JP2650208B2 - Echo canceller - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller applied to a telephone line such as a loudspeaker and a satellite communication.

(Prior Art) Conventionally, as this kind of echo canceller, there is one having a configuration applied to a loudspeaker as disclosed in, for example, Japanese Patent Application Laid-Open No. 61-56526. FIG. 8 is a block diagram showing a configuration example of this conventional echo canceller. In the figure, 800 is an echo canceller, and 802 is a loudspeaker. In this apparatus, in a loudspeaker 802, an audio signal R _out output from a speaker 804 is input to a microphone 806 as an echo path (Echo Path, hereinafter sometimes referred to as EP) in a room, and a transmission signal is output. Echo signal S
_in (k) and leak the echo canceller (Echo C
anceller. Hereinafter, it may be referred to as EC. ) 800 to delete. 808 is a speaker amplifier, and 810 is a microphone amplifier.

The echo canceller 800 controls an adaptive digital filter (Adaptive Digital Filer; hereinafter, also referred to as ADF) 812 that receives a received signal and outputs a pseudo echo signal S _in (k), and an adaptive estimation operation of the ADF 812. A Double Talk Detector (hereinafter referred to as DTD) 814 and an addition that subtracts the pseudo echo signal S _in (k) from the echo signal S _in (k) to generate a residual signal R _es (k). It is mainly composed of the container 816. In addition, here, 818, 82
0 is an A / D converter, 822 and 824 are D / A converters, and k
Represents a sampling time synchronized with a synchronous clock pulse of 8 kHz, for example.

Adaptive estimation operation of ADF812 only the received signal R _in (k) is present, only the single-talk does not exist near-end talker's signal N, because it is performed correctly, the received signal R _in the (k) Kintanhanashi In the double talk state where the user signal N is present at the same time, the DTD 814 for inhibiting the adaptive estimation operation of the ADF 812 is indispensable.

By the way, in the technique disclosed in the above-mentioned Japanese Patent Application Laid-Open No. 61-56526, the detection sensitivity of DTD814 is increased. FIG. 9 is a block diagram showing a DTD of this conventional configuration. In this configuration, 901 to 903 and 923 are peak value detectors, and 904 to 90
6 is a square circuit, 907 and 908 are priority encoders, 9
09 is AT memory, 910, 918, 924, 925 are adders, 911, 91
2, 916 is a comparator, 913, 914, and 917 are switches, 915 is a shift circuit, 919 and 920 are limiters, and 921 and 922 are correction amount memories. , The respective signal levels L _Rin (k) and L _Res (k) of the received signal R _in (k) and the residual signal R _es (k) are obtained in the logarithmic domain, and this level difference is smaller than the threshold value AT. At this time, double talk is detected, and the ADF adaptive operation inhibition signal INH = 1 is output. In this case, the threshold value AT is controlled as in non-double talk (INH = 0) AT (k + 1) = AT (k) + δD in double talk (INH = 1) AT (k + 1) = AT (k) −δD, AT (k) ≧ L _Rin (k) −L _Res (k) is established by the fluctuation of EP, and even if a double talk is erroneously detected, the threshold AT decreases with the passage of time, so that AT (k) <L _Rin (K) −L _Res (k), and the ADF 812 has the characteristic of starting the adaptive estimation operation again.

(Problems to be Solved by the Invention) However, in the case of the echo canceller having the conventional configuration, when a sudden echo path fluctuation occurs and howling occurs, AT (k) ≧ L _Rin (k) −L _Res (K), the adaptive operation of the ADF is prohibited and the howling state continues until the threshold value AT decreases gradually and AT (k) <L _Rin (k) −L _Res (k) holds. There was a serious drawback.

An object of the present invention is to eliminate such conventional disadvantages,
An object of the present invention is to provide an echo canceller that suppresses the occurrence of howling slightly and has excellent speech quality.

(Means for Solving the Problems) In order to achieve this object, according to the present invention, there is provided an adaptive digital filter for generating a pseudo echo signal by inputting a received signal, and converting the pseudo echo signal from the echo signal. In an echo canceller configured to perform echo cancellation by subtraction, a silent detector that detects a silent state of a received signal, a double talk detector that detects a double talk state, and a howling detector that detects a howling state Based on the detection results from the silence detector, the double talk detector, and the howling detector, control is performed to perform adaptive estimation at least when the received signal is in a sound state and howling occurs, and Digital Filter with Control to Prohibit Adaptive Estimation in the Talking State and Double Talk An echo canceller including a control unit, wherein the howling detector identifies an echo signal, a residual signal obtained by subtracting a pseudo echo signal from the echo signal, or a received signal (hereinafter, also referred to as a howling signal). A frequency component detection unit that detects that the frequency of the received signal is an emphasized signal; and a signal level detection unit that detects that the signal level of the reception signal or the transmission signal exceeds a predetermined threshold. It is characterized by comprising.

(Operation) As described above, according to the present invention, each of the howling detector, the received signal silence detector, and the double talk detector generates a detection signal of one logical value, for example, “1” when detected, and the other signal when not detected. A detection signal having a logical value of, for example, "0" is generated, and these detection signals are sent to the ADF control unit. Based on the results of these detection signals, an ADF adaptive estimation operation prohibition signal is output to the ADF, and the ADF adaptive estimation is performed. It controls the prohibition of operation and its release.

Therefore, during howling, the AD
Since F performs adaptive estimation, howling stops immediately,
Therefore, the speech quality is not impaired, and the speech quality equivalent to the conventional one is guaranteed by the double talk detection.

Further, in the present invention, since the howling detector and the double talk detector provided separately with the howling and the double talk which are difficult to distinguish from each other are detected, there is no mutual detection of the howling and the double talk. Therefore, the above-described adaptive estimation control can be properly performed, and the above-described adaptive estimation prohibition or cancellation control can be properly performed.

Further, in the present invention, when detecting a howling state, the howling signal is a signal in which a specific frequency is emphasized, and the signal level of a reception signal or a transmission signal is a signal level exceeding a predetermined threshold. The howling state is detected based on the two characteristics. For this reason, in the present invention, the howling detector is constituted by a frequency component detecting section for detecting the characteristic of the howling detector, and a signal level detecting section for detecting the characteristic of the howling detector. As a result, in the present invention, the circuit configuration of the howling detector can be simplified. Therefore, it is possible to avoid a complicated circuit configuration of the echo canceller combining the howling detector and the double talk detector.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the configuration of the echo canceller of the present invention.

First, an outline of the present invention will be described with reference to FIG. In the figure, 100 is an echo canceller, 102
Is an adaptive digital filter (ADF) that receives a received signal R _in (k), generates a pseudo echo signal _in (k), and performs an adaptive estimation operation, as in the related art. An adder 104 subtracts the pseudo echo signal _in (k) from the echo signal S _in (k) as in the prior art to generate a residual signal.

The echo canceller of the present invention provides a received signal R _in (k)
Silence detector (ID) 10 to detect the presence or absence of silence
6, a double talk detector (DTD) 108 for detecting the presence or absence of a double talk state, a howling detector (HD) 110 for detecting howling, a silence detector 106, a double talk detector 108, and a howling detector 110. And an adaptive digital filter (ADF) control unit 112 for controlling the adaptive operation of the adaptive digital filter (ADF) 102 in response to the detection result from.

Silence Detector This silence detector 106 is, in this embodiment, a reception signal R _in
When (k) is input and a silent state is detected, a detection signal
The J ₁ outputs the one logic value "1", for example, when a non-silent state detection signal J ₁ double-talk detector 108 to the detection signal J ₁ outputs the other logic value "0", for example
And input to the ADF control unit 112.

FIG. 2 is a block diagram showing a specific configuration example of the silence detector 106. In this embodiment, the absolute value circuit 114
And an integrating circuit 116 for the absolute value | R _in (k) | of the received signal R _in (k), and a comparator 118. Received signal R
_in (k) is input to the absolute value circuit 114 and the absolute value | R _in (k)
| = AR _in (k), and then calculate the absolute value
116 is sent to a δ ₁ multiplier 120, where the absolute value AR _in (k) is multiplied by δ ₁ (0 <δ ₁ <1) which is a small positive amount, sent to an adder 122, and this adder 122 Then, the (1-δ ₁ ) multiplier
The output P _Rin (k) is obtained by adding to the output from 124. The output P _Rin (k) of the adder 122 is determined by the comparator 118 and the unit delay element 126.
Sent to The process of obtaining P _Rin (k) of the integration circuit 116 is performed by the following equation: P _Rin (k) = (1−δ ₁ ) P _Rin (k−1) + δ ₁ | R
_Rin (k) | This circuit receives the signal R
Since the integration circuit of the absolute values of the _in (k), by setting [delta] ₁ suitably, approximately (k-20) Toka (k-1
The average amplitude in a short time such as (0), for example, (k-16) hours can be obtained. For example, if you set the [delta] ₁ 2 ^-4 and time for obtaining the average amplitude becomes 2 ⁴ Sample about.
The output P _Rin (k) obtained in this manner is compared with a preset threshold value X _th to determine the presence or absence of a silent state. In this case, if P _Rin (k)> X _th , the detection signal J ₁ = 0 as a reception signal sound, and if P _Rin (k) ≦ X _th , the detection signal J ₁ = 1 as a reception signal silence. and outputs the result as a detection signal J _1.

Note that the configuration of the silence detector described above is not limited to any way above configuration, if it is configured so as to output a detection signal J ₁ which was determined silence and sound.

Next, the double talk detector 108 will be described. Usually, the behavior of double talk and howling can be extremely confusing in terms of signal. Therefore, in the present invention, a detector for detecting whether double talk is single talk is provided separately from the howling detector.

In this embodiment, the double talk detector 108 is connected to the power detector 302 and the logarithmic converter 304 of the received signal R _in (k).
, A power detector 306 and a logarithmic converter 308 of the residual signal R _es (k), an adder 310 of the signals from the log-logarithmic converters 304 and 308, a comparator 312, and a doubler in the comparator 312. double-talk detection threshold value control circuit for setting a talk detection threshold a _t (k) (hereinafter, referred to as a _t control circuit.) 314
It consists of:

The received signal R _in (k) and the residual signal R _es (k) are power-detected by power detectors 302 and 306, respectively.
After obtaining P _x (k) and P _e (k), respectively, they are converted into logarithmic domain values L _x (k) and L _e (k) by log converters 304 and 308 connected at the subsequent stage, respectively. You. Adder
At 310, the difference signal A _com (k) = L _x (k) −L
In one outputs the _e (k) is supplied to the comparator 312 on the other hand supplied to A _t the control circuit 314.

In the comparator 312, in advance by A _t the control circuit 314 (k-
The threshold A _t (k) calculated at the sampling time of 1),
Performing a comparison of the difference signal A _com (k), A _t (k) <A when _com a (k) is the detection signal J ₂ is a logical value such as "0 while indicating that a single-talk state "and, in the case of _{a t (k) ≧ a com} (k) , it serves as the other logic value, for example," 1 "representing the detection signal J ₂ is a double talk state.

The detection signal J ₂ is to output to the ADF controller 112 shown in FIG. 1, it is input as A _t control selection signal to the A _t the control circuit 314.

On the other hand, the A _t control circuit 314 detects the above results J
_2, silence detector 106 (first view) on the basis of the detection result J ₁ input as A _t control prohibition signal, updating of the double-talk detection threshold value A _t (k), a circuit for performing each of the storage and subtraction It is.

Figure 4 is a block diagram showing a specific configuration example of the A _t control circuit 314. The A _t the control unit 314, the averaging circuit 4
02, r adder 404, selector switch 406, adder
And 408, a limiter 410 for outputting a threshold value A _t (k + 1) corresponding to the next sampling time, which is the output signal of the A _t control circuit 314, a delay unit 412 for delaying the output by a unit sampling time, constant 0 or δ ₂ (this δ ₂
Is a constant independent of δ ₁ already described. ), A multiplier 416, and an adder 42.
0.

The difference signal A _com (k) is supplied to the average value circuit 402 and the adder 420, respectively. This averaging circuit 402 generates the signal A
_A long-term average value of _com (k), for example, 128 or 256 sampling intervals is obtained. Then enter the r adder 404 as a signal of the long-time average value, gives a voltage level difference of only r between the threshold A _t (k) by subtracting the r from the long-term mean value, double-talk detection Keep the sensitivity constant.

The output of the r adder 404 is supplied to the selector switch 406 to the A side terminal. Note that a unit delay element 412 is connected to the B-side terminal of the selector switch 406. The selector switch 406 has a detection result J ₁ from the silence detector (ID) 106.
When a detection result J ₂ from the double-talk detector (DTD) has been inputted, it operates in relation as shown in the following Table I, based on the detection result J ₁ and J _2.

Table I J ₁ J ₂ selection terminal 0 0 A 0 1 B 1 0 B 11 B As can be understood from Table I, the selector switch 406 selects the terminal A in the case of sound and single talk. In other states, the terminal B is selected.

The adder 420 outputs the signal A _com (k) and the output of the unit delay element 412.
The difference signal from A _t (k) is output to multiplier 416.

The multiplier 416 supplies the result to a multiplicative constant which is selected by the selector switch 414 (0 or [delta] ₂₎ the difference signal _{A t (k) -A com (} k) to the adder 408.

The adder 408 is a multiplier based on the output of the selector switch 406.
The output of 416 is subtracted and output to the limiter 410. The limiter 410 has a function of threshold A _t (k) limits the value of the threshold A _t (k) so as not to be set unreasonably high or low. The output of this limiter 410 is the updated threshold A
It is output to the unit delay element 412 and the comparator 312 as _t (k + 1).

The selector switch 414 on the basis of the detection result J ₁ and J ₂ as described above, performs the operation shown in the following Table II.

Table II J ₁ J ₂ selection terminal 0 0 0 0 1 δ ₂ 1 0 0 1 1 1 0 (where 0 <δ ₂ <1) As can be understood from Table II, the selector switch 414 has a sound and In the case of double talk, the constant δ ₂ is selected, and in other states, the constant 0 is selected.

This control, A _t the control circuit outputs a double-talk detection threshold value A _t (k) in the following relationship.

The received signal R _in (k) at the time of _{silence: A t (k + 1)} = A t (k) single talk _{time: A t (k + 1)} = A long-term average value during the double-talk of _{com (k): A t (} k + 1) = A _t (k) -δ ₂ (A _t (k) -A
_com Note (k)), the structure of DTD108 described above is merely an example, in the echo canceller of the present invention, any way described above if a construction such DTD108 outputs a detection signal J ₂ was determined doubletalk It is not limited to the configuration described above.

Howling Detector Next, howling detector (HD) 110 will be described.

FIG. 5 is a block diagram showing an example of the configuration of the howling detector 110. FIG. 6 is a block diagram showing a second-order acyclic adaptive prediction filter (hereinafter referred to as a second-order FIR) incorporated as a frequency component detector. FIG.

This howling detector 110 outputs the echo signal y (k) (S _in
(K)) is input to the second-order FIR and 502, coefficients ₁ (k) and ₂ (k) of the second-order FIR 502, a control unit 504, and a control unit 506 of a prediction output control coefficient g (k) of the adaptive prediction filter. And g
(K), ₂ (k) and ADF102 register power MPOW
Is composed of a judgment unit 508 that detects howling from (k), at the time of howling detection, and outputs a detection signal J ₃ as one logic value "1", for example to the ADF controller 112, the detection signal J is otherwise ₃ is the other logical value, for example "0"
Is output to the ADF control unit 112. still,
The determination unit 508 is also a signal level detector that detects whether the signal level of the reception signal or the transmission signal exceeds a predetermined threshold by observing the register power MPOW (k). .

 Next, the principle of howling detection will be described.

Now, echo canceller 100 when howling occurs
If the input / output R _in (k), R _es (k), and S _in (k) are x (k), e (k), and y (k), these x
(K), e (k) and y (k) are almost single sine waves, and the spectrum becomes linear spectral. Second-order recursive filter with transfer function of
Called. Considering the impulse response of ()), Q (Z) = 1− (Z) in order to sequentially adaptively predict this sine wave and make it uncorrelated. May be input to a second-order FIR having a transfer function of

The circuit of the secondary FIR shown in FIG. 6 receives the echo signal y (k) and receives a dummy residual signal from the dummy prediction value _d (k).
adder 602 for outputting r _d (k), and this signal y (k)
Delay element 60 that delays by the unit sampling time
4, a unit delay element 606 for further delaying the delayed y (k), and a signal y−
A multiplier 608 for multiplying (k-1) by a coefficient ₁ (k);
The coefficient is added to the signal y (k−2) from the unit delay element 606
₂ (k), multipliers 610,
An adder 612 that adds the signal from 610 to output a predicted value (k) and a multiplication that multiplies the predicted value (k) by a coefficient g (k) and outputs the dummy predicted value _d (k) described above. vessel
614, and an adder 616 for subtracting the predicted value (k) from the echo signal y (k) and outputting a residual signal r (k). The residual signal r obtained here
(K) and the dummy residual signal r _d are both of FIG. 5
₁ (k), ₂ (k) are sent to the control unit 504 and the g coefficient control unit 506.

In the configuration shown in FIG. 6, an algorithm for adaptively outputting a prediction value (k) and making the residual signal r (k) uncorrelated regardless of howling occurs at any frequency will be described below. explain.

The differential term obtained by differentiating r ² (k) with the coefficient _i (k) of the prediction filter is r (k) = y (k) − (k) Therefore, the method of sequential update is Or, _i (k + 1) = _i (k) + δ · sgn {r (k)} · sgn {y (ki)} where 0 <α <2δ is a positive minute value. Therefore, when A (z) = (z), the input y (k) ≒ the predicted value (k), and the residual signal r (k) is uncorrelated.

Further, if the purpose is to make only r (k) uncorrelated by predicting only a sine wave, it may be ₂ (k). When the root P of 1−A (z) = 0 is represented on the Z plane in polar coordinates, P = r · e ^{± jω} where r: the absolute value of the complex number P ω: the argument of the complex number P e: natural It can be expressed as the base of the logarithm, and by substituting this for 1-A (z) = 0, The reason why the impulse response of the second-order IIR becomes a sine wave is only when r ² = 1.

_{The 1} (k) and ₂ (k) control units 504 are as described above,
The coefficients ₁ (k) and ₂ (k) are controlled based on the following equations.

Next, a predicted value output control coefficient g (k) is introduced as a method of indicating how close the signal is to a sine wave. The coefficient g (k) is adaptively changed to obtain a dummy residual signal r _d (k).
Is given as described below.

r _d (k) = y (k) _-d (k) ... Therefore, Or 0 ≦ g (k) ≦ 1 The g (k) control unit 506 calculates g
(K) is controlled.

As described above, by the two controls, if y (k) is uncorrelated, g (k) 予 測 0, and the predicted value (k) = 0. If the echo signal y (k) is a sinusoidal (howling state), the coefficient _{g (k) ≒ 1 2 (} k) = 1 y (k) ≒ (k) = dummy residual becomes d (k) The signals r _d (k) and r (k) are decorrelated.

However, if howling is detected based on these two coefficients, a malfunction may occur due to the near-end speaker signal. Therefore, at the time of howling, utilizing the fact that the signal level of transmission and reception becomes high, the ADF performing the adaptive operation by the learning identification method calculates the total power MPOW (k) of the input signal register of the ADF calculated for the adaptive operation. _Is observed by the determination unit 508 as a signal level detection unit, and M _pow (k)> M _pth ... G (k)> g _th ... ₂ (k)> a _2th ... M _pth : power M _pow ( k) Threshold value g _th : g (k) threshold value a _2th : ₂ (k) threshold value If howling is detected at the same time, erroneous detection is eliminated. Of course, the coefficient
_{When 2} (k) = 1, the equation may be omitted.

Incidentally, those configurations howling detector 110 described above is merely an example, this configuration of the howling detector 110 to be anything like limited to the configuration of the embodiment described the above have a configuration which can output a howling detector J ₃ is not. Further, in the above-described example of the howling detector, the echo signal S _in (k), that is, y (k) is input and detected, but the received signal R _in (k) (x (k)) or Residual signal _Res
Even if the processing is performed by inputting (k) (e (k)), the same effect as in the case of the signal y (k) can be obtained.

ADF control unit The ADF control unit 112 is a device for controlling the silence detector (ID) 106 described above.
Detection signal J ₁ , detection signal J ₂ from double talk detector (DTD) 108, detection signal from howling detector (HD)
With J _3, it has a configuration that outputs the ADF adaptive estimation operation inhibition signal INH in accordance with these detection results.

FIG. 7A is a circuit diagram showing a configuration example of the ADF control section 112. In this configuration example, a logic circuit is used, and a truth table, a detection mode, and an operation of the ADF for explaining the logic combination operation are shown in FIG. 7B.
In this circuit, 702 is an OR gate, 704 is an AND gate, 7
06 is a NOT gate, and the NOT gate has a detection signal J ₃
The AND gate 704 is supplied with the detection signal J ₂ and the inverted detection signal J ₃ inverted by the NOT gate. The OR gate 702 is further supplied with the detection signal J ₁ , the detection signal J ₂ and the inverted detection signal J _2.
The logical product signal of ₃ is supplied, and the ADF adaptive estimation operation inhibition signal INH is finally output from the OR gate 702. Therefore, the ADF control unit 112 outputs the adaptive estimation inhibition signal INH = 0 at the time of howling occurrence and single talk,
Let 102 perform adaptive estimation. On the other hand, at the time of double talk and at the time of reception signal silence, this signal INH = 1 is output, and A
DF102 adaptive estimation is prohibited.

(Effects of the Invention) As is clear from the above description, according to the echo canceller of the present invention, a howling detector, a received signal silence detector, and a double talk detector are provided, and ADF adaptation is performed based on each detection result. Since the prohibition and release of the estimation operation are controlled, the ADF immediately performs adaptive estimation based on the detection result at the time of howling, so the howling stops immediately and does not hinder the call, and therefore the call quality is reduced. Are better.

For double talk detection, the same performance as the conventional one can be expected.

Further, in the present invention, since the howling detector and the double talk detector which individually provide the howling and the double talk which are hard to distinguish from each other are detected in combination, there is no mutual detection of the howling and the double talk. Therefore, the above-described adaptive estimation control can be properly performed, and the above-described adaptive estimation prohibition or cancellation control can be properly performed.

In addition, since the howling detector includes the frequency component detection unit and the signal level detection unit, the circuit configuration of the howling detector can be simplified. as a result,
It is possible to avoid a complicated circuit configuration of an echo canceller combining a howling detector and a double talk detector.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the echo canceller of the present invention, FIG. 2 is a block diagram showing an example of the configuration of a silence detector used in the present invention, and FIG. 3 is double talk used in the present invention. FIG. 4 is a block diagram showing one configuration example of a detector, FIG. 4 is a block diagram showing one configuration example of a double-talk detection threshold control circuit provided in the double-talk detector shown in FIG. 3, and FIG. FIG. 6 is a block diagram showing an example of the configuration of a howling detector used in the present invention. FIG.
FIG. 7 is a circuit diagram showing an example of the configuration of the next FIR. FIG. 7 is a diagram showing an example of the configuration of an ADF control unit used in the present invention and its operation. FIG. 8 is a block diagram showing an example of the configuration of a conventional echo canceller. FIG. 9 is a block diagram showing a configuration example of a double talk detector provided in the echo canceller of FIG. 100 Echo canceller 102 Adaptive digital filter (ADF) 104, 122, 310, 408, 420, 602, 612, 616 Adder 106 Silence detector (ID) 108 Double talk detector (DTD) 110 Howling detector (HD) 112 ADF control unit, 114 ...... absolute value circuit 116 ...... integrating circuit, 118,312 ...... comparator 120 ...... [delta] ₁ multiplier 124 ...... _{(1-δ 1)} multipliers 126,412,604,606 ...... unit delay elements 302, 306 ...... power detection Units 304, 308 Logarithmic converter 314 Double threshold threshold (A _t ) control circuit 402, Average circuit, 404 r Adder 406, 414 Selector switch 416, 608, 610, 614 Multiplier, 410 Limiter 502 ... Second-order acyclic adaptive prediction filter 504..., ₁ and ₂ control units 506... G coefficient control unit 508... Judgment unit 702... OR gate 704.

 ──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Yoshikazu Nakano 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) References JP-A-58-142641 JP-A-56-112112 (JP, A) JP-A-57-87246 (JP, A)

Claims (1)

(57) [Claims] An echo canceller for inputting a received signal to generate a pseudo echo signal and for generating a pseudo echo signal and for canceling the echo by subtracting the pseudo echo signal from the echo signal detects a silent state of the received signal. A silence detector, a double talk detector for detecting a double talk state, a howling detector for detecting a howling state, a silence detector, the double talk detector, and a detection result from the howling detector. Adaptive control for performing adaptive estimation at least when the reception signal is in a sound state and howling is occurring, and performing control for inhibiting adaptive estimation when the reception signal is in a sound state and double talk. An echo canceller comprising: a digital filter control unit; A frequency component detection unit that detects that the echo signal, or a residual signal obtained by subtracting a pseudo echo signal from the echo signal, or a received signal is a signal in which a specific frequency is emphasized; An echo canceller, comprising: a signal level detection unit that detects that a signal level of a transmission signal exceeds a predetermined threshold.