JP5333307B2 - Noise estimation method and noise estimator - Google Patents

Description

  The present invention relates to a noise estimation method and a noise level estimator for estimating a level change over time such as background noise in an echo canceller, for example.

FIG. 9 is a functional block diagram showing a conventional noise estimator.
The conventional noise estimator includes a short-term power estimator 1, a long-term power estimator 2, and a comparison / determination unit 3. The short-term power estimator 1 is constituted by an integration circuit or a digital filter having a characteristic that the rise is steep and the fall is gentle, and estimates the short-time power of the input signal e.

  The long-term power estimator 2 is constituted by an integration circuit or a digital filter having a characteristic that the rise is gradual and the fall is steep, and estimates the long-term average power of the input signal e.

The comparison / determination unit 3 compares the ratio between the instantaneous power estimation value obtained by the short-term power estimator 1 and the average power estimation value obtained by the long-term power estimator 2 with a predetermined threshold value, so that it is included in the reference signal. The background noise that exists constantly, for example, the noise level is estimated.
-Output signal of short-term power estimator 1: Pow_S
-Output signal of long-term power estimator 2: Pow_L
・ Threshold: Pow_TH
・ Noise level estimate: BNL
Then,
(Pow_S / Pow_L) ≦ Pow_TH (A)
Here, 1.0 <Pow_TH
The input signal is assumed to be only a noise signal,
BNL = Pow_S
Is output as an estimate of the noise level.
Pow_TH <Pow_S / Pow_L (B)
In this case, it is estimated that the input signal includes a signal component such as speech other than noise, and the value of Pow_L immediately before the estimated noise level is shifted from Equation (A) to Equation (B) is output as BNL. To do.

  Patent Document 1 discloses a method of learning a noise pattern as a method corresponding to a noise level change. Patent Document 2 discloses a method for specifying only a noise signal from an input signal including speech other than noise.

JP 2000-330587 A JP 2009-153053 A

  However, in the conventional noise estimator, in an environment where the noise of the input signal changes with time, the long-term power estimator 2 follows and stabilizes the change in the signal level of the noise. Since it takes time, there is a problem that the noise level cannot be estimated correctly during that time.

  Patent Document 1 discloses a method for learning a noise pattern as a method for dealing with a change in noise level in order to solve the above-described problem. There is a problem that a large amount of database memory is required.

  In order to solve the above-described problem, Patent Document 2 discloses a method for specifying only a noise signal from an input signal including speech other than noise. However, for implementation, a plurality of input devices (for example, There was a problem that a microphone) was required.

  The present invention has been made in view of such problems, and it is possible to appropriately follow and estimate a change in noise level even in an environment where background noise increases or decreases over time with a simple device configuration. It is an object of the present invention to provide a noise estimation method and a noise estimator that enable the above.

  The noise estimation method of the present invention samples a signal to be processed and / or a noise signal, inputs this as an input signal, calculates a short-time power average of the input signal, and outputs a short-term power estimation value. A noise estimation method using an estimator and a level change detector that monitors the increase and decrease of the short-term power estimation value to estimate the noise level of the noise signal. And a power estimation process for calculating the short-time power average of the input signal as the discrete time elapses to obtain the short-term power estimated value.

  Furthermore, in the noise estimation method of the present invention, the level change detector monitors the increase / decrease of the short-term power estimation value at each time, and when the short-term power estimation value increases / decreases within a predetermined range, The signal is determined to be only the noise signal, and the short-term power at each time is determined by a temporary noise level holding process for holding the short-term power estimation value at that time as a temporary noise level and the level change detector. When an increase or decrease in the estimated value is monitored and the short-term power estimated value increases and the number of increases reaches a first threshold, it is determined that the input signal includes the signal to be processed, and the provisional noise Noise level estimation processing for estimating the level as the noise level of the noise signal.

  The noise estimator of the present invention samples a signal to be processed and / or a noise signal and inputs it as an input signal, and a short-time power average of the input signal with the lapse of discrete time for each sampling period And a short-term power estimator that outputs as a short-term power estimate.

  Furthermore, the noise estimator of the present invention inputs the short-term power estimation value, monitors increase / decrease of the short-term power estimation value at each time, and when the short-term power estimation value increases / decreases within a predetermined range The input signal is determined to be only the noise signal, the short-term power estimate at that time is held as a provisional noise level of the noise signal, the short-term power estimate increases, and the number of increases When the signal reaches the first threshold, the input signal includes a level change detector that determines that the signal to be processed includes the signal to be processed and estimates the provisional noise level as a noise level of the noise signal. Yes.

  The noise estimation method and noise estimator of the present invention have the following effects (1) to (3).

  (1) The increase / decrease in the short-term power estimation value is monitored, and when the short-term power estimation value increases / decreases within a predetermined range, it is determined that the input signal is only a noise signal, and the short-term power estimation value at that time is When the short-term power estimation value increases and the number of increases reaches the first threshold, it is determined that the input signal includes a signal to be processed, and the provisional noise level is maintained. Since the level is estimated as the noise level of the noise signal, it is possible to appropriately estimate the noise level in a short time even in an environment where the noise changes over time.

  (2) Since it comprised as said (1), the database which memorize | stores a noise pattern is unnecessary, and it becomes possible to estimate a noise level appropriately, without requiring a lot of memory.

  (3) Since it comprised as said (1), it becomes possible to estimate a noise level appropriately with a simple hardware constitution.

FIG. 1 is a functional block diagram showing the noise estimator of FIG. 2 in Embodiment 1 of the present invention. FIG. 2 is a block diagram showing an outline of the noise estimator in Embodiment 1 of the present invention. FIG. 3 is a circuit diagram showing a schematic configuration of the low-pass filter in FIG. FIG. 4 is a circuit diagram showing a configuration of the noise estimator in FIG. FIG. 5 is a waveform diagram showing an input signal and an output signal of the short-term power estimator in FIG. FIG. 6 is a time chart showing a noise estimation method using the noise estimator of FIG. FIG. 7 is an explanatory diagram showing state transitions of the noise estimator of FIG. FIG. 8 is a flowchart showing a noise estimation method using the noise estimator of FIG. FIG. 9 is a functional block diagram showing a conventional noise estimator.

  Modes for carrying out the present invention will become apparent from the following description of the preferred embodiments when read in light of the accompanying drawings. However, the drawings are only for explanation and do not limit the scope of the present invention.

(Configuration of Example 1)
FIG. 2 is a configuration diagram illustrating an outline of the noise estimator according to the first embodiment of the present invention.

  In FIG. 2, the noise estimator samples a signal to be processed (for example, an audio signal) and / or a noise signal and inputs the sampled signal as an input signal e, which is input with the lapse of discrete time t for each sampling period. A short-term power estimator 10 that calculates a short-time power average of the signal e and outputs it as a short-term power estimated value Pow_S is provided. The short-term power estimated value Pow_S changes its value over time to form an envelope.

  A level change detector JDG 20 is connected to the output side of the short-term power estimator 10.

  The short-term power estimator 10 receives an absolute value circuit ABS11 that inverts the negative polarity portion of the input signal e and converts it into the absolute value signal abs, and the absolute value signal abs, and outputs the absolute value at the first time t1. A low-pass filter (hereinafter referred to as “LPF”) that calculates a power average between the value signal abs and the short-term power estimated value Pow_S at the second time t2 that is a sampling period before the time t1 and outputs the result as a new short-term power estimated value Pow_S .) 12. The LPF 12 is composed of, for example, an infinite impulse response (hereinafter referred to as “IIR”) type digital filter.

  The level change detector JDG 20 inputs the short-term power estimated value Pow_S at time t1, calculates the power ratio ΔP with the short-term power estimated value Pow_S_OLD at time t2 immediately before time t1, and calculates the power ratio ΔP and the third It has a function of comparing the threshold value ΔX and the fourth threshold value ΔY to determine the increase / decrease state of the short-term power estimation value Pow_S of the input signal e.

  Further, the level change detector JDG 20 has a small increase / decrease in the short-term power estimation value Pow_S of the input signal e. When the increase / decrease change remains within a predetermined range, the input signal e includes only a noise signal. The temporary noise level BNL_X is updated and held with the short-term power estimation value Pow_S at that time. The number of increases in the short-term power estimation value Pow_S is counted for each sampling period, and when the number of increases reaches the first threshold value M, it is determined that the signal to be processed is included in the input signal e and held. The provisional noise level BNL_X has been estimated as the noise level BNL of the noise signal.

FIG. 3 is a circuit diagram showing a schematic configuration of the low-pass filter 12 in FIG.
The LFP 12 includes a multiplier 12-a whose coefficient is 1-X. The multiplier 12-a has an input terminal IN, and an adder 12-b is connected to the output side of the multiplier 12-a. An output terminal OUT and a delay element 12-c are connected to the output side of the adder 12-b. A multiplier 12-d whose coefficient is X is connected to the output side of the delay element 12-c, and an adder 12-b is connected to the output side of the multiplier 12-d.

  The absolute value signal abs is input to the input terminal IN at time t1, the multiplier 12-a multiplies the absolute value signal abs and the coefficient 1-X, and the result is input to the adder 12-b. The The delay element 12-c holds the short-term power estimated value Pow_S_OLD at the time t2 immediately before the time t1, and is input to the multiplier 12-d at the time t1.

  In the multiplier 12-d, the short-term power estimated value Pow_S_OLD is multiplied by the coefficient X, and the result is input to the adder 12-b. In the adder 12-b, the absolute value signal abs × (1-X) and the short-term power estimated value Pow_S_OLD × X are added and output as a new short-term power estimated value Pow_S from the output terminal OUT. As a result, a weighted average of the absolute value signal abs and the short-term power estimated value Pow_S_OLD at time t2 is output as a new short-term power estimated value Pow_S.

  FIG. 1 is a functional block diagram showing the noise estimator of FIG. 2 in Embodiment 1 of the present invention.

  The short-term power estimator 10 outputs a short-term power estimated value Pow_S, and the level change detector JDG 20 inputs this.

  The level change detector JDG 20 includes a power ratio calculation unit 30, an input signal state determination unit 40, a provisional noise level update unit 50, a level decrease count unit 60, a level increase count unit 70, and a noise level output unit 80. And provisional noise level update prohibition canceling means 90.

  The power ratio calculation means 30 receives the short-term power estimation value Pow_S output from the short-term power estimator 10 at each time, and calculates the short-term power estimation value Pow_S at time t1 and Pow_S_OLD at time t2 immediately before time t1. It has a function of calculating the power ratio ΔP. The input signal state determination unit 40 receives the power ratio ΔP, and the short-term power estimation value Pow_S of the input signal e is in an increasing state, a decreasing state, or a steady state depending on the magnitude of the power ratio ΔP. It has a function to determine whether.

  When the input state determination unit 40 determines that the state is an increase state, the level increase count unit 70 increments the level increase counter UpCNT to set an update prohibition state of the provisional noise level BNL_X, and the level increase counter UpCNT is set to the threshold value M. It is configured to determine whether or not it has been reached.

  The level increase counting means 70 further determines whether or not the level decrease counter DnCNT is equal to or smaller than a fifth threshold value (for example, “1”) when the input state determination means 40 determines that the increase state is present. When it is, it is determined that the short-term power estimated value Pow_S has increased within a predetermined range, and the slight increase determination unit 70a that clears the level decrease counter DnCNT to 0 is provided.

  The level decrease counting unit 60 is configured to increment the level decrease counter DnCNT when the input state determining unit 40 determines that the level is decreasing, and to determine whether or not the level decrease counter DnCNT has reached the second threshold value N. have.

  The level decrease counting unit 60 further determines whether or not the level increase counter UpCNT is 1 when the input state determining unit 40 determines that the level is decreasing, and when it is 1, the short-term power estimated value Pow_S is determined. Is slightly decreased within a predetermined range, and a slight decrease determination unit 60a for clearing the level increase counter UpCNT to 0 is provided.

  The provisional noise level update unit 50 determines that the input signal e includes only a noise signal when the input state determination unit 40 determines that the input signal e is in a steady state. At this time, when the provisional noise level BNL_X is prohibited from being updated, the provisional noise level BNL_X is updated and held by the short-term power estimation value Pow_S input at the time.

  The noise level output means 80 is configured to output the provisional noise level BNL_X as the noise level BNL when the level increase counter 70 determines that the level increase counter UpCNT has reached the threshold value M. Temporary noise level update prohibition canceling means 90 has a function of clearing level increase counter UpCNT to 0 and canceling the temporary noise level BNL_X update prohibition state when level decrease counter DnCNT reaches threshold value N in level decrease count means 60. have.

FIG. 4 is a circuit diagram showing a configuration of the noise estimator in FIG.
FIG. 4 is a circuit diagram in which the noise estimator of FIG. 1 is realized by an individual circuit, and includes a short-term power estimator 10 and a level change detector JDG 20. The level change detector JDG 20 compares a plurality of memories 31, 53, and 82 that store variables such as a short-term power estimation value Pow_S, a level increase counter UpCNT72 and a level decrease counter DnCNT, and threshold values such as ΔX and ΔY and input data. In addition, a plurality of selectors 41, 42, 51, 63, 64, 73, 81 for selecting subsequent paths, a plurality of adders 61, 71 and a divider 32 are configured. Each selector 41, 42, 51, 63, 64, 73, 81 is configured by a comparator.

  In FIG. 4, an arrow whose arrowhead is a black triangle represents a data signal, an arrow whose arrowhead is> represents a control signal, and On attached to the arrow indicating the control signal is on the start point side of the arrow. It shows that the selector controls and operates the selector on the end point side of the arrow.

  The level change detector JDG 20 includes a memory 31 that stores the short-term power estimated value Pow_S_OLD input at time t2 immediately before time t1. When the short-term power estimation value Pow_S is newly input, the level change detector JDG 20 is configured to store the short-term power estimation value Pow_S in the memory 31 as the short-term power estimation value Pow_S_OLD.

  The divider 32 is connected to the memory 31 and the short-term power estimator 10, and is configured to receive the short-term power estimated value Pow_S and the short-term power estimated value Pow_S_OLD and obtain a power ratio ΔP that is a ratio thereof. ing.

  On the output side of the divider 32, a selector 41 and a selector 42 are connected in parallel. The selector 41 has a function of selecting the subsequent processing by comparing the power ratio ΔP and the threshold value ΔX.

  One end of the output terminal of the selector 41 is connected to a selector 74 for determining whether or not the level decrease counter DnCNT is 1 and an adder 71 for adding 1 to the level increase counter UpCNT. A level increase counter UpCNT and a selector 73 are connected to the adder 71. The selector 73 has a function of determining whether or not the level increase counter UpCNT has reached the threshold value M. On the output side of the selector 73, a selector 81 for comparing the noise level upper limit value BNL_max of the noise signal with the provisional noise level BNL_X and a level decrease counter DnCNT are connected.

  On the output side of the level increase counter UpCNT, an adder 71, a selector 64 for determining whether or not the level increase counter UpCNT is 1 and a selector 51 for determining whether or not the level increase counter UpCNT is 0 are connected. The output side of the selector 64 is connected to the level increase counter UpCNT. A level decrease counter DnCNT is connected to the output side of the selector 74, and a selector 74 and an adder 61 that adds 1 to the level decrease counter DnCNT are connected to the output side of the level decrease counter DnCNT.

  The other end of the output terminal of the selector 41 is connected to a selector 42 that compares the power ratio ΔP with the threshold value ΔY. A selector 64 and an adder 61 are connected to one end on the output side of the selector 42. The output side of the adder 61 is connected to a level decrease counter DnCNT and a selector 63 that determines whether or not the level decrease counter DnCNT has reached the threshold value N. A level increase counter UpCNT is connected to the output side of the selector 63. A selector 51 is connected to the other end of the selector 42, and a switch 52 is connected to the output side of the selector 51.

  A memory 53 for storing the provisional noise level BNL_X is connected to the output side of the switch 52, and a selector 81 for comparing the provisional noise level BNL_X and the noise level upper limit value BNL_max is connected to the output side of the memory 53. Yes. A memory 82 for storing the noise level BNL is connected to the output side of the selector 81.

(Noise estimation method of Embodiment 1)
A noise estimation method using the noise estimator of the first embodiment will be described with reference to FIG.

  At time t1, the input signal e is input to the short-term power estimator 10, and the short-term power estimator 10 calculates a short-time power average of the input signal e and outputs it as a short-term power estimated value Pow_S. The memory 31 stores the short-term power estimated value Pow_S_OLD input at time t2 immediately before time t1. When the short-term power estimated value Pow_S is newly input, the level change detector JDG 20 stores the short-term power estimated value Pow_S in the memory 31 as the short-term power estimated value Pow_S_OLD.

  The short-term power estimate value Pow_S and the short-term power estimate value Pow_S_OLD are input to the divider 32, and the power ratio ΔP is obtained. The power ratio ΔP is input to the selector 41 and compared with the threshold value ΔX. When ΔP ≧ ΔX, the selector 74 determines whether DcCNT is 1. When it is 1, the level decrease counter DnCNT is cleared to 0.

  Further, the adder 71 adds 1 to the level increase counter UpCNT, and the selector 73 determines whether or not the level increase counter UpCNT has reached the threshold value M. When the threshold value M is reached, the selector 81 compares the noise level upper limit value BNL_max with the provisional noise level BNL_X. When BNL_X ≦ BNL_max, the value of the provisional noise level BNL_X is stored in the memory 82 as the noise level BNL. When BNL_X> BNL_max, the noise level upper limit value BNL_max is stored in the memory 82 as the noise level BNL.

  In the selector 41, when ΔP <ΔX, the selector 42 compares the power ratio ΔP with the threshold value ΔY, and when ΔP <ΔY, the selector 64 determines whether or not the level increase counter UpCNT is 1. When it is 1, the level increase counter UpCNT is cleared to 0. Further, the adder 61 adds 1 to the level decrease counter DnCNT, and the selector 63 determines whether or not the level decrease counter DnCNT has reached the threshold value N. When the level decrease counter DnCNT reaches the threshold value N, the level increase counter UpCNT is cleared to zero.

  In the selector 42, when ΔY ≦ ΔP, the selector 51 determines whether the level increase counter UpCNT = 0. A switch 52 is connected to the output side of the selector 51. As a result of the determination in the selector 51, when the level increase counter UpCNT = 0, the switch 52 is turned on and the short-term power estimated value Pow_S is stored in the memory 53 as the provisional noise level BNL_X.

  5A and 5B are waveform diagrams showing the input signal e and the output signal of the short-term power estimator 10 in FIG.

  The horizontal axes in FIGS. 5A and 5B are the number of samplings, which corresponds to 8000 samplings = 1 second. 5A and 5B, the vertical axis represents the signal level X, which corresponds to 20 log (X / 5564) [dBm]. That is, X = 5664 corresponds to 0 [dBm]. For example, X = 2000 corresponds to −9 [dBm], X = 1000 corresponds to −15 [dBm], and X = 500 corresponds to −21 [dBm]. The unit [dBm] is expressed in decibels based on 1 [mW], and is expressed by 10 log (power 1 [W] / 1 [mW]). For example, 1 [mW] = 0 [dBm], 10 [mW] = 10 [dBm], 100 [mW] = 20 [dbm], 1 [W] = 30 [dBm].

  In FIG. 5A, the input signal e has a shape of a vertical object centered on 0 on the vertical axis, and the amplitude is changed in small increments. In the vicinity of 0 on the vertical axis, a band-like waveform with a substantially constant amplitude extends in the horizontal axis direction. This band-like waveform corresponds to a portion below the broken line A in FIG. 5B and indicates background noise. For example, in FIG. 5B, the noise level BNL is 50, which corresponds to 20 log (50/5564) = − 41.08 [dBm].

FIG. 6 is a time chart showing a noise estimation method using the noise estimator of FIG.
In FIG. 6, the horizontal axis represents discrete time at regular intervals. The uppermost short-term power estimated value Pow_S row indicates the level of the short-term power estimated value Pow_S at each time inputted to the level change detector JDG 20.

  The next BNL row indicates the noise level BNL, and its initial value is zero. The next ΔP row is a graph showing whether ΔP has increased, is steady, or has decreased as a result of calculation in the level change detector JDG 20. The next line level increase counter UpCNT line is a graph showing the increase or decrease of the level increase counter UpCNT. The next line level decrease counter DnCNT line is a graph showing the increase or decrease of the level decrease counter DnCNT.

  For example, the short-term power estimated value Pow_S increases from time t1 to t2. Therefore, the increase in ΔP rows is plotted, and the level increase counter UpCNT is incremented by one. When the short-term power estimated value Pow_S decreases at time t3, the level decrease counter DnCNT is incremented by 1 and the level increase counter UpCNT is 1, so this is cleared to 0. Similarly, when the short-term power estimated value Pow_S increases at time t4, the level increase counter UpCNT is added and the level decrease counter DnCNT is 1, so it is cleared to 0. These operations prevent the counter values of the level increase counter UpCNT and the level decrease counter DnCNT from being accumulated and reaching the threshold value M or the threshold value N while the short-term power estimation value Pow_S repeats slight increase and decrease. It is out.

  Thereafter, the short-term power estimated value Pow_S repeats a slight increase and decrease, and when a steady state occurs at time t5, the level change detector JDG20 is in a state where the provisional noise level BNL_X is prohibited from being updated because the level increase counter UpCNT is 0. The short-term power estimated value Pow_S is updated and held as the temporary noise level BNL_X by recognizing that it has been canceled. That is, at this time, since the level increase counter UpCNT is 0, it is determined that the input signal is not yet in the rising phase and includes only the noise signal.

  An increase state occurs from time t6, the level increase counter UpCNT is incremented by 1, and an update prohibition state of the provisional noise level BNL_X is set. When a steady state occurs at time t7, the level change detector JDG 20 recognizes that the update prohibition state of BNP_X is set because the level increase counter UpCNT is not 0, and does not update the provisional noise level BNL_X. . Thereafter, every time an increase state occurs, the level increase counter UpCNT is incremented by 1. At time t8, the level increase counter UpCNT reaches the threshold value M, the level change detector JDG20 clears the level decrease counter DnCNT to 0, and is temporarily The value of the noise level BNL_X is stored as the noise level BNL. Thereafter, each time the increase state occurs, the level increase counter UpCNT is incremented by one.

  On the other hand, if the decrease state frequently occurs after time t9, the count value of the level decrease counter DnCNT sequentially increases. When the level decrease counter DnCNT reaches the threshold value N at time t10, the level change detector JDG20 determines that the input signal e has passed the decreasing phase, clears the level increase counter UpCNT to 0, and updates the provisional noise level BNL_X. Release the prohibited state.

  When the steady state occurs at time t11, the level change detector JDG 20 recognizes that the update prohibition state of the provisional noise level BNL_X has been canceled because the level increase counter UpCNT is 0, and the short-term power estimated value at that time Pow_S is updated and held as the provisional noise level BNL_X. Similarly, when the steady state occurs at time t12, the level change detector JDG 20 recognizes that the update prohibition state of the provisional noise level BNL_X is released because the level increase counter UpCNT is 0, and the short-term at that time point. The power estimated value Pow_S is updated and held as the provisional noise level BNL_X.

  At time 13, the level increase counter UpCNT reaches the threshold value M, the level change detector JDG20 clears the level decrease counter DnCNT to 0, and stores the value of the provisional noise level BNL_X as the noise level BNL.

FIG. 7 is an explanatory diagram showing state transitions of the noise estimator of FIG.
The level change detector JDG 20 takes in the short-term power estimation value Pow_S from the short-term power estimator 10 and calculates a power ratio ΔP with the short-term power estimation value Pow_S_OLD which is the output result immediately before the short-term power estimator 10.

  When the calculation result is ΔX ≦ ΔP, it is determined in step S4 that the input power has increased, and in step S32, the value of the level increase counter UpCNT is incremented by one. In step S5, in order to suppress the fluctuation of the counter, when the level decrease counter DnCNT is updated immediately, that is, when the level decrease counter DnCNT is 1, the value of the level decrease counter DnCNT is cleared to zero.

  When the level increase counter UpCNT reaches the threshold value M, the value of the level decrease counter DnCNT is cleared in step S5, and the value of the temporary noise level BNL_X is stored as the noise level BNL in step S9. However, when the provisional noise level BNL_X exceeds the noise level upper limit value BNL_max, the noise level upper limit value BNL_max is set to BNL. Note that the noise level upper limit value BNL_max is appropriately determined according to the environment to which the noise estimator is applied.

  If the result of the calculation is ΔY ≦ ΔP <ΔX, it is determined in step S13 that it is in a steady state. If the value of the level increase counter UpCNT is 0, the short-term power estimated value Pow_S at this time is updated as the provisional noise level BNL_X. And hold.

  If the result of calculation is ΔP <ΔY, it is determined in step S15 that the input power has decreased, and in step S17, the value of the level decrease counter DnCNT is incremented by one. In step S19, in order to suppress the fluctuation of the counter, when the level increase counter UpCNT is updated immediately, that is, when the level decrease counter DnCNT is 1, the value of the level increase counter UpCNT is cleared to zero. When the level decrease counter DnCNT reaches the threshold value N, the value of the level increase counter UpCNT is cleared in step S19, and the provisional noise level BNL_X update prohibition state is cancelled.

  FIG. 8 is a flowchart showing a noise estimation method using the noise estimator in FIG. 4, and steps common to those in FIG. 7 are denoted by common reference numerals.

When the power to the noise estimator is turned on, provisional noise level BNL_X = 0, level increase counter UpCNT = 0, and level decrease counter DnCNT = 0 are set as initial values. Furthermore, ΔX, ΔY, M, and N which are threshold values are
ΔY ≦ 1.0 <ΔX, 2 ≦ M, 2 ≦ N
It is preset to be

  In step S1, the absolute value circuit ABS11 in the short-term power estimator 10 converts a signal having a positive polarity out of the input signal e into a signal obtained by inverting a signal having a negative polarity to a positive polarity. As a result, the input signal e is converted into an absolute value signal abs. The absolute valued signal is output as a short-time envelope of the input signal e by the LPF 12 having a characteristic that the rise is steep and the fall is gentle.

  In step S2, the level change detector JDG 20 takes the result of the short-term power estimator 10 as the short-term power estimation value Pow_S, and the power ratio ΔP = the output result of the immediately preceding short-term power estimator 10 and the short-term power estimation value Pow_S_OLD. Short-term power estimated value Pow_S / short-term power estimated value Pow_S_OLD is calculated. After calculating the power ratio ΔP, the short-term power estimated value Pow_S_OLD is updated to the short-term power estimated value Pow_S for the calculation of the power ratio ΔP in the next processing.

  When ΔX ≦ ΔP in step S3, the process proceeds to step S4. In step S4, when the level decrease counter DnCNT is updated immediately in order to suppress the counter fluctuation, that is, when the level decrease counter DnCNT = 1. (YES) In step S5, the value of the level decrease counter DnCNT is cleared to zero. When the level decrease counter DnCNT ≠ 1 (NO), the process proceeds to step S6. In step S6, the value of the level increase counter UpCNT is incremented by 1 and updated.

  When the level increase counter UpCNT reaches the threshold value M in step S7 (YES), the value of the level decrease counter DnCNT is cleared in step S8, and the value of the provisional noise level BNL_X is set to the noise level BNL in step S9. . In step S10, the noise level BNL is compared with the noise level upper limit value BNL_max. If the noise level BNL exceeds the noise level upper limit value BNL_max, the noise level upper limit value BNL_max is set as the noise level BNL in step S11. The process ends. Note that the noise level upper limit value BNL_max is appropriately determined according to the environment to which the noise estimator is applied. In step S7, when the level increase counter UpCNT ≠ M (NO), this process ends.

  If ΔP <ΔX in step S3, the process proceeds to step S12, where the power ratio ΔP and the threshold value ΔY are compared. When ΔY ≦ ΔP, it is determined as a steady state, and the process proceeds to step S13. When the level increase counter UpCNT = 0 in step S13 (YES), the provisional noise level BNL_X is prohibited from being updated. In S14, the short-term power estimation value Pow_S is updated and held as the provisional noise level BNL_X, and this process is terminated. When the level increase counter UpCNT ≠ 0 (NO), this process is terminated.

  If ΔP <ΔY in step S12, it is determined that the state is decreasing, and the process proceeds to step S15. In step S15, immediately after the level increase counter UpCNT is updated, that is, when the level increase counter UpCNT = 1 (YES), the value of the level increase counter UpCNT is cleared to 0 in step S16 in order to suppress the counter fluctuation. Then, the process proceeds to step S17. When the level increase counter UpCNT ≠ 1 (NO), the process proceeds to step S17.

  In step S17, 1 is added to the value of the level decrease counter DnCNT. In step S18, when the level decrease counter DnCNT reaches the threshold value N (YES), in step S19, the value of the level increase counter UpCNT is cleared to 0 to cancel the update prohibition state of the provisional noise level BNL_X and perform this processing. finish. When the level decrease counter DnCNT is not equal to the threshold value N (NO), this process ends.

(Effect of Example 1)
According to the first embodiment, there are the following effects (1) to (3).

  (1) The short-term power estimator 10 calculates a short-time power average of the input signal e as time elapses, and outputs a short-term power estimated value Pow_S. The level change detector JDG 20 inputs this to calculate the time Pow_S increase / decrease is monitored every time, and when the increase / decrease in Pow_S is very small, it is determined that the input signal e is only a noise signal, Pow_S at that time is held as the provisional noise level BNL_X, and Pow_S increases and increases When the number of times reaches the threshold value M, it is determined that the input signal e includes a signal to be processed, and BNL_X is estimated as the noise level BNL. For this reason, it is possible to appropriately estimate the noise level BNL in a short time even in an environment where the noise changes with time.

  (2) Since it is configured as described in (1) above, it is possible to appropriately estimate the noise level BNL without requiring a database for storing noise patterns and without requiring a large amount of memory.

  (3) Since it comprised as said (1), it becomes possible to estimate the noise level BNL appropriately with a simple hardware configuration.

(Modification)
The present invention is not limited to the above-described embodiments, and various usage forms and modifications are possible. For example, the following forms (a) to (d) are used as the usage form and the modified examples.

  (A) The present invention can be applied to control of coefficient update of an applied filter provided in an echo canceller by appropriately estimating a change in the background noise level when background noise is superimposed on the audio signal. .

  (B) Although the noise estimator of the present invention is configured by the circuit shown in FIG. 4 in the first embodiment, the level change detector JDG 20 is executed by program control using a digital signal processor or the like. May be.

  (C) In the first embodiment, a noise signal with respect to an audio signal is assumed. However, the present invention can be applied to, for example, electromagnetic induction noise from the outside in an electronic circuit.

  (D) In the first embodiment, the means for realizing the envelope output has been described with respect to the IIR LPF 12, but the present invention is not limited to this. For example, it may be a finite impulse response LPF, or may be an integration circuit or the like having a characteristic that the rise is steep and the fall is gentle.

10 Short-term power estimator 11 Absolute value circuit (absolute value circuit ABS)
12 Low-pass filter (LPF)
20 Level change detector JDG
30 power ratio calculation means 40 input signal state determination means 50 provisional noise level update means 60 level decrease count means 70 level increase count means 80 noise level output means 90 provisional noise level update prohibition release means

Claims (10)

A short-term power estimator that samples a signal to be processed and / or a noise signal, inputs this as an input signal, calculates a short-time power average of the input signal, and outputs a short-term power estimate;
A level change detector for monitoring the increase and decrease of the short-term power estimate to estimate the noise level of the noise signal;
A noise estimation method using
The short-term power estimator calculates a short-time power average in the input signal with the lapse of discrete time for each sampling period to obtain the short-term power estimate, and
The level change detector monitors the increase / decrease of the short-term power estimated value at each time, and when the short-term power estimated value increases / decreases within a predetermined range, the input signal is only the noise signal. A provisional noise level holding process for determining and holding the short-term power estimate at that time as a provisional noise level;
The level change detector monitors the increase / decrease of the short-term power estimate at each time, and when the short-term power estimate increases and the number of increases reaches the first threshold, the input signal is: A noise level estimation process for determining that the signal to be processed is included, and estimating the provisional noise level as a noise level of the noise signal;
A noise estimation method comprising: Sampling a signal to be processed and / or a noise signal and inputting it as an input signal, calculating a short-time power average of the input signal with the lapse of discrete time for each sampling period, and a short-term power estimation value A short-term power estimator that outputs as
The short-term power estimation value is input to monitor the increase / decrease of the short-term power estimation value at each time. When the short-term power estimation value increases / decreases within a predetermined range, the input signal is only the noise signal. The short-term power estimate at that time is held as a provisional noise level of the noise signal, the short-term power estimate increases, and when the number of increases reaches a first threshold, A level change detector that determines that the input signal includes the signal to be processed and estimates the provisional noise level as a noise level of the noise signal;
A noise estimator characterized by comprising: The short-term power estimator is
An absolute value conversion circuit that inverts a negative polarity portion of the input signal and converts it into an absolute value signal;
The absolute value signal is input, and the power average of the absolute value signal at the first time and the short-term power estimated value at the second time immediately before the first time is calculated. A filter that outputs as a new short-term power estimate;
The noise estimator according to claim 2, comprising: The level change detector is
The short-term power estimation value output from the short-term power estimator at each time is input, and a power ratio between the short-term power estimation value at the first time and the short-term power estimation value at the second time is calculated. Power ratio calculation means for calculating,
Input signal state determination means for inputting the power ratio and determining whether the input signal is in an increasing state, a decreasing state, or a steady state according to the magnitude of the power ratio;
A level increase counter, and when the input state determination means determines that the increase state, the count value of the level increase counter is incremented and the count value of the level increase counter reaches the first threshold value Level increase counting means for determining whether or not,
A level decrease counter, and when the input state determination means determines that the decrease state, the count value of the level decrease counter is incremented to determine whether the count value of the level decrease counter has reached a second threshold value Level decrease counting means for determining whether or not,
In the input state determination means, when the steady state is determined, the input signal is determined to be only the noise signal, and the provisional noise level is determined by the short-term power estimation value input at the time at that time. Provisional noise level updating means for updating and holding
In the level increase counting means, when it is determined that the count value of the level increase counter has reached a first threshold value, a noise level output means for outputting the provisional noise level as the noise level;
The noise estimator according to claim 2 or 3, characterized by comprising: The input signal state determination means includes
When the power ratio is input and the power ratio is greater than or equal to a third threshold, the input signal is determined to be in the increased state, and the power ratio is less than a fourth threshold that is smaller than the third threshold. Sometimes it is determined that the input signal is in the decreasing state, and when the power ratio is greater than or equal to a fourth threshold and less than the third threshold, it is determined that the input signal is in the steady state. The noise estimator according to claim 4. The provisional noise level update means
When the input state determination means determines that the steady state is present and the count value of the level increase counter is 0, the input signal is only the noise signal, and the update prohibition state of the provisional noise level is released. The provisional noise level is updated and held by the short-term power estimation value inputted at the time at that time, and when the count value of the level increase counter is not 0, the input signal is 6. The noise estimator according to claim 4, wherein the noise estimator includes the processing target signal and determines that the update of the provisional noise level is prohibited, and does not update the provisional noise level. The noise level output means includes
A noise level upper limit value, and in the level increase counting means, when the count value of the level increase counter reaches the first threshold value, the temporary noise level is compared with the noise level upper limit value, When the noise level exceeds the noise level upper limit value, the noise level upper limit value is output as the noise level. When the temporary noise level is equal to or lower than the noise level upper limit value, the temporary noise level is set as the noise level. It outputs, The noise estimator of any one of Claims 4-6 characterized by the above-mentioned. The level change detector further includes:
In the level decrease counting means, when the count value of the level decrease counter reaches the second threshold value, the count value of the level increase counter is cleared to 0 to cancel the update prohibition state of the provisional noise level. The noise estimator according to claim 4, further comprising provisional noise level update prohibition canceling means. The level increase counting means further includes:
When the input state determination means determines that the increase state, it determines whether or not the count value of the level decrease counter is less than or equal to a fifth threshold, and when the count value is less than or equal to the fifth threshold, A slight increase determination unit that determines that the power estimation value has increased in the predetermined range and clears the count value of the level decrease counter to 0;
The level decrease counting means further includes:
In the input state determination means, when it is determined that the decrease state, it is determined whether or not the count value of the level increase counter is less than or equal to the fifth threshold, and when it is less than or equal to the fifth threshold, 9. The slight decrease determination unit that determines that the power estimation value has decreased within the predetermined range and clears the count value of the level increase counter to 0, 9. Noise estimator.   The noise estimator according to claim 2, wherein the processing target signal is an audio signal.

Images