JPH0793595B2 - Double talk detection method and double talk detector - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a double talk detection method and a double talk detector of an echo canceller applied to a telephone line such as a public address telephone and satellite communication.

(Prior Art) Conventionally, an echo canceller of this type is disclosed in
As disclosed in Japanese Laid-Open Patent Publication No. 61-56526, there is a configuration applied to a public address telephone.

FIG. 5 is a block diagram showing a configuration example of this conventional echo canceller. In the figure, reference numeral 800 is an echo canceller, and 802 is a loudspeaker telephone. This device is a loudspeaker phone 8
In 02, the audio signal R _out output from the speaker 804
Is the echo path in the room (Echo Path. Below EP
Sometimes called. ) Is input to the microphone 806 and leaks as an echo signal S _in (k) _in the transmission signal, which is canceled by an echo canceller (hereinafter sometimes referred to as EC) 800. Reference numeral 808 is a speaker amplifier, and 810 is a microphone amplifier.

The echo canceller 800 is an adaptive digital filter (Ada) that receives a received signal and outputs a pseudo echo signal S _in (k).
ptive Digital Filer. Hereinafter, it may be referred to as ADF. ) 812, a double talk detector (hereinafter referred to as DTD) 814 for controlling the adaptive estimation operation of the ADF 812, and the pseudo echo signal S _in (k) as the echo signal S
It is mainly composed of an adder 816 that subtracts from _in (k) to generate a residual signal R _es (k). Here, 818 and 820 are A / D converters, 822 and 824 are D / A converters, and k is 8 for example.
It represents the sampling time that is performed in synchronization with the synchronization clock pulse of kHz.

In the adaptive estimation operation of the ADF812, only the received signal R _in (k) exists,
Since it is accurately performed only in the single talk in which the near-end talker signal N does not exist, the ADF812 is used in the double-talk state in which the reception signal R _in (k) and the near-end talker signal N exist at the same time.
The DTD814 that prohibits the adaptive estimation operation of is required.

By the way, in the technique disclosed in Japanese Patent Laid-Open No. 61-56526, the detection sensitivity of DTD814 is increased. FIG. 6 is a block diagram showing a DTD of this conventional configuration. In this configuration, 901 to 903 and 923 are peak value detectors, 904 to 906 are squaring circuits, 907 and 908 are priority encoders, and 909.
Is AT memory, 910, 918, 924, 925 is adder, 911, 912,
916 is a comparator, 913, 914, 917 is a switch, 915 is a shift circuit, 919, 920 is a limiter, and 921, 922 is a correction amount memory. Signal R _in (k) and echo signal S _in (k) and residual signal R
A peak value detection circuit 90 corresponding to _es (k) (in FIG. 6, these signals are designated as x (k), y (k) and e (k))
After detecting the peak value at 1, 902, 903, the squaring circuit 904, 90
Calculate peak power at 5,906. And the received signal x
(K), each signal level L of the residual signal e (k)
_{x (k)} and Le _(k) are obtained as logarithmic domain values by the priority encoders 907 and 908, and the difference between the received signal level _{Lx (k)} and the threshold value AT (k) in the comparator 912 is the signal of the residual signal. Level L
When it becomes larger than _{e (k)} , that is, when L _{x (k)} −AT (k)> L _{e (k)} , it is judged as non-double talk and output as ADF adaptive operation prohibition signal INH = 0 and, on the other hand, L when _x in _{(k) -AT (k) ≦} L e (k), ADF adaptive operation prohibition is determined that the double-talk signal I
NH = 1 is output to control the ADF adaptive operation. Based on this detection result, the threshold value AT is as in non-double talk (INH = 0) AT (k + 1) = AT (k) + δD during double talk (INH = 1) AT (k + 1) = AT (k) −δD. Have control over. _{L x (k) -AT (k} ) ≦ L e (k) is established by variations in the double talk time and EP, it is prohibited adaptive operation of ADF812, threshold AT (k) for decreases with the passage of time , L _{x (k)} −AT (k)> L _{e (k)} , and the ADF 812 starts the adaptive estimation operation again.

(Problems to be Solved by the Invention) However, in the case of the loudspeaker telephone system of the echo canceller having the conventional configuration, if a sudden echo path variation occurs and howling occurs, L _{x (k)} −AT (k ) ≦ L _{e (k)} , until the threshold AT (k) is successively decreased and L _{x (k)} −AT (k)> L _{e (k)} is satisfied, adaptive operation of the ADF is prohibited and howling There was a serious drawback that the condition would continue.
Therefore, there is a drawback that the far-end speaker and the near-end speaker cannot talk because of the howling sound.

An object of the present invention is to eliminate the conventional drawback that howling caused by such a rapid echo path variation connects during the ADF adaptive operation prohibition period, and to obtain a double talk detection method for obtaining an echo canceller with excellent speech quality. And to provide the device.

(Means for Solving the Problems) In order to achieve this object, according to the double-talk detection method of the present invention, an adaptive digital filter for inputting a received signal to generate a pseudo echo signal and a double-talk state are provided. An echo canceller having a double-talk detector for detecting and a howling detector for detecting a howling state, and controlling an adaptive estimation operation of the adaptive digital filter based on detection results of the housing detector and the double-talk detector. In the double-talk detection method according to, the level difference between the residual signal obtained by subtracting the pseudo echo signal from the echo signal and the received signal is calculated, and a double-talk state is determined from the level difference based on a double-talk detection threshold. In order to detect whether or not the double talk detection threshold is (i) the double talk detection threshold, In single talk, the value of which is smaller than the level difference, and when there is sound in the received signal, the long-term average value of the level difference is set, and (ii) during single talk and the received signal When there is no sound or when only the near-end signal is transmitted, the current threshold value is held, and (iii) the double talk detection threshold value is larger than the level difference, and the reception signal is present. It is characterized in that it is gradually lowered when the sound changes or when the echo path changes to the extent that howling is not reached, and (iv) at the time of howling detection, it is lowered faster than in the case of (iii).

Further, according to the double talk detector of the present invention, an adaptive digital filter for receiving a received signal to generate a pseudo echo signal, a double talk detector for detecting a double talk state, and a howling detector for detecting a howling state. In the double-talk detector in the echo canceller for controlling the adaptive estimation operation of the adaptive digital filter based on the detection results of the housing detector and the double-talk detector, the received signal and the received signal In response, a residual signal obtained by subtracting the pseudo echo signal output from the adaptive digital filter from the echo signal is supplied, and a signal level calculation circuit for calculating the level difference between the received signal and the residual signal And a ratio of a level difference signal between the received signal and the residual signal and a double talk detection threshold value. And outputs an adaptive estimation operation prohibition signal, and a comparator that outputs a talk state detection signal of a single talk state or a double talk state, a howling detector that detects howling and outputs a howling detection signal, and A silence detector that detects a silence state of a reception signal and outputs a silence detection signal; and the level difference signal, the talk state detection signal, the howling detection signal, and the silence detection signal are supplied, (i) the double talk During single talk, in which the detection threshold is smaller than the level difference, and when there is sound in the received signal, a long-term average value of the level difference is used. (Ii) During single talk and the received signal When there is no sound or when only the near-end signal is transmitted, the current threshold value is held, and (iii) the double talk detection threshold value exceeds the level difference. Also during large double talk, and when there is a sound of the received signal or when the echo path changes to the extent that howling does not occur, (iv) at the time of howling detection, compared to the case of (iii) above. A double-talk detection threshold value control circuit is provided for controlling the double-talk detection threshold value in each case so that the double-talk detection threshold value can be quickly reduced.

Further, in carrying out the double talk detection according to the present invention, the howling detector comprises a second-order acyclic adaptive prediction filter to which a received signal, an echo signal or a residual signal is input, It is preferable that the howling is detected by the following coefficient, predicted value output control coefficient, and received signal register power of the adaptive digital filter for pseudo echo output.

(Operation) As described above, according to the double-talk detecting method and the apparatus for carrying out the same according to the present invention, the double-talk is detected and the detection threshold value is controlled if the howling does not occur, and the double-talk is detected when the howling occurs. The detection threshold of is reduced faster than in the case of double talk, and the inhibition of the adaptive estimation operation of the ADF is released faster.

Therefore, when howling, the ADF is immediately determined based on the detection results.
Since the adaptive estimation is performed, the howling is stopped immediately, and thus the call quality is not impaired, and the double talk detection guarantees the call quality equivalent to the conventional one.

(Embodiment) An embodiment of the method and apparatus of the present invention will be described together with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the configuration of the double talk detector of the present invention used in an echo canceller.

In this embodiment, the double-talk detector 100 for controlling the adaptive estimation operation of the adaptive digital filter of the echo canceller includes a power detector 102 and a logarithmic converter 104 for the received signal x _{k (k)} (R _in (k)). And a residual signal e _(k) R _es (k)) including a power detector 106 and a logarithmic converter 108, which output a signal level L _x (k) and L _e (k), respectively.
And a level signal adder from the logarithmic conversion units 104 and 108.
And 112, a comparator 114, double-talk detection threshold value control circuit for setting a double-talk detection threshold value A _t (k) in the comparator 114 (hereinafter, referred to as A _t control circuit.) And 116, the silence detector 118 And howling detector 120.

The received signal R _in (k) and the residual signal R _es (k) are detected by the power detector 102.
And 106 respectively, the power is detected, and the power signals P _x (k) and P _e (k) are obtained, respectively, and then the logarithmic converters 104 and 108 connected in the subsequent stage respectively output the signal level L _x of the value in the logarithmic domain. (k) and L _e (k), respectively. In the adder 112, the level difference signal A _com (k), which is the difference between the signal levels of the two, is added.
= L _x (k) at one output a -L _e (k) is A _t the control circuit 11
6 and the comparator 114 on the other hand.

In the comparator 114, in advance by A _t the control circuit 116 (k-1)
The threshold value A _t (k) calculated at the sampling time of is compared with the level difference signal A _com (k), and when A _t (k) <A _com (k), the comparison result is generated. The talk state detection signal J ₂ has one logical value, for example, “0” indicating that it is in the single talk state, and outputs the adaptive operation prohibition signal INH = 0 to enable the ADF to perform the adaptive operation.

In the case of _{A t (k) ≧ A com} (k) , the other logic value, for example, "1" indicating that the talk state detecting signal J ₂ is a double-talk state, and the outputs the INH = 1 Disables adaptive operation of the ADF.

A _t the control circuit the talk state detecting signal J ₂ is shown in Figure 1
116 to be input as A _t control selection signal.

In the present embodiment, the silence detector 118 described above has a signal level L _x (k) of the received signal and a preset silence detection threshold L _x t.
It is composed of a comparator which compares with h and detects silence in the received signal. In this comparator 118, L _x th> L _x (k)
And when it is judged outputs the adaptive operation inhibition signal INH = 1 in order to prevent the ADF estimation operation is disturbed, A _t
Detected as control selection signal signal J ₁ (J ₁ = ₁₎ to A _t control circuit
Input to 116. On the other hand, when L _x th ≤L _x (k), there is sound, so INH = 0 and the detection signal J ₁ is output as J ₁ = 0.

Adaptive hauling detector 120, also described in detail later, the residual signal e (k), to detect the howling with ADF of the received signal register power MPOW (k), A _t the control circuit 116 when the howling detection to A _t control selection signal J ₃ = 1
To enter. The received signal register power M of this ADF
POW (k) is a value calculated for the ADF to perform the adaptive estimation operation by the learning identification method, and is the received signal (R _in) corresponding to each sampling time stored in each unit delay element of the ADF. (k), R _in (k-1), ..., R _in (kn) (where n is the order of the ADF) is the total power (total power).

On the other hand, the A _t control circuit 116 and J ₂ of the detection result of the comparator 114 described above will be silence detection result J ₁ and described below is input as A _t control prohibition signal from the silence detector 118 (Figure 1) This is a circuit for updating, storing, and subtracting the double-talk detection threshold A _t (k) in response to these signals based on the howling detection result J _{3 of the} adaptive howling detector (AHD) 120. Figure 2 is a block diagram showing a specific configuration example of the A _t control circuit 116. The A _t the control unit 116, the averaging circuit 202, a selector switch 204, an adder 206 for reducing the later-described threshold value, corresponding to the next sampling time, which is the output signal of the A _t control circuit 116 Threshold A _t (k + 1)
A limiter 208 for outputting the signal, a unit delay element 210 for delaying this output by a unit sampling time, constants 0, δ ₁
Or (the [delta] ₁ and [delta] ₂ is a constant independent from each other.) [Delta] ₂ and a selector switch 212 for selecting a multiplier 214 multiplying the selected constant, the threshold value A _t (k) and the level difference signal described later And an adder 216 that takes the difference from A _com (k).

The level difference signal A _com (k) is supplied to the average value circuit 202 and the adder 216, respectively. This averaging circuit 202 produces a signal A _com (k)
For example, in a circuit that calculates a long-term average value such as 128 or 256 sampling intervals, The average value is calculated in accordance with.

Then, if necessary, a process for holding the double-talk detection sensitivity constant for the long-time average value signal is performed, and then the signal is sent to the selector switch 204.

The output of this average value is supplied to the A-side terminal of the selector switch 204. A unit delay element 210 is connected to the B side terminal of the selector switch 204. This selector switch 204 has a silence detection result J ₁ from the silence detector 118,
The detection results (control prohibition signal) J ₂ from the comparator 114 for double-talk detection are input, respectively, and operation is performed based on these detection results J ₁ and J ₂ in the relationship shown in Table I below. To do.

Table I J ₁ J ₂ Select terminal 0 0 A 0 1 B 1 0 B 1 1 B As can be understood from this Table I, this selector switch 204 selects the terminal A in the case of sound and single talk. outputs long-term mean value calculated by the average value circuit 202, in other words if only the idle calling or near-end signal is transmitting the other state, a _t the control circuit 116 terminal B is selected threshold It is configured to be in an operating state in which hold or subtraction is performed.

The adder 216 outputs the signal A _com (k) and the output A _t (k) of the unit delay element 210.
And the difference signal to the multiplier 214.

The multiplier 214 multiplies the difference signal A _t (k) −A _com (k) by the constant (0, δ ₁ or δ ₂ ) selected by the selector switch 212 and supplies the result to the adder 206.

The adder 206 is a multiplier from the output of the selector switch 204.
The output of 214 is subtracted and output to the limiter 208. The limiter 208 has a function of limiting the value of the threshold A _t (k) if desired such that the threshold A _t (k) is not set unreasonably high or low. The output of the limiter 208 is output to the unit delay element 210 and the comparator 114 shown in FIG. 1 as the updated threshold value A _t (k + 1).

The selector switch 212 detects the above-mentioned detection results J ₁ , J ₂ and J ₃
Based on three constants “0”, “δ ₁ ” and “δ ₂ ”.
(However, it is set _{0 ≦ δ 1 << δ 2 ≦} 1 and.) Select one constant depending on the state of the subtraction of the hold threshold A _t (k) of the, A _t (k), It operates as shown in Table II below.

As can be understood from Table II, in the selector switch 212, the constant "0" is selected in the single talk state of J ₁ = 0, J ₂ = 0 and J ₃ = 0, and the difference signal A _t (k) is selected. −A _com (k)
Send multiplied by 0, while the output value from the adder 206 is the signal of the long-term average value selected from the averaging circuit 202 by the selector switches 204 A _t the (k + 1) to the limiter 208.

If no call or only near-end signals with J ₁ = 1, J ₂ = 0 and J ₃ = 0 are being transmitted, the selector switch
204 since the constant "0" in B is connected to the terminal and the selector switch 212 is selected, the threshold value _{A t (k + 1) =} A t (k) , and becomes a hold state.

Further, in a double talk state or an echo path fluctuation in which J ₁ = 0, J ₂ = 1 and J ₃ = 0, the selector switch 204 is connected to the B side terminal, and the selector switch 212 selects the constant δ ₁ to select the multiplier 214. Δ ₁ (A _t (k) −A _com (k)) is output from the adder 206, the output value from the adder 206 is A _t (k + 1) = A _t (k) −δ ₁ (A _t (k ) −A _com (k)), which is the output value, that is, the double-talk detection threshold A _t (k + 1) at the next sampling time (k + 1) is A
gradually subtracted in proportion to the difference between the _com (k) and A _t (k).

Further, in the howling state due to rapid echo path fluctuations of J ₁ = 0, J ₂ = 1 and J ₃ = 1, the selector switch 204
Is connected to the B-side terminal and the selector switch 212 selects the constant δ ₂ , the threshold A _t (k + 1) is subtracted as in the case of the double talk state or the echo path fluctuation.
However, in this case, since the constant δ ₂ is considerably larger than δ ₁ , this threshold A _t (k + 1) is A _t (k + 1) = A _t (k) −δ ₂ (A _t (k) -A _com (k)) is rapidly subtracted. Therefore, A
_com (k)> A _t (k), which causes the condition of L _x (k) −A _t (k)> L _e (k) to be restored, and thus the ADF can be adaptively operated. That is, in the howling state, A _com (k) decreases and A _com (k) decreases.
Adaptive operation of DF is prohibited, but this is immediately detected and AD
The adaptive movement of F starts rapidly and the howling stops.
For this reason, the double-talk detection sensitivity becomes constant, the followability to the minute fluctuation of the echo path is improved, and the communication quality of this telephone line is kept high.

In addition, (J ₁ , J ₂ , J ₃ ) is ( ₁ , ₁ , 0), (1, 1, 1)
Since the states and (1, 0, 1) are inconsistent, these states are excluded from Table II. Also, (0, 0,
1) indicates that the ADF can be adaptively operated in the howling state, and is excluded from this Table II.

Next, the adaptive howling detector (AHD) 120 will be described.

FIG. 3 is a block diagram showing a configuration example of the adaptive howling detector 120, and FIG. 4 is a second-order acyclic adaptive prediction filter (hereinafter sometimes referred to as second-order FIR) incorporated therein. 2 is a circuit configuration diagram of FIG.

This howling detector 120 has a residual signal e (k) (R
_es (k)) is input to the secondary FIR502 and the secondary FIR502 coefficient
₁ (k), ₂ (k) control unit 504, control unit 506 for predictive output control coefficient g (k) of the adaptive prediction filter, g (k),
₂ (k) and is composed of an ADF received signal register power Mpow (k) determining unit 508 detects howling more, when howling detection, A _t control circuit howling detection signal J ₃ as one logic value "1", for example and outputs to 116, in other cases has a configuration to be output to a _t the control circuit 116 the howling detection signal J ₃ as other logic value "0", for example.

Next, the principle of howling detection will be described.

Now, input and output R _in (k), R _es (k), and R _in (k) of the echo canceller 800 when howling occurs are x (k) and e, respectively.
(K) and y (k), these x (k) and e
(K) and y (k) are almost a single sine wave, and the spectrum has a line spectrum property. Second-order recursive filter with transfer function of
Called. ), In order to make this sine wave sequentially adaptively predicted and uncorrelated, Q (Z) = 1- (Z). It is sufficient to input it to the second-order FIR having the transfer function of.

The circuit of the secondary FIR shown in FIG. 4 receives the residual signal e (k) and outputs the dummy residual signal r _d (k) from the dummy prediction value _d (k) and the adder 602. A unit delay element 604 for delaying (k) by a unit sampling time, and a unit delay element for delaying the delayed e (k) again
606 and the coefficient e (k-1) from the unit delay element 604
A multiplier 608 that multiplies ₁ (k) and a multiplier 610 that multiplies the signal e (k−2) from the unit delay element 606 by a coefficient ₂ (k).
And an adder 612 that adds the signals from the multipliers 608 and 610 and outputs the predicted value (k), and the dummy predicted value described above by multiplying the predicted value (k) by a coefficient g _d (k). _d
(k) is output from the multiplier 614, and the predictor (k) and the residual signal e (k) are subtracted from the adder 616 to output the residual signal r (k). There is. The residual signal r (k) and the dummy residual signal r _d (k) obtained here are both sent to the ₁ (k), ₂ (k) control unit 504 and the g coefficient control unit 506 in FIG.

In the configuration shown in FIG. 4, an algorithm for adaptively outputting the prediction value (k) to make the residual signal r (k) uncorrelated regardless of howling occurs will be described below. To do.

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

Further, for the purpose of predicting only the sine wave and making r (k) uncorrelated, ₂ (k) = 1 ... When the root P of 1-A (z) = 0 is displayed on the Z plane in polar coordinates, P = r · e ^{± jω,} where r: absolute value of complex number P ω: declination of complex number P: It can be expressed as the base of natural logarithm, and by substituting this into 1-A (z) = 0, 1-A (z) = 0 (1-re ^jω Z ⁻¹ ) · (1-re ^−jω Z ^{− 1} ) = 0 1-2 r · cos ω · Z ⁻¹ + r ² Z ⁻² = 0 …… The reason why the impulse response of the second-order IIR becomes a sine wave is because it is only when r ² = 1. _{The 1} (k), ₂ (k) control unit 504 controls the coefficients ₁ (k) and ₂ (k) based on the above equation.

Next, the predicted value output control coefficient g (k) is introduced as a method of expressing how close the signal is to a sine wave. An algorithm for adaptively changing the coefficient g (k) to make the dummy residual signal r _d (k) uncorrelated is given as described below.

r _d (k) = e (k) _-d (k) ...... Therefore, Or, g (k + 1) = g (k) + δ · sgn {r d (k)} · sgn {
(K)} ...... 0 ≦ g (k) ≦ 1 The g (k) control unit 506 calculates g (k) based on the equation.
To control.

With the above two controls, if e (k) is uncorrelated, g (k) ≈0 and the predicted value (k) = 0. If the residual signal e (k) is sinusoidal (howling state), the coefficient g (k) ≈1 ₂ (k) = 1 e (k) ≈ (k) = _d (k) and the dummy residual The difference signals r _d (k) and r (k) are decorrelated.

However, if howling is detected by only these two coefficients, there is a possibility of malfunction due to the near-end talker signal. Therefore, at the time of howling, the fact that the signal level of transmission and reception becomes high level is used to calculate the total power MPOW of the ADF input signal register calculated for ADF adaptive operation.
Observing (k), M _pow (k)> M _pth …… g (k) ＞ g _th …… ₂ (k) ＞ a _2th …… M _pth : Power M _pow (k) threshold g _th : g ( k) Threshold value a _2th : ₂ (k) threshold value If the _howlings are detected at the same time, the erroneous detection is eliminated. Of course, depending on the formula
_{When 2} (k) = 1, the formula may be excluded.

The configuration of the adaptive howling detector 120 described above is merely an example, and the configuration of the howling detector 120 is not limited to the configuration of the above-described embodiment as long as the howling detection signal J ₃ can be output. Not a thing. Also,
In the example of the howling detector described above, the residual signal R _es (k), that is, e (k) is input and detected, but the received signal R _in (k) (x (k)) or the echo is received. Signal S
_{Even if in} (k) (y (k)) is input and processed, the same effect as that of the signal e (k) can be obtained.

The invention is not limited to the embodiments described above. For example, although the adaptive howling detector and the silent detector are incorporated in the double-talk detector in the above-described embodiment, the adaptive howling detector and the silent detector are provided outside the double-talk detector and used in combination therewith. In the case, it is also included in the scope of the present invention.

(Effect of the invention) As is apparent from the above description, according to the double-talk detecting method and the apparatus thereof of the present invention, in any case, howling is higher than the descending speed of the double-talk detecting threshold value during normal double-talk. Even if howling occurs due to rapid echo path fluctuations and ADF is disabled for adaptive operation, the threshold for double talk detection is reduced at a high speed, and ADF is adaptively estimated immediately because the double-talk detection threshold at the time of detection is configured to be very fast. Howling is stopped as it becomes operational.
Therefore, the call quality is less deteriorated and the call quality of the telephone line can be kept high.

Further, according to the apparatus of the present invention, the howling detector is constituted by a simple second-order acyclic adaptive prediction filter, so that howling can be performed with high accuracy by using the detection parameters represented by Can be detected.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the double talk detector of the present invention, and FIG. 2 is a structural example of a double talk detection threshold value control circuit provided in the double talk detector shown in FIG. FIG. 3 is a block diagram showing the configuration example of the adaptive howling detector used in the present invention, and FIG. 4 is a configuration of a secondary FIR provided in the adaptive howling detector shown in FIG. FIG. 5 is a block diagram showing a configuration example of a conventional echo canceller, and FIG. 6 is a block diagram showing a conventional configuration of a double talk detector provided in the echo canceller of FIG. . 112,206,216,602,612,616 …… Adder 118 …… Silence detector (comparator) 100 …… Double talk detector (DTD) 110 …… Signal level calculation circuit 120 …… Adaptive howling detector (AHD) 114 …… Comparator 210,604,606 …… Unit delay element 102,106 …… Power detector 104,108 …… Logarithmic converter 116 …… Double-talk detection threshold (A _t ) control circuit 202, …… Average value circuit 214,608,610,614 …… Multiplier 204,212 …… Selector switch 208 …… Limiter 502 ...... Second-order acyclic adaptive prediction filter 504 …… ₁ , ₂ control unit 506 …… g coefficient control unit, 508 …… Determination unit.

Claims (3)

[Claims] 1. An adaptive digital filter for receiving a received signal to generate a pseudo echo signal, a double-talk detector for detecting a double-talk state, and a howling detector for detecting a howling state. Detector and double-talk detection method in an echo canceller for controlling the adaptive estimation operation of the adaptive digital filter based on the detection results of the double-talk detector, a residual signal obtained by subtracting the pseudo echo signal from an echo signal, and the reception In order to detect a level of double talk from the level difference by calculating a level difference from the signal and using the level of the double talk detection threshold as a reference, (i) the double talk detection threshold is set to the level During single talk, which is smaller than the difference, and when there is sound in the received signal Set long-term average value of the level difference, a time (ii) single talk, and, when only silence or when the near-end signal of the reception signal is being sent,
Hold the current threshold value, (iii) gradually during double talk when the double talk detection threshold value is larger than the level difference, and when there is a sound in the received signal or when the echo path changes to the extent that howling does not result. And (iv) at the time of howling detection, the double-talk detection method is characterized in that it is made faster than in the case of (iii) above. 2. An adaptive digital filter for inputting a received signal to generate a pseudo echo signal, a double-talk detector for detecting a double-talk state, and a howling detector for detecting a howling state. In the echo canceller for controlling the adaptive estimation operation of the adaptive digital filter based on the detection result of the detector and the double talk detector, the received signal and the adaptive digital filter in response to the received signal. A signal level calculation circuit that is supplied with a residual signal obtained by subtracting the output pseudo echo signal from the echo signal and that calculates a level difference between the received signal and the residual signal, the received signal and the residual signal Output the adaptive estimation operation prohibition signal by comparing the level difference signal between the difference signals and the double talk detection threshold. In addition, a comparator that outputs a talk state detection signal indicating a single-talk state or a double-talk state, a howling detector that detects howling and outputs a howling detection signal, and a silence state by detecting a silent state of the received signal. A silence detector that outputs a detection signal, the level difference signal, the talk state detection signal, the howling detection signal, and the silence detection signal are supplied, and (i) a single whose double talk detection threshold is smaller than the level difference. At the time of talk and when the received signal has a sound, the long-term average value of the level difference is used. (Ii) At the time of single talk and when the received signal has no sound or near-end signal only. Is being sent,
The current threshold value is held, and (iii) during double talk in which the double talk detection threshold value is larger than the level difference, and when there is sound in the received signal or when the echo path changes to the extent that howling does not occur, gradually And (iv) at the time of howling detection, the double talk detection threshold value control for separately controlling the double talk detection threshold value so as to decrease the double talk detection threshold value faster than in the case of (iii). A double talk detector comprising a circuit. 3. The howling detector comprises a quadratic acyclic adaptive prediction filter to which a received signal, an echo signal or a residual signal is input, and a quadratic coefficient of the adaptive prediction filter, a prediction value output control coefficient and The double-talk detector according to claim 2, wherein the howling is detected by the received signal register power of the adaptive digital filter for pseudo echo output.