected to be 0.5 when it is worse than 0 dB, and the addition ratio is selected to be 0.05 when the long-term average value is better than 10 dB. 8. The noise reduction processing apparatus according to claim 7, wherein the long-term average value is 15 when the number of frequency bands is 32.
When the average is longer than 15 dB, the addition ratio is set to 0.5 when the average is longer than 15 dB.
05. A noise reduction processing apparatus characterized by being selected as No. 05. 9. A noise reduction process for reducing a noise signal from an input signal in which a target speech signal and a noise signal are mixed, wherein: a frequency band division process of dividing the input signal into a plurality of frequency bands; An input signal power calculation process of calculating power for each of the frequency bands with respect to the input signal divided into each frequency band by the band division process, and an input signal divided into each frequency band by the frequency band division process. A process of estimating the ratio of noise to the instantaneous target audio signal in each of the frequency bands, and a gain factor calculation process of determining a gain factor for each frequency band based on the instantaneous target audio signal to noise ratio. Based on the gain factor for the input signal divided into each frequency band by the frequency band dividing process Noise reduction processing for reducing sound, processing for calculating a long-term average value of the instantaneous target sound signal-to-noise ratio, input signal divided into each frequency band by the frequency band division processing, and noise due to the noise reduction processing. A process of determining the addition ratio with the reduced signal for each of the frequency bands based on the long-term average value, and an input signal divided into each frequency band by the frequency band division process and the noise reduction process. A program for performing a noise reduction process including a process of obtaining a band output signal by adding a signal with reduced noise based on the addition ratio and a process of combining a signal over the entire frequency band from the band output signal. A computer-readable storage medium having stored therein.