JP3205463B2 - Noise reduction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is applied to a fixed or mobile voice communication system which is sometimes used in an environment having a large background noise, and determines a filter coefficient from voice statistics and noise statistics. The present invention also relates to a noise reduction method for filtering an input signal using the filter coefficient to reduce noise superimposed on original speech.

[0002]

2. Description of the Related Art A conventional noise reduction method is based on the premise that the statistical characteristics of noise are known in advance.
D. Gibson (“Filtering of Co
lored Noise for Speech En
Hancement and Coding ”, IEEE
E Trans. SP. Vol. 39, no. 8,19
91) et al. This method includes a Kalman filtering process for each frame. The Kalman filtering process requires statistics of speech and noise. The method of Gibson et al. Is a method in which the statistic of speech for each frame is estimated as unknown for each frame, and the statistic of noise for each frame is known and filtering is performed. Therefore, the estimation accuracy of the noise statistic assumed to be known affects the effect of noise reduction.

[0003]

However, it is not very easy to estimate an accurate noise statistic in advance. It is necessary to preliminarily record only the noise signal in a state where the influence of the voice is eliminated, and estimate the statistics of the noise. There is a disadvantage that the trouble of performing noise recording processing before a call is increased, which leads to an increase in system cost. In addition, in an application area such as mobile radio, the noise statistic estimated in advance may fluctuate with time. In this case, the estimation accuracy of the noise statistic estimated in advance decreases with time. As a result, it also has the disadvantage that the noise cannot be sufficiently reduced.

[0004] In order to follow the temporal variation, a method of having a microphone for noise recording simultaneously with the microphone for voice recording, that is, a method of having a noise recording system of a different system from the voice recording system can be considered. By using this method, the estimation accuracy of the noise statistic is improved, but there is a disadvantage that the system cost is increased. Also, the conventional noise reduction method does not consider the state of the input signal, that is, the input signal is only noise, only voice, the state where noise is superimposed on the voice, the state where voice rises, the state where voice falls, etc. However, there is a disadvantage that the sound may be unnaturally felt.

An object of the present invention is to provide a noise reduction method capable of sufficiently reducing noise. Another object of the present invention is to provide a noise reduction method for sufficiently reducing noise by improving the estimation accuracy of a noise statistic without increasing the price. It is still another object of the present invention to provide a noise reduction method for enhancing the noise reduction effect by adaptively processing according to the state of voice and the state of noise.

[0006]

According to the present invention, a silent section in an input signal is detected, and a statistic of noise is estimated from the input signal in the detected silent section. That is, in a two-way telephone call or the like, when attention is paid to one of the speakers, there is a certain amount of time during which no sound is emitted (the time during which the other party is talking). This state is a state in which only noise is input. In this state, the statistics of noise can be estimated with high accuracy.

Further , the states of voice and noise in the input signal are divided into at least four predetermined states, and each state is stored as a frame state table, and the input signal is stored at predetermined time intervals (frames). The processing is performed to determine the state of each frame with reference to the frame state table, and the speech statistics and noise statistics are adaptively modified according to the determined frame states.

According to a second aspect of the present invention, the frame state table according to the first aspect of the present invention is prepared, noise is recorded by a noise microphone, and a noise statistic is estimated from an input signal recorded in the noise microphone. On the other hand, the state of voice and noise is determined for each frame from both input signals of the microphone for voice and the microphone for noise, and the statistic of voice and the statistic of noise are determined by referring to the frame state table according to the state. Is modified adaptively.

According to the third aspect of the present invention, the frame state table in the first aspect of the present invention is prepared, and a noise statistic is estimated before a call, and the noise statistic is estimated from the noise statistic and an input signal. The input signal is processed at regular time intervals (frames) using the determined voice statistics and the state of the frame is determined with reference to the frame status table, and the voice statistics are determined according to the determined frame status. Adaptively modify volume and noise statistics.

[0010] According to the invention 請 Motomeko 4 obtains speech, the signal-to-noise ratio from the estimated statistics of the noise in any one of the above, depending on the signal-to-noise ratio, or in response to said noise power Control adaptive modification.

According to a seventh aspect of the present invention, in any one of the first to fourth aspects, each of the first n samples of each frame (n is an integer of 2 or more and smaller than the number N of samples in one frame) The filter coefficients are updated for each sample, and the filter coefficients are not updated for the remaining (N−n) samples.

[0012]

EXAMPLES showing a processing procedure of embodiment of the invention of claim 1 in FIG. 1, illustrating a noise reducing apparatus to which the embodiment is applied in FIG. The input signal from the input terminal 11 is divided for a predetermined time (frame), for example, every several tens of milliseconds (S ₁ ), and the state of each frame is determined by the state determination unit 12 (S ₂ ). That is, the relationship between the voice and the noise in the input signal is shown in FIG.
As shown in the figure, there is no background noise and only voice exists, state 10 where neither background noise nor voice exists, complete silence state 0, only background noise exists state 20, background noise exists and voice is present. The state is divided into a speech head or tail state 21 and a stationary state 22 of speech with background noise. These states are given codes and stored in the frame state table 13 in advance.

The input signal makes such a state transition as shown in FIG. 3B. This state is determined by the audio power in the input signal and its fluctuation, the audio spectrum characteristic and its fluctuation, and the like. When the result of the state determination of each frame is state 20 in which only noise exists (S20)
_{In 3} ), the noise statistic estimating unit 14 estimates noise statistics, for example, an AR model and power of the noise, from the input signal at that time (S ₄ ). In states other than the state 20, the voice statistic estimation unit 15
In audio statistics, for example, it estimates the AR model and power of the voice (S _5).

The estimated noise statistics and speech statistics are adaptively modified by the correction units 16 and 17 according to the determined state of the frame (S
₆ ). This correction is performed, for example, as shown in FIG. That is, in the state 20, in order to perform the noise reduction processing strongly, the power value of the audio signal is set to a value smaller than the power of the input signal (power of the noise). The coefficient a multiplied by the power of the input signal is set to a value of 1 or less, for example, 0.1, and the power of the voice with respect to the noise power is reduced so that the strong noise reduction processing is performed by the Kalman filter. Also, in this case, the AR model of the voice is blunted, and the AR model and the power of the noise used at that time are used as they are.

In a state 21, the noise power is multiplied by a coefficient of 1 or less, for example, 0.7 in the immediately preceding state 20, that is, the noise power is reduced to weaken the noise reduction processing. The AR model and the power of the voice used in the frame are used as they are, and the AR model of the noise used in the previous state 20 is used as it is. In state 22, the voice uses the AR model and power estimated in the frame as it is, and the noise uses the AR model and power estimated in state 20 immediately before, that is, performs noise reduction processing of normal strength.
That is, in the prior art, the estimated statistics of speech and noise were used as they are, but in this embodiment, in state 20 other than state 22, the estimated power of the speech is modified small to enhance the noise reduction processing, and in state 21, the noise reduction processing is performed. , The estimated statistic is adaptively modified according to the state so that the estimated power is corrected to be small to weaken the noise reduction processing.

A filter coefficient is calculated using the statistics of speech and noise adaptively modified as described above (S
_7), and sets the filter coefficient in the Kalman filter 18, an input signal from the terminal 11 to filtering processing by the filter 18 for each corresponding frame suppresses a noise component in the input signal (S _8). The signal filtered by the filter 18 is voice-encoded and transmitted or stored as needed.

As described above, since the statistic of the noise is estimated in the silent section, that is, in the state 20, it is possible to realize a highly accurate estimation, there is no need to perform the estimation of the statistic of the noise in advance, and only one recording system is required. When the noise state changes, the noise statistic corresponding thereto is estimated, so that the noise can be reduced satisfactorily. Also, the degree of noise reduction is increased, weakened, or adaptively changed according to the state, so that noise reduction is effectively performed.

The number of states and the definition of the states are not limited to the above example. For example, in the above-described embodiment, the speech head and the speech tail are defined as the state 21 together. However, the speech head is the state 21 and the speech tail is the state 23. In the state 23 as shown in FIG. The coefficient by which the power is multiplied is gradually reduced, for example, 0.9, 0.89, 0.8
Multiply by gradually smaller values such as 8, 0.87..., 0.7. By this operation, a sound, that is, a state 21 or 22
From state to silence, that is, state 20, the same state 20
By controlling the multiplier a in accordance with the transition of the state even if it is determined that in this example, the noise is gradually reduced in this example, and a natural feeling in audibility can be obtained.

FIG. 7 shows an embodiment in which the state transition is further complicated. The number of states is increased and the state transition is diversified as compared with the embodiment shown in FIG. In this embodiment, the configuration shown in FIG. 7 is employed to prevent erroneous noise reduction processing when no noise is superimposed. FIG.
8 and 9 show the conditions of the state transition in. The explanation of each variable in FIGS. 8 and 9 is shown below.

P: Average power per sample of the input signal. Ratio: Assuming that the average power of the current frame is P and the average power of the immediately preceding frame is P ′, Ratio = P / P ′ st-cnt: Transition control auxiliary variable. The default value is 8. In the transition condition (a1, b1, d1, e1, f1, g1, j1), it is reduced by one. In the number (e1), it is initialized to 8.

Ps1, Pn, Psp: constants. Threshold for average power of input voice. Ps1 = 3.0, Pn = 640.0, Psp = 2500.
0 spg: Variable. Dimensionless number. The initial value is 1.0. Transition condition (i2) and 'P ≦ Psp or Ratio ≦ 0.3 or P
If ratio ≧ 1.0 / 0.3 'holds 10 times in a row,
Let psg = 3.0 × Pmin / Psp. Set to 1.0 in the transition condition (c1).

Pmin: Pmin = MIN (Pmin, P) (number (i2) and 'P ≦ Psp or Pratio ≦ 0.3 or Pratio ≧ 1.
0 / 0.3 'is satisfied) Pmin = positive maximum value (other cases) Pow10: a variable having a power dimension. The initial value is 0.0.
In the number (e2), Pow10 = 0.9 × Pow10 + 0.1 ×
The processing of P is performed.

Pow10cnt: Variable. The initial value is 0. In the number (e2), Pow10cnt = Pow10cnt + 1 (Pow10> 2.0 ×
In the case of Pn), the processing of Pow10cnt = 0 (in the case of Pow10 ≦ 2.0 × Pn) is performed. Pr: Pr = P / Pow20 Pow20: Pow20 = P (under condition d1), Pow20 = 0.9 ×
Pow20 = 0.1 × P (at condition 11), Pow20 = Min
(P) (when st22 is repeated 50 times) K: K = (1.0−ref1 × ref1) × (1.0−ref
2 × ref2). Where ref1 and ref2
Is the k parameter when the input is passed through a second-order linear prediction filter.

Var: residual power when the input is passed through a second-order linear prediction filter. sstt: takes one of the values 0, 2, 4, 6, 8, and 10. Transition conditions (k1, k2, l1, m1, o1, o2,
In (p1, q1, s1, t1, t2), a value is obtained based on the following formula. Where Var 'is the value of Var of the immediately preceding frame, and Ratio = Var' / Var.
Is defined by

When sstt = 0 or sstt = 2: 0.6 <Ratio <1.4: sstt = sstt + 2 0.5 <Ratio ≦ 0.6 or 1.4 ≦ Ratio <1.5: sstt =
2 Ratio ≦ 0.5 or 1.5 ≦ Ratio: sstt = 2 When sstt = 4: 0.7 <Ratio <1.3: sstt = 6 0.6 <Ratio ≦ 0.7 or 1.3 ≦ Ratio <1.4: sstt =
4 0.5 <Ratio ≦ 0.6 or 1.4 ≦ Ratio <1.5: sstt =
2 Ratio ≦ 0.5 or 1.5 ≦ Ratio: sstt = 0 When sstt = 6: 0.8 <Ratio <1.2: sstt = 8 0.7 <Ratio ≦ 0.7 or 1.2 ≦ Ratio <1.3: sstt =
6 0.6 <Ratio ≦ 0.7 or 1.3 ≦ Ratio <1.4: sstt =
4 0.5 <Ratio ≦ 0.6 or 1.4 ≦ Ratio <1.5: sstt =
2 Ratio ≦ 0.5 or 1.5 ≦ Ratio: sstt = 0 When sstt = 8 or sstt = 10: 0.9 <Ratio <1.1: sstt = 10 0.8 <Ratio ≦ 0.9 or 1 .1 ≦ Ratio <1.2: sstt =
8 0.7 <Ratio ≦ 0.7 or 1.2 ≦ Ratio <1.3: sstt =
6 0.6 <Ratio ≦ 0.7 or 1.3 ≦ Ratio <1.4: sstt =
4 0.5 <Ratio ≦ 0.6 or 1.4 ≦ Ratio <1.5: sstt =
2 Ratio ≦ 0.5 or 1.5 ≦ Ratio: sstt = 0 States 10 and 11 in FIG. 7 are states where the background noise is estimated to be sufficiently small. By making the transition conditions from these states to other states (20 or more) sufficiently strict, it is possible to avoid performing a noise reduction filter operation erroneously despite the absence of background noise.

Normally, the coefficients of the Kalman filter 18 are calculated for each sample. The amount of processing required to calculate this coefficient is larger than the filtering operation itself. on the other hand,
Since the values of the AR model coefficient and the power used for calculating the filtering coefficient take a constant value for each frame (for example, the number N of samples in one frame is 160), the filtering coefficient converges to a constant value in the latter half of the frame. Therefore, instead of updating the filter coefficients for every sample, the filter coefficients are updated for the first n samples of the frame, for example, three samples, and the obtained coefficients are used to perform a filtering process. (N-n)
For the sample, the filter coefficient is not updated, and filtering is performed using the coefficient obtained for the n-th sample.
Thus, the amount of calculation and the processing time can be reduced.

In the above description, the coefficient a for multiplying the input signal power
Controls adaptively according to the estimated noise power and voice power levels. That is, when the noise power is small or the input SRN (signal-to-noise ratio) is high, the direction is set so that the Kalman filter 18 operates strongly as described above.
When the noise power is large or the input SRN is low, the Kalman filter 18 is set to operate in a weaker direction. If strong noise reduction is performed when the SNR of the input is low, the unnaturalness of the audio signal increases, and the coefficient a for avoiding this is adaptively controlled. Input SNR
For example, as shown in FIG. 10A, the coefficient a changes linearly between a large value at a small SNR and a small value at a large SNR as shown in FIG. 10A.

Further, as shown in FIG. 10B, when the amplitude of the input signal is equal to or smaller than a predetermined value, that is, equal to or less than 1 / α of the maximum amplitude (for example, when the maximum amplitude is 1, α = 1000), the input signal amplitude is reduced. After the value is set to zero, the processing shown in FIG. 2 may be performed. That is, when the input signal is at a low level, it may be regarded as noise and may be suppressed. FIG. 10C
As shown in the figure, the power of the input signal is a predetermined value β (for example, β /
If the maximum input power is 30 dB or less, the input signal may be attenuated according to the power. That is, after performing the same processing as that of the so-called expander, the processing illustrated in FIG. 2 may be performed. A temporal control (using a time constant) may be applied to the attenuation by the expander.
This expander processing is performed, for example, in the Slavic compander described in "Application and Effect of Sirabic Compander in 800 MHz Band Car Telephone System", Vol. 30, No. 3 (1981), p. A time constant may be provided like the panda, and the attack time may be, for example, about 3.0 ± 0.1 mS, and the recovery time may be about 1.35 ± 0.6 mS.

The various signal processings shown in FIG. 10 are not limited to the case where the level (ie, the amplitude or the power) of the input signal itself is used, but the power of the estimated noise statistic or the estimated speech statistic. Is also good. Further, these signal processings are shown in FIG.
As shown in FIG. 11A, the signal is supplied to the noise reduction unit 32 shown in FIG.
The various processes shown in FIG. 10 may be performed on the output of the noise reduction unit 32 as shown in FIG.

The signal processed by the noise reduction means 32 shown in FIG. 2 may be encoded and output by the speech encoding means 33 in FIG. 11C. For example, V
SELP, PSI-CELP (IEICE RC593-7)
8-1993 November "Pitch Synchron
ous Innovation CELP (PSI-C
ELP) ”).

As shown in FIG. 11D, between the noise reduction means 32 and the encoding means 33, the processing means 31 shown in FIG.
May be inserted. As shown in FIG. 11E, a noise reduction unit 32 may be provided on the output side of the audio decoding unit 34 corresponding to the audio encoding unit 33. Although not shown in the figure, in FIG. 11E, a signal processing unit 31 may be provided before or after the noise reduction unit 32.

[0032] In FIG. 1, a statistic correction of Step S ₆ may be omitted, which is the embodiment of the invention of claim 1. The silent interval whether the check in step S _3, and the noise statistics estimation step S ₄ is omitted in FIG. 1, prior to the call, performs the estimation of the noise statistics, using the estimated statistics, The statistics may be modified adaptively according to the state of each frame. This is an embodiment of the fourth aspect of the present invention, and is particularly effective when the noise state does not fluctuate greatly as in fixed communication. Furthermore, a noise microphone is provided in addition to the audio microphone, a voice statistic is estimated from an input signal from the audio microphone, a noise statistic is estimated from an input signal of the noise microphone, and various types of noise are estimated from both input signals. A state or a state transition may be detected, and the statistics may be adaptively modified accordingly. This is the third aspect of the present invention.

In FIG. 10 and FIG.
This applies not only to the invention of claim 1 but also to the invention of claim 2 or 3 . The filtering process is not limited to the Kalman filter process.

[0034]

As described above, according to the first aspect of the present invention, the trouble of estimating the noise statistic prior to a call is saved, and one recording system is sufficient, and the system cost is reduced. In addition, when the noise state fluctuates with time as in the case of mobile communication, the statistic of the noise is estimated following the fluctuation, and the noise is reduced more satisfactorily.

According to the second aspect of the present invention, the noise reduction is enhanced or weakened according to the state of the input signal every short time, so that the audibility is improved. 請 Motomeko 7 calculation amount is reduced processing time according to the invention of is increased. FIG. 6 shows the relationship between the S / N ratio of the input signal (before noise reduction) and the S / N ratio of the output signal (after noise reduction) when the invention of claim 1 is applied. From this figure, it can be seen that the S / N ratio of the output signal is always higher than the S / N ratio of the input signal, and especially when no input signal is present, that is, when the background noise alone is used, the application effect is large.

As described above, according to the present invention, the effect of noise reduction is great. Therefore, when performing various kinds of coding on a speech signal, the noise reducing processing according to the present invention is performed prior to coding, so that good quality can be obtained. An encoded audio signal can be obtained.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an embodiment of the invention of claim 1 ;

FIG. 2 is a block diagram showing a noise reduction apparatus to which the invention of claim 1 is applied.

3A is a diagram illustrating an example of a definition of a state of an input signal, and FIG. 3B is a diagram illustrating a state transition of an input signal;

FIG. 4 is a view showing an example of correction of an estimated statistic in each state according to the invention of claim 1 ;

FIG. 5 is a diagram showing another example.

FIG. 6 shows the S of the input signal when the invention of claim 1 is applied.
FIG. 4 is a diagram showing an example of the relationship between / N and the S / N of an output signal.

FIG. 7A is a diagram showing another example of the definition of the state of the input signal;
B is a diagram showing another example of the state transition of the input signal.

FIG. 8 is a diagram showing conditions for state transition in FIG. 7;

FIG. 9 is a view showing a continuation of FIG. 8;

FIG. 10 is a view showing an example of characteristics of various signal processings used before and after the noise reduction processing.

11A and 11B are block diagrams each showing a combination of a noise reduction unit and a signal processing unit, and FIG.
It is a block diagram which shows the combination with a decoding means.

Continuation of front page (56) References JP-A-62-294315 (JP, A) JP-A-5-49054 (JP, A) ROBERT J. McAULAY and MARILYN L. MARP ASS; "Speech Enhancement Using a Soft-Decision Noise Suppression Filter", IEEE TRANSACTIONS ON ACOUSTICS, SPEECH H, AND SIGNAL GROUP. ASSP-28, NO. 2 (April 1980) p. 137-145 Sharon Gannot, David Burshtein and Weinstein; 6, NO. 4 (July 1998) p. 373-385 (58) Fields surveyed (Int.Cl. ⁷ , DB name) H04B 1/10-1/14 H04B 15/00 H03H 17/00-21/00

Claims (7)

(57) [Claims] 1. A speech statistic is estimated from an input signal, a filter coefficient is calculated using the statistic of the speech and a statistic of noise in the input signal, and the input signal is calculated based on the filter coefficient. In the noise reduction method of performing the filtering process, the state of the voice and the noise in the input signal, the state where only the voice exists, the state where only the background noise exists, the state where the background noise and the voice are the beginning or end of the speech, Background noise is present and speech is at least divided into steady states. Each state is stored in a frame state table, and the input signal is processed for each fixed time (frame), and the frame state table is referred to. The state of each frame is determined by the following procedure. If the determined frame state is a state where background noise is present and the sound is in a steady state, the sound is
The power of the voice estimated in the frame of
Is the noise estimated when only the background noise
If the sound power is used as is and only background noise is present, the power value of the sound is
Correct the power estimated in that frame to a smaller value
The noise uses the power estimated in the frame as it is.
If the background noise and the speech are at the beginning or end of the speech, the speech will be
Using the power estimated in that frame as it is, the noise is
Noise estimated when only the background noise immediately before exists
To weaken the noise reduction processing is used to fix small power
Adaptively the statistics of the speech and the statistics of the noise
A noise reduction method characterized by correcting . 2. A statistic of speech is estimated from an input signal recorded by a microphone for speech, and a statistic of noise is estimated from an input signal recorded by a microphone for noise. A noise reduction method for calculating a filter coefficient using the statistic and filtering the input signal using the filter coefficient, wherein a state of the voice and the noise in the input signal is determined based on a state where only the voice exists and a background noise. Only, the background noise and speech are at the beginning or end of the speech, and the background noise is present and the speech is in a steady state, and each state is stored in the frame state table. The input signal is processed at fixed time intervals (frames), and the state of each frame is determined by referring to the above-mentioned frame state table. If the background noise is present and the sound is in a steady state, the sound
The power of the voice estimated in the frame of
Is the noise estimated when only the background noise
If the sound power is used as is and only background noise is present, the power value of the sound is
Correct the power estimated in that frame to a smaller value
The noise uses the power estimated in the frame as it is.
If the background noise and the speech are at the beginning or end of the speech, the speech will be
Using the power estimated in that frame as it is, the noise is
Noise estimated when only the background noise immediately before exists
To weaken the noise reduction processing is used to fix small power
Adaptively the statistics of the speech and the statistics of the noise
A noise reduction method characterized by correcting . 3. A noise reduction method for estimating voice statistics from an input signal, calculating a filter coefficient using the voice statistics and noise statistics, and filtering the input signal using the filter coefficients. In the above, the noise statistic is estimated from an input signal in a silent section before a call, the states of voice and noise in the input signal are divided into predetermined states, and each state is stored as a frame state table. The state of speech and noise in the input signal is defined as follows: a state in which only speech is present, a state in which only background noise is present, a state in which background noise and speech are at the beginning or end of speech, and a state in which background noise is present. The voice is at least divided into a steady state, and each state is stored in a frame state table, and the input signal is processed at regular time intervals (frames). Referring to determine the status of each frame, the determined frame state, if the steady state sound exists background noise, audio its
The power of the voice estimated in the frame of
Is the noise estimated when only the background noise
If the sound power is used as is and only background noise is present, the power value of the sound is
Correct the power estimated in that frame to a smaller value
The noise uses the power estimated in the frame as it is.
If the background noise and the speech are at the beginning or end of the speech, the speech will be
Using the power estimated in that frame as it is, the noise is
Noise estimated when only the background noise immediately before exists
To weaken the noise reduction processing is used to fix small power
Adaptively the statistics of the speech and the statistics of the noise
A noise reduction method characterized by correcting . 4. A correction amount of said adaptive correction is adaptively controlled so that said filter processing is emphasized when a signal-to-noise ratio is high or noise power is low. The noise reduction method according to any one of the above. 5. The noise reduction method according to claim 1, wherein the input signal or the output signal is a predetermined value or less, and the input signal or the output signal is attenuated according to the level. . 6. The noise reduction method according to claim 5, wherein temporal control is performed on the attenuation. 7. In the first n samples of each frame (n is an integer of 2 or more and smaller than the number N of samples in one frame), the filter coefficient is updated for each sample, and the remaining (N−n−n) 7. The noise reduction method according to claim 1, wherein the filter coefficient is not updated in the sample.