JPH10161694A - Band split type noise reducing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent deterioration of quality from being accompanied even when the band split number is minimized. SOLUTION: An input signal is band-divided, and with respect to a signal XK(n) every band, a power PX.k (n) is determined (24), whereby a noise average power PavN.k (n) is estimated (51). SNR)k (n) is estimated by the PX.k (n), PavN.k (n) (27), and a gain factor G(SNRk (n)) is determined by SNRk (n) (30). The addition ratio αaccording to the ratio of voice to noise is determined by SNRk (n) so that it is large when S/N is poor, Xk (n) is multiplied by G(SNRk (n)) to provide a signal Pk (n)' reduced in noise (28), Xk (n) is added to Yk (n)' according to αby αX.k (n)+(1-α)Yk (n)' to provide Yk (n) (53), and all the bands are composed and converted into time area.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention
The present invention relates to a noise reduction method for outputting a noise-reduced voice signal from an input signal in which a target voice signal and a signal such as unnecessary noise are mixed in a voice / sound device such as a TV conference device.

[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. 2A shows a sound collection system, which will be used to explain a conventional noise reduction method. In this specification, the time representation of a signal is represented by, for example, X (n) using an integer value n representing a discrete time. The target audio signal 1 uttered by the speaker 12 now
3 is S (n) and unnecessary ambient noise 14 such as air conditioning is N
(N) The input signal 15 received by the microphone 11 when the voice signal 13 and the noise 14 are received by the microphone 11 and the input signal 15 is input to the noise reduction device 16 is X (n), and the output signal 17 of the noise reduction device 16 is output.
Is Y (n). Input signal X to noise reduction device 16
(N) contains ambient noise N (n) 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. 3 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. The same symbols are used for the same elements as in FIG. 2A.

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 a plurality of 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. Thereafter, P _{X, k} (n) is transferred to the S / N ratio estimating unit 27 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 = 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 noise band 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 transferred to the S / N ratio estimating unit 27. 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).

The S / N ratio estimator 27 calculates P _{X, k} (n) calculated by the input signal power calculator 24 and the noise power Pav calculated by the noise power calculator 26 for each band.
_The S / N ratio, which is the ratio of the target speech signal to the noise signal, is estimated using _{N, k} (n). Regarding the S / N ratio estimating unit 27, a flowchart of the processing is shown in FIG. 4A. Detailed description of the processing will be made using this figure.

[0007] 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 calculation unit 30.

[0008] In the gain factor calculation section 30, S / N
Using the SNR _k (n) transferred from the ratio estimator 27 and possibly the SNR _k (n) ′, the gain factor G (SNR _k ( n)) is calculated. The gain factor G (SNR _k (n)) is transferred to the gain factor insertion unit 28. FIG. 5 shows a gain factor according to each method. That is, in the upper three methods in FIG. 5, the gain factor is obtained using only SNR _k (n), but the lowermost MM
When using the SE method, SNR _k (n) ′ in addition to 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 30. That is, the band signal P transferred from the input signal power calculator 24
_With respect to _{X, k} (n), P _{Y, k} (n) = G (SNR _k (n)) × P _{X, k} (n) (13), and a noise-reduced band output P _{Y, k} ( n) is output.
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), is synthesized into a full-band signal, and further converted into a time-domain signal by, for example, an inverse discrete Fourier transform. The result is converted into an analog signal by the D / A converter 34, and the signal 17, Y (n), in which noise is reduced, is output. Spectr
The al subtraction is a result of subtracting the noise power P _{N, k} (n) from the input signal power P _{X, k} (n) by the calculation of the equation (13).

In these conventional systems, 256 to 102
In many cases, processing is performed by dividing into four frequency bands,
If this is used for the conference system as it is, a large delay occurs. When a signal is delayed during a conference, there is a problem that the call performance deteriorates. On the other hand, reducing the number of frequency division in order to reduce the delay, the following (15) Defined S / N ratio by the _{formula, S / N = 10 log 10} (P S / P N) , however, P _S : Average power of the target signal P _N : Average power of the noise signal (15) When the average power is about 15 dB or more, the quality of voice is improved by noise reduction, but when the average power is about 10 dB or less,
It has been found that the noise is reduced, but the sound signal is distorted in accordance with the noise, and the remaining noise is temporally changed. This is because, in the conventional noise reduction method, when processing is performed on a signal divided into a small number of bands, distortion due to nonlinear processing increases.

That is, in the conventional method, while keeping the delay small,
There is a problem that noise cannot be effectively reduced.
For this reason, the sound quality is important for the purpose of listening, such as a voice conference device and a TV conference device, and this method cannot be applied as it is to a sound collection that requires a real-time property.

[0012]

As a method for reducing the noise of an input signal mixed with ambient noise, the conventional method of subtracting the amplitude of the noise from a speech signal increases the delay when the number of frequency band divisions is large. If the number of frequency band divisions is reduced, there is a problem in that distortion occurs in the audio signal, and undesired sound is produced because the remaining noise changes with time. SUMMARY OF THE INVENTION It is an object of the present invention to provide a noise reduction processing method which has a small processing delay and a small deterioration in sound quality in audibility.

[0013]

SUMMARY OF THE INVENTION The present invention divides a signal received by a microphone in which a target voice signal and ambient noise are mixed into a plurality of bands, and reduces noise power for each band signal. Estimating, comparing the estimated noise power with the input signal power that has actually been input, estimating the ratio of the audio signal to the noise signal, and estimating the amplitude of the audio signal as a noise-reduced signal. The gain factor for noise suppression for each band calculated based on the ratio of the speech signal to the noise signal is multiplied by the input signal of the corresponding band. The estimation of the noise-reduced signal is performed by masking the resulting distortion by adding a small amount of the signal for each band according to the ratio of the voice signal to the noise signal as the noise increases, thereby increasing the noise. As a result, the distortion is reduced. Enables effective noise reduction.

The present invention has the following features. First,
The distortion caused by noise reduction is masked by adding a small amount of signal for each band to prevent deterioration of sound quality. Also, by changing the addition rate of the signal for each band according to the ratio of the audio signal to the noise signal, when the noise is small, that is, when the processing distortion due to the noise reduction is small, the addition amount of the signal for each band is reduced. However, it is possible to maximize the noise reduction effect.

[0015]

FIG. 1 shows a functional configuration of a noise reduction apparatus to which an embodiment of the present invention is applied, and common symbols are used for those corresponding to FIG. For noise power estimation,
Although it may be the one shown in FIG. 3, here, Japanese Patent Application No. Hei 8-68
548, Sasaki, Haneda, "On band-division noise reduction using loss control," Spring Meeting of the Acoustical Society of Japan, 515-5.
16 (1996), the estimation is performed using the method described below. According to this method, the noise power can be detected if there is a temporally non-voice section even in a section in which voice and noise are regarded as being mixed.

First, an input signal 15 mixed with a target signal and unnecessary noise received by the microphone 11 is digitized in an A / D converter 21 and transferred 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 and the gain factor inserter 28. Hereinafter, the k-th band signal of the input signal as X _k (n), for explaining the flow of processing for X _k (n).

The input signal power calculation unit 24 transfers the
X_kCalculate the power level of (n) by the above equation (7)
Then, the data is transferred to the S / N ratio estimation unit 27 and the noise power estimation unit 51.
Is done. In the noise power estimating unit 51, the transferred P
_{X, k}noise power Pav using (n) _{N, k}(n) is estimated
It is. FIG. 2A is a flowchart of the processing of the noise power estimation unit 51.
Show.

First, in step 61, the time average power level P _{X, k} (n) of the transferred P _{X, k} (n) is calculated by using the above equations (8), (9) and (10). calculate. The average time in this case is, for example, 5 to 6 msec. Next, at step 62, Pav for a certain period of time.
Take a histogram of the level distribution of _{X, k} (n). That is, P
Add 1 to the number of power sections to which av _{X, k} (n) belongs, that is, h _k (int Pav _{X, k} (n)) = h _k (int Pav _{X, k} (n)) + 1 (16) . int (*) indicates that the decimal part is truncated to an integer. Further, in step 63, the histogram h _k (i)
Are detected and stored. That is, i that satisfies h _k (i ′) ≦ h _k (i) (17) with respect to values before and after is obtained. The smallest i of the i is the noise power Pav _{N, x} (n). That is, when a plurality of peak values are obtained, the peak value of the minimum value is determined by the noise power Pav _{N, k} (n)
This is transferred to the S / N ratio estimation unit 27.

The S / N ratio estimator 27 uses 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. hand,
In the method shown in FIG. 4A, SNR _k (n) ′ and SNR _k (n)
Is estimated. SNR estimated by S / N ratio estimating section 27
_k (n) is transferred to the gain factor calculation unit 30 and the input signal addition rate determination unit 52. Gain factor calculator 3
Depending on the calculation method used at 0, SNR _k (n) ′ is also input.

In the gain factor calculating section 30, the S / N
Using the SNR _k (n) transferred from the ratio estimating unit 201 and, if necessary, the SNR _k (n) ′, the gain factor G
(SNR _k (n)) is determined. Here, the specific calculation of the gain factor uses the method shown in FIG. The gain factor G (SNR _k (n)) estimated by the gain factor calculation unit 30 is transferred to the gain factor insertion unit 28. Note that the method of the S / N ratio estimating unit 27 and the method of the gain factor calculating unit 30 may be methods other than those described in this specification.

The gain factor insertion unit 28 performs noise reduction using the gain factor G (SNR _k (n)) calculated by the gain factor calculation unit 30. That is, the band signal X _k transferred from the frequency band dividing unit 22
The operation of equation (18) is performed on (n). Y ′ _k (n) = G (SNR _k (n)) × X _k (n) (18) The noise-reduced signal Y ′ _k (n) is input to the input signal adder 5.
Transfer to 3.

The input signal addition rate determination section 52 uses the transferred SNR _k (n) to determine the input signal addition rate α ′ based on the S / N ratio. α ′ desirably takes a small value when SNR _k (n) is large. For example, 15 ≦ 10 log ₁₀ [SNR _k (n)] ⇒ α ′ = 0 5 <10 log ₁₀ [SNR _k (n)] <15 ⇒ α ′ = 0.10 log ₁₀ [SNR _k (n)] <5 ⇒ α '= 0.3 (19) Further, in this example, the frequency band dividing unit 2
In FIG. 2, the input signal addition rate α is obtained by multiplying α ′ by a band division number factor β determined according to the band division number M. For example, α = β × α ′ 1024 ≦ M⇒β = 0.01 M <1024 → β = 1 (20) That is, when the number M of band divisions is large, the addition rate α is reduced. The input signal addition rate α is determined by the input signal addition unit 53
Is forwarded to

In the input signal adding section 53, the band signal X _k (n) transferred from the frequency band dividing section 22, the gain factor inserting section 28, and the input signal addition rate determining section 52, the signal Y ′ _k with reduced noise (n) and outputs the band output signal Y _k (n) using the input signal addition rate α. That is, Y _k (n) = α × X _k (n) + (1−α) × Y ′ _k (n) (21) Y _k (n) is transferred to the time domain conversion unit 29, where the entire area is synthesized and then converted to the time domain. The result is converted into an analog signal by the D / A converter 34 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 ms.
ec, the background noise level was set in a variable reverberation room of 46 dB (A). The number M of band divisions was set to 64 (this is a preferable number of divisions in the subband echo canceller). The evaluation was performed on 30 subjects. The created stimulus sound was played from a speaker, and the quality was evaluated by the subject as the received voice in a public conference. 8d
B, 6 dB, and 1 dB, three types of noise-added voices with S / N ratios were processed by the four types of noise reduction schemes shown in FIG.

FIG. 2B (a) shows the change in the MOS value in each noise reduction method. Comparing before and after the noise reduction processing, when the S / N ratio is good, almost
While the MOS value is higher than that of the noise-added voice (original voice) by the processing, when the S / N ratio is poor, the MOS value is lowered. This is considered to be because when the number of band divisions is small, voice distortion due to processing and degradation due to residual noise are more severe than quality improvement by noise reduction, and the MOS value is lowered. From this fact, when the S / N ratio is poor, not only does the conventional method not improve the sound quality due to noise reduction,
It turns out that it is decreasing.

FIG. 2B (b) shows the result of adding the original sound to the noise reduction sound according to the present invention. here,
Α was set to 0.3 regardless of the S / N ratio. When the original sound is added when the S / N ratio is poor, the MOS value is increased by approximately 0.5 to 1 as compared with before the addition, and as a result, the sound quality is improved by the noise reduction processing. Further, it can be understood that the addition rate α is changed according to the S / N ratio, and when the S / N ratio is good, the sound quality can be more effectively improved by reducing the addition rate α.

The gain factor may be inserted as shown in FIG. 3, that is, by the equation (13), and Y _k (n) ′ may be obtained from the equation (14). As can be understood from the above embodiment, from the input signal power for each band and the noise power for each corresponding band, the ratio of noise to the target audio signal in that band is estimated, and the target voice in that band is estimated. What is necessary is just to estimate the signal amplitude as a noise-reduced signal.

[0028]

As described above, the present invention divides a signal in which the target voice and noise are mixed into frequency bands, estimates the ratio of the noise signal to the target signal for each band, and based on the estimation result, The noise is reduced every time, but in order to reduce the distortion of the target sound that occurs when the number of frequency band divisions is small, the original signal is added more to the noise-reduced sound according to the ratio of the noise to the sound as the noise increases. As a result, noise-reduced speech with little distortion can be obtained. At this time, the smaller the number of band divisions, the larger the ratio to be added.

The noise reduction for the purpose of listening according to the method of the present invention makes it possible to obtain an easy-to-listen target signal with a small delay. As a result, even in a voice communication system such as a voice conference or a TV conference, even if ambient noise other than the target voice is mixed in a transmission signal received by the microphone and transmitted to the other party, the method of the present invention can be used. The reduction of the noise makes it possible to maintain the clarity of the voice, and the communication quality is improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a functional configuration of a noise reduction device to which an embodiment of the present invention is applied.

2A is a flowchart illustrating the processing performed by the noise power estimating unit 51 in FIG. 1, FIG. 2B is a diagram illustrating the result of a subjective evaluation of the noise reduction effect according to the conventional method, and FIG. 2B is a diagram illustrating the noise reduction effect according to the present invention; It is a figure showing the result of subjective evaluation.

FIG. 3 is a block diagram showing a functional configuration of a device to which a conventional noise reduction method is applied.

FIG. 4A is a flowchart of a process in a gain factor calculation unit, and FIG. 4B is a diagram illustrating the principle of a noise reduction device.

FIG. 5 is a diagram showing a calculation formula of a typical gain factor.

Claims (3)

[Claims] 1. A noise reduction method for outputting an audio signal obtained by removing a noise signal from an input signal, comprising: a band dividing step of dividing the input signal into a plurality of frequency bands; An input signal power calculating step of calculating power, a noise power calculating step of calculating noise power in the input signal divided into the respective bands, an input signal power of each of the bands, and a noise power of each of the bands. An estimation step of estimating the ratio of noise to the target audio signal for each band from the above, and estimating the amplitude of the target audio signal for each band as a signal with low noise, and the estimated target voice A step of determining an addition ratio of the noise-reduced signal for each band and the input signal divided into the respective bands, based on a signal-to-noise ratio; and Adding the noise-reduced signal for each band and the input signal divided into each band based on the addition ratio, and outputting a band output signal; Combining a band signal with a band signal. 2. The noise reduction method according to claim 1, wherein the estimation step uses the input signal power for each of the bands and the noise power for each of the bands to generate the target audio signal pair for each of the bands. Estimating a noise ratio; calculating a gain factor for each of the bands based on the target audio signal-to-noise ratio; and calculating each of the bands with respect to the input signal divided into the respective bands. A gain factor inserting step of inserting the gain factor determined for each band to obtain the noise-reduced signal for each band. 3. The noise reduction method according to claim 1, wherein the number of frequency bands to be divided in the band division step is:
A band division type noise reduction method, wherein an addition ratio of the noise-reduced signal and the input signal divided into each band is determined for each band.