JPH0613940A - Echo canceller - Google Patents

Abstract

PURPOSE:To provide a small sized echo canceller which is able to discriminate various talking states with high accuracy independently of the presence of noise and whose echo cancellation accuracy is improved. CONSTITUTION:A transmission voice detector 8 outputs a significant detection signal in the double talk state and the transmission state and outputs an insiginificant signal in the reception state and the non-voice state based on autocorrelation of a signal on a transmission signal line. A reception voice detector 10 outputs a significant detection signal in the double talk state and the reception state and outputs an insiginificant detection signal in the transmission state and the non-voice state based on a signal on a reception signal line. Then, a talking state discrimination device 11 discriminates the talking state based on the detection signal from the transmission voice detector 8 and the reception voice detector 10 and a coefficient control section 6 controls the revision of the coefficient of an adaptive filter 4 based on the discriminated talking state.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller, and is particularly suitable for application to a hands-free telephone.

[0002]

2. Description of the Related Art In a hands-free telephone, an omnidirectional microphone is used to increase the positional freedom of a near-end speaker. It is unavoidable to be captured and is sent as an echo towards the far-end speaker. In order to avoid such inconvenience, many hands-free telephones are provided with an echo canceller.

FIG. 2 shows an example of a conventional echo canceller (Japanese Patent Laid-Open No. 3-80628). It should be noted that FIG. 2 shows the principle configuration, and the analog / digital converter and the digital / analog converter are omitted.

In FIG. 2, the received signal R relating to the utterance of the far-end speaker is given to the speaker 104 to be sounded, and is also inputted to the adaptive filter 101. Speaker 104
The received voice (echo) R2 sounded from is sneak into the omnidirectional main microphone 102, and the main microphone 10
It is captured by 2 and given to the adder 105 as a subtracted input. When the receiving signal R is input, the adaptive filter 101 performs a convolution operation with the filter coefficient (estimated value of echo path gain) at that time, and supplies the calculated value to the subtraction input terminal of the adder 105 as a pseudo echo signal. Thus,
From the adder 105, a signal obtained by removing the echo component from the captured signal from the microphone 102 is output, and this signal is transmitted and given to the coefficient control circuit 106.
As will be described later, the coefficient control circuit 106 uses the adaptive filter 1 based on the signal output from the adder 105 during the period when the update of the coefficient of the adaptive filter 101 is permitted.
The coefficient of 01 is updated.

The echo becomes a problem when the reception signal R is received. In such a case, only the reception signal R is received, and the reception signal is received. There is a case of so-called double talk in which the near-end speaker 108 is making a voice. In the former case, the adder 105 outputs an echo component removal residual signal, which is a case suitable for updating the filter coefficient. On the other hand, in the latter case, the adder 105 outputs the transmission signal relating to the near-end talker in addition to the echo component removal residual signal, and when the filter coefficient is updated by this signal, the content of the transmission signal is mainly changed. Since the coefficient is updated based on this, it is generally not suitable for updating.

Therefore, at the time of double talk, a structure for prohibiting the update of the coefficient of the adaptive filter 101 is provided. That is, the sub microphone 103 having directivity and the double talk detection circuit 107 are provided. The microphone 103 has a directivity and an installation position so as to satisfy the expressions (1) to (4) described later. Double talk detection circuit 107
Detects the double-talk state from the captured signal of the main microphone 102 and the captured signal of the sub-microphone 103, and prohibits the coefficient update from the coefficient control circuit 106 when the double-talk state is detected.

The principle of detecting the double talk state in this conventional example will be described below. Here, among the signals captured by the omnidirectional microphone 102, the input levels of the reception signal component, the transmission signal component, and the noise (background noise) component are R2, S2, and N2, respectively, and the directional microphone 103 is used.
The input levels of the reception signal component, the transmission signal component, and the noise component of the captured signal by R3, S3, and N3 are respectively
As described below.

By properly selecting the directivity of the sub microphone 103 and the positional relationship between the microphones 102 and 103 and the speaker 104, the following expressions (1) to (4) are satisfied. That is, this conventional example
It is assumed that Eqs. (1) to (4) hold.

R2 + N2> R3 + N3 (1) S3 + N3> S2 + N2 (2) R2> R3 (3) S3> S2 (4) In the transmission state in which only the near-end speaker 108 is uttering voice, The signals captured by the microphones 102 and 103 are composed of transmission signal components S2 and S3 and noise components N2 and N3. Therefore, the power ratio Ssend of the captured signals of both microphones 102 and 103 in the transmitting state can be expressed by the following equation.

Ssend = (S3 + N3) / (S2 + N2) (5) Similarly, when the captured signal is divided into its components, both microphones 102 and 103 are captured in a receiving state in which only the far-end speaker is making a voice. The signal power ratio Srec is (6)
The power ratio Sdouble of the captured signals of both microphones 102 and 103 in the double-talk state in which both the near-end speaker 108 and the far-end speaker are uttering voice.
Can be expressed by equation (7).

Srec = (R3 + N3) / R2 + N2 (6) Sdouble = (R3 + S3 + N3) / (R2 + S2 + N2) (7) Power ratio Ssend in the transmitting state, power ratio Srec in the receiving state, and power ratio S in the double-talking state
When the magnitude relation of double is obtained by using the above equations (1) to (4), the following equation is obtained.

Srec <Sdouble <Ssend (8) Accordingly, the power ratio between the captured signals of both microphones 102 and 103 is obtained, and this power ratio is compared with a two-step threshold value to determine whether the state is one of three states. Can be determined.

[0013]

However, the conventional echo canceller shown in FIG. 2 has the following drawbacks regarding the detection of double talk.

(1) When the noise source is close to the extension of the directional microphone and the near-end speaker, when the noise is large, the above-described assumptions of the expressions (1) to (4) are satisfied. Cannot be done and the double-talk detection performance deteriorates.

(2) If the noise is large, even if there is no received signal, it is misunderstood that the receiving state is due to only the noise, and the coefficient of the adaptive filter is updated. This may cause deterioration of the erase amount.

(3) The reason why there is a difference between the powers of the captured signals of both microphones is that the sound from one near-end speaker is picked up by one directional microphone. Therefore, when there are a plurality of speakers at the near end and they are speaking at the same time, the double-talk detection performance deteriorates.

(4) Since a sub microphone is required in addition to the main microphone, and moreover, directivity, installation position, etc. are specified, the entire hands-free telephone becomes large.

The present invention has been made in consideration of the above points, and various call states including the double talk state can be determined with high accuracy regardless of the presence of noise, and the echo cancellation accuracy can be improved. It is intended to provide a small-sized echo canceller that can be used.

[0019]

In order to solve such a problem, according to the present invention, an adder removes an echo signal sneaking from a speaker to a microphone by using a pseudo echo signal formed by an adaptive filter. The following means are provided in the echo canceller provided in the telephone.

That is, it is provided on the transmission signal line, and outputs a significant detection signal in the double talk state and the transmission state based on the autocorrelation of the signal on the signal line, and also in the reception state and the silent state. Sometimes a sending voice detector that outputs a non-significant detection signal, and provided on the receiving signal line,
Based on the signal on the signal line, it outputs a significant detection signal in the double talk state and the receiving state, and outputs a non-significant detection signal in the transmitting state and the silent state. A call state determiner that determines a call state based on detection signals from the transmitted voice detector and the received voice detector, and a coefficient control unit that controls coefficient update of an adaptive filter based on the determined call state And.

Here, the echo canceling component consisting of the adaptive filter and the adder is multiplexed so that the time when the coefficient controller creates the coefficient in the adaptive filter of each echo canceling component is different, and the echo canceling component is different. The power comparator that determines the post-removal signal to be the transmission signal in the double-talk state by comparing the power of the post-removal signal output from the unit and the post-removal signal determined by the power comparator in the double-talk state are selected as the transmission signal. A transmission signal selecting means for selecting a signal after being removed by the echo removing component that is predetermined in the receiving state as a transmission signal and selecting a signal captured by the microphone as a transmission signal in the transmitting state and the silent state. preferable.

In this preferred embodiment, the coefficient of the adaptive filter in each echo canceling component is fixed in the double talk condition, except for the predetermined echo canceling component whose output signal is selected as the transmission signal in the receiving state. Is more preferable.

[0023]

In the present invention, the transmission voice detector outputs a significant detection signal in the double talk state and the transmission state based on the autocorrelation of the signal on the transmission signal line, and
The non-significant detection signal is output in the receiving state and the silent state, and the receiving voice detector outputs a significant detection signal in the double talk state and the receiving state based on the signal on the receiving signal line, An insignificant detection signal is output in the transmitting state and the silent state. Then, the call state determiner determines the call state based on the detection signals from the transmitted voice detector and the received voice detector, and the coefficient control unit determines the coefficient of the adaptive filter based on the determined call state. Control updates.

Here, as the signal on the transmission signal line, various combination signals of the transmission signal component, the echo component, and the noise component are conceivable. However, the transmission voice detector detects the self of the signal on the transmission signal line. Since the detection operation is performed based on the correlation, the detection accuracy is good. It should be noted that the one using autocorrelation can also be applied to the received voice detector.

As described above, since the transmitting voice detector and the receiving voice detector are provided, the silent state can be detected independently, and the operation suitable for the silent state can be executed. In addition, a microphone used only for determining the call state is unnecessary.

In order to appropriately control the echo canceling operation in accordance with the four call states, the echo canceling component consisting of the adaptive filter and the adder is multiplexed, and the coefficient in the adaptive filter of each echo canceling component is set as a coefficient. It is assumed that the time points created by the control unit are different, and the power comparator that determines the post-removal signal to be used as the transmission signal in the double-talk state by comparing the powers of the post-removal signals output from the echo cancellation components, and the double-talk In the state, the post-removal signal determined by the power comparator is selected as the transmission signal, and in the receiving state, the post-removal signal by the echo cancellation component that is predetermined is selected as the transmission signal, and the microphone is captured in the transmitting state and the silent state. It is preferable to provide transmission signal selecting means for selecting the selected signal as a transmission signal.

Here, in the double talk state, each echo canceling constituent unit is excluded, except for a predetermined echo canceling constituent unit whose output signal is selected as a transmission signal in the receiving condition so that the stable coefficient in the past can be used. It is preferable to fix the coefficient of the adaptive filter in (4) in the double talk state.

[0028]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a conceptual configuration of an embodiment of the present invention. First, the conceptual configuration and operation of the embodiment will be described with reference to FIG. It should be noted that FIG. 1 is not an accurate representation of the configuration of the embodiment, but is shown conceptually so that the greatest feature thereof becomes clear. The exact configuration of the embodiment will be clarified through the description of FIG.

Also in this embodiment, the basic structure and operation for removing echoes are the same as in the conventional case. That is, the received signal is given to the speaker 2 to be sounded, and is also input to the adaptive filter 4. The received voice (echo) y generated from the speaker 2 wraps around in the omnidirectional microphone 1, is captured by the microphone 1 and is given to the adder 3 as a subtracted input. When the reception signal is input, the adaptive filter 4 performs a convolution operation with the filter coefficient (estimated value of the echo path gain) at that time, and supplies the calculated value as a pseudo echo signal Y to the subtraction input terminal of the adder 3. Thus, the adder 3 outputs the signal e from which the echo component has been removed from the signal captured by the microphone 1, which is transmitted and also given to the coefficient control circuit 6. The coefficient control circuit 6 updates the coefficient of the adaptive filter 4 based on the signal e output from the adder 6 during the period in which the update of the coefficient of the adaptive filter 4 is allowed.

However, the configuration for controlling whether or not the coefficient of the adaptive filter 4 is updated is different from the conventional configuration.

That is, the transmitted voice detector 8, the received voice detector 10 and the call state judging device 11 are provided. The transmitted voice detector 8 detects whether or not there is a transmitted signal based on the output signal e from the adder 3 using autocorrelation, and gives the detection result to the call state determination device 11. The reception voice detector 10 detects whether or not there is a reception signal using autocorrelation, and gives the detection result to the call state determination device 11. Based on the detection results of the transmitted voice detector 8 and the received voice detector 10, the call state determiner 11 determines whether the call state is a double talk state, a transmitted state, a received state, or a silent state. The coefficient control unit 6 when in the receiving state
The coefficient updating operation by the coefficient control unit 6 is permitted, the coefficient updating operation by the coefficient control unit 6 is permitted or prohibited in the double talk state, and the coefficient updating operation by the coefficient control unit 6 is prohibited in other states.

Incidentally, conventionally, the silent state cannot be determined alone, but is determined as the receiving state.

Transmitted voice detector 8 and received voice detector 10
As described above, the use of autocorrelation detects the presence of a transmission signal or a reception signal. This detection method will be described in detail.

When the speaker signal (transmitting signal or receiving signal) at the sampling period i is ai, the autocorrelation function S (j) when shifted by j times the sampling period is expressed by the following equation.

S (j) = Σ ai · ai + j / Σ ai · ai (9) Here, the sum of the two is such that i is 1 to N (a period longer than the maximum possible pitch of the speaker signal ai). Value corresponding to). The transmitted voice detector 8 and the received voice detector 10 change a plurality of autocorrelation functions S (j) by changing the parameter j within a predetermined range centered on a value corresponding to the average pitch of the speaker signal. Ask. Then, for example, the maximum value Smax (j) of the plurality of autocorrelation functions S (j) is obtained,
The maximum value Smax (j) is compared with a predetermined threshold value, and when the maximum value Smax (j) is large, it is determined that a speaker signal (sending signal or receiving signal) exists, and the maximum value Smax (j) is determined.
Is small, it is determined that there is no speaker signal (sending signal or receiving signal).

As is well known, the audio signal has a periodicity, and the autocorrelation function corresponding to the period has a very large value. On the other hand, noise is generally random, and a plurality of autocorrelation functions obtained by changing the parameter j have small values. Therefore, the autocorrelation function can be used to distinguish between noise and speaker signals.

Here, the signal input to the transmission voice detector 8 changes depending on the call state, switching of the call state, and the like. For example, immediately after the talk state is changed to the double talk state, when the call state determiner 11 continuously determines the talk state, a signal obtained by adding the echo component of the received signal to the talk signal is the detector 8 Entered in. Both the transmitted signal and the received signal have a high autocorrelation, and the signal obtained by adding the echo component of the received signal to the transmitted signal also shows a slight decrease in value, but the power of the transmitted signal is large, so the autocorrelation is still high. Is. Further, for example, immediately after changing from the silent state to the receiving state, when the call state determining unit 11 determines that the state is still silent, a signal to which the echo component of the receiving signal is added is input to the detector 8. Strictly speaking, there are many echo paths, and since the power ratio between the echo component and the noise component is small, the autocorrelation of the echo component is small even if the received signal has pitch characteristics. According to the same examination, the transmission voice detector 8 can output a detection signal that is easy in the transmission state and the double talk state by appropriately selecting the threshold value.

Further, even when there are a plurality of near-end speakers uttering at the same time, the synthesized voice has periodicity, and the transmission voice detector 8 can detect the presence of the transmission signal.

On the other hand, since the received voice detector 10 has only two states, that is, the reception signal is input or not input, the determination can be easily performed.

A well-known method, for example, an algorithm such as a learning identification method can be applied to the tap coefficient updating method executed by the coefficient control unit 6 described above. FIG. 3 shows an example of changes in the echo removal amount when there is only a received signal. The echo removal amount is an example in which 10 · log (input power of adder 3 / output power of adder 3) is defined. A constant value v1 of the amount of echo consumption from the time of startup from the silent state
Until it comes to take, the tap coefficient is updated by performing sequential processing by an algorithm such as a well-known learning identification method.
Even after the echo consumption amount takes a value near a constant value v1, the tap coefficient is updated every sampling period.

FIG. 4 shows in detail the components of the adder 3 and the adaptive filter 4 according to this embodiment. In FIG. 4, the adaptive filter portion of this embodiment is adapted to generate two types of pseudo echo signals Y1 and Y2 simultaneously or selectively, which is also a feature of this embodiment.

In order to generate the first pseudo echo signal Y1, the delay element group 55 (551 to 55 (n-1)), the coefficient multiplier group 54 (541 to 54n) and the adder group 52 (5).
21-52 (n-1)) are provided. Further, the switch 6 for opening and closing the path of the coefficient given to the coefficient multiplier group 54.
3 and a switch 62 for opening and closing the output path of the first pseudo echo signal.

In order to generate the second pseudo echo signal Y2, the delay element group 55 and the coefficient multiplier group 53 (531 to 53)
n) and an adder group 51 (511 to 51 (n-1)) are provided, and the delay element group 55 has a common configuration for generating the first pseudo echo signal. Also, the second pseudo echo signal Y
Switch group 56 (561
56n) are provided.

The switch groups 54 and 56, the switch 6
2, 63 and switches 60, 61 described later are controlled by the coefficient control unit 6 and / or the power comparator 59 described later, and the control content will be clarified in the operation description described later.

The received signal (received signal series data) from the far-end speaker is input to one sampling period delay elements 551 to 55 (n-1) connected in cascade, and one sampling extracted from n taps. Each tap data having a different timing for each cycle has corresponding variable coefficient multipliers 541 to 541.
54n. Each variable coefficient multiplier 541-54n
Each of the input tap data is multiplied by the coefficient set at that time, and the corresponding adder 521 to 52 (n-
Give to 1). The adders 521 to 52 (n-1) are connected in cascade, and the total sum of the tap data multiplied by the coefficient is calculated for the entire adder group 52. The final stage adder 52 (n) The total sum data output from (-1) is given as a first pseudo echo signal Y1 to the echo removing adder 58 as a subtraction input when the switch 62 is closed. The coefficients of the variable coefficient multipliers 541 to 54n are updated by the coefficient control unit 6 every sampling period when the switch 63 is closed (a part of the double talk state and the receiving state).

Switches 561-5 that open and close in conjunction with each other
In the state where 6n is closed, the delay element group 55
The tap data of different timings for each sampling period extracted from the n taps of the variable taps are output to the variable coefficient multipliers 531 to 5n via the corresponding switches 561 to 56n.
Given to 3n. Each variable coefficient multiplier 531 to 53n
Each of the input tap data is multiplied by the coefficient set at that time, and the corresponding adder 511 to 51 (n-
Give to 1). The adders 511 to 51 (n-1) are connected in cascade, and the sum of the tap data multiplied by the coefficient is calculated for the entire adder group 51. The final stage adder 51 (n -1) The sum total data output from -1) is used as the second pseudo echo signal Y2, and the echo removing adder 5
7 as a subtraction input.

Here, in the receiving state, the coefficients are transferred from the variable coefficient multipliers 541 to 54n for forming the first pseudo echo signal Y1 to the variable coefficient multipliers 531 to 53n, This transfer cycle is a predetermined time T
2 times the time. In other call states, the variable coefficient multipliers 531 to 53n maintain the set coefficients.

FIG. 5 is an explanatory diagram showing the relationship between the detection response time T1 of the transmitted voice detector 8 and the predetermined time T2 that defines the transfer cycle. From the receiving state to the double talk state in which the coefficient update is not performed. It shows the case of migration. Even if the state changes to the double talk state at the time point B, the transmitted voice detector 8 cannot immediately detect it, but does detect it for the first time at the time point A after some delay. A predetermined time is set to a time T2 that is longer than the detection response time T1 (which slightly changes depending on the signal content), and is set to the variable coefficient multiplier coefficients 541 to 54n at every 2 * T2, which is twice the time in the listening state. The transfer operation is performed so that the existing coefficient is set in the variable coefficient multipliers 531 to 53n. That is, in the period T2 (k + 1) before the transfer operation, the period T2 (k
The coefficients transferred from the variable coefficient multipliers 541 to 54n to (-1) are set in the variable coefficient multipliers 531 to 53n. Therefore, the variable coefficient multiplier 5 can be changed depending on the state switching.
Even if the coefficients of 41 to 54n are disturbed, the variable coefficient multipliers 531 to 531
The coefficient of 53n can be stable.

As described above, the two types of pseudo echo signals Y
The reason why 1 and Y2 are generated is that an echo is to be removed by a good pseudo echo signal even in a call state in which at least one coefficient is not updated.

The capture signal P by the microphone 1 is subtracted from the echo removing adder 58 to which the first pseudo echo signal Y1 is input and the echo removing adder 57 to which the second pseudo echo signal Y2 is input. The echo removing adder 58 or 57, which is provided as an input, subtracts the first or second pseudo echo signal Y1, Y2 from the captured signal P, and outputs the signal after the subtraction as the first and second transmission candidate signals. Output as E1 and E2.

These first and second transmission candidate signals E1 and E2 become transmission signals through the switch 60 or 61 which is closed at that time and are given to the transmission line and the transmission voice detector 8. .

Further, the first and second transmission candidate signals E1 and E2 are given to the power comparator 59. The power comparator 59 obtains an average power of the first and second transmission candidate signals E1 and E2, and a value obtained by adding an offset to the average power of the first transmission candidate signal and the average power of the second transmission candidate signal. It compares the magnitude with and, according to the comparison result,
It controls the opening and closing of the switches 60, 61, 63. The switch control by the power comparator 59 is designed to function effectively only when the determination result of the call state determiner 11 is the double talk state.

The operation of the echo canceller of the embodiment having the above configuration will be described below depending on the case in a call state.

(A) Receiving state In the receiving state where there is only the receiving signal, the receiving voice detector 10 gives a significant detection signal to the call state judging device 11,
The transmitted voice detector 8 does not output a significant detection signal.

Thus, the call state judging device 11 judges that the call is in the receiving state and gives the judgment result to the coefficient control unit 6. At this time, the coefficient control unit 6 closes the switches 61 to 63 and opens the switch 60 and the switch group 56. Therefore, in the receiving state, only the first pseudo echo signal Y1 is formed and given to the adder 58, and the microphone 1
The echo is removed by subtracting the first pseudo echo signal Y1 from the captured signal P from the signal, and the signal (including only the removal residual) E1 after removal is transmitted as a transmission signal. The transmitted voice detector 8 to which such a transmission signal E1 is applied continuously outputs an insignificant detection signal.

The signal E1 after removal, which contains only the removal residual, is
The coefficient control unit 6 is supplied to the coefficient control unit 6, and updates the coefficients of the variable coefficient multipliers 541 to 54n every sampling cycle. The coefficients of the variable coefficient multipliers 541 to 54n are
The variable coefficient multiplier 531 to the variable coefficient multiplier 531 for each time twice the predetermined time T2
The coefficient which has been transferred to the variable coefficient multiplier 53n and has been set in the variable coefficient multipliers 541 to 54n in the past is set to the multipliers 531 to 53n.

As described above, in the receiving state, the first
The echo removing operation using the pseudo echo signal Y1 and the coefficient updating operation are repeatedly executed.

(B) Double-Talk State In the double-talk state in which both the transmission signal and the reception signal are present, the reception voice detector 10 gives a significant detection signal to the communication state judging device 11, and the transmission voice detector 8 Also provides a significant detection signal to the call state determiner 11. Although the captured signal P captured by the microphone 1 also includes an echo component, the transmitted voice detector 8 outputs a significant detection signal because it performs the detection operation using the autocorrelation function as described above.

Thus, the call state determiner 11 determines the double talk state and gives the determination result to the coefficient controller 6. Immediately after the determination of the double talk state, the coefficient control unit 6 opens the switch 63 so that the coefficient update is not performed, and closes the switch 62 and the switch group 56 to close the first and second pseudo states. Echo signal Y
Both 1 and Y2 are supplied to the adders 58 and 57. In the double talk state, the switch 60
The control of switches 61 and 61 is entrusted to the power comparator 59, and the control of the switch 63 is entrusted to the power comparator 59 when the initial state is exceeded.

Therefore, immediately after the switching to the double talk state, the first transmission candidate signal (removal residual and transmission) in which the echo is removed by subtracting the first pseudo echo signal Y1 from the captured signal P by the microphone 1 E1)
And a second transmission candidate signal (which is composed of a removal residual and a transmission signal component) E2 in which the echo is removed by subtracting the second pseudo echo signal Y2 from the capture signal P captured by the microphone 1 to the power comparator 59. Is entered. The power comparator 59 compares the average power of the first transmission candidate signal E1 with a value obtained by adding an offset to the average power of the second transmission candidate signal E2. If the former is larger, it is determined that the removal residual component included in the first transmission candidate signal E1 is larger, the switch 60 is closed, and the switch 61 is closed.
Is opened to select the second transmission candidate signal E2 having the smaller removal residual component as the transmission signal. On the contrary, when the latter is large, it is judged that the removal residual component included in the first transmission candidate signal E1 is smaller, and the switch 61 is closed and the switch 60 is opened to remove the removal residual component. The first transmission candidate signal E1 having a small component is selected as the transmission signal. Further, when the first transmission candidate signal E1 is selected in this way, the switch 63 is closed and the variable coefficient multiplier 5
The coefficients of 41 to 54n are updated. However, the coefficient updating of the variable coefficient multipliers 531 to 53n is not performed.

As described above, the switch 63 is opened in the initial state, but thereafter, the switch 63 is closed or opened depending on the comparison result of the power comparator 59.

It is to be noted that the reason for determining the magnitude by adding the offset is that the coefficient relating to the latest estimated value of the impulse response of the echo path is set in the variable coefficient multipliers 541 to 54n, and therefore the first pseudo echo signal This is because E1 is given a little priority over the second pseudo echo signal E2. Also, the first
The transmission candidate signal E1 is mainly composed of the transmission signal component in the double talk state. However, based on this signal, the coefficients of the variable coefficient multipliers 541 to 54n are updated depending on the case. For the reason. This is because it is possible to instantly improve the echo removal characteristics before the reception state is determined by the change from the double talk state to the reception state, and when the double talk period becomes long, the variable coefficient multiplier 531 This is because the coefficients of ~ 53n and 541-54n are no longer valid and it may be possible to obtain a better output signal by updating in this way.
This is to prevent howling in which the transmission candidate signal E1 and the second transmission candidate signal E2 are alternately and repeatedly selected in a short cycle.

Since most of the transmission signals selected and formed in the double talk state are transmission signals, the transmission voice detector 8 continuously outputs a significant detection signal in this state.

As described above, in the double talk state, the echo removing operation using the first pseudo echo signal Y1 and the echo removing operation using the second pseudo echo signal Y2 are performed in parallel, The one with better removal is selected to be a transmission signal, and the coefficient updating operation is controlled according to the selected content.

(C) Speaking state In a speaking state in which there is only a speaking signal, the receiving voice detector 10 gives an insignificant detection signal to the speaking state judging device 11, and the transmitting voice detector 8 is significant. The detection signal is given to the call state judging device 11.

In this way, the call state judging device 11 judges that it is in the transmitting state and gives the judgment result to the coefficient control unit 6. At this time, the coefficient control unit 6 closes only the switch 61 and opens the other switches 60, 62, 63 and the switch group 56. Therefore, the first pseudo echo signal Y1 input to the adder 58 is 0, and the captured signal (only the transmission signal) P by the microphone 1 becomes the transmission signal as it is via the adder 58 and the switch 61. . Therefore, the transmitted voice detector 8 continuously gives a significant detection signal to the call state determiner 11. In the transmitting state, there is no echo component of the received signal, so it is not necessary to remove the echo.

Here, since all the switches other than the switch 61 are open, the coefficient updating operation is not performed, and the variable coefficient multipliers 531 to 53n, 541 to 541 are opened by opening the other switches. The coefficient held in 54n is protected without being affected by noise or a received signal.

As described above, in the transmitting state, the captured signal P of the microphone 1 is directly used as the transmitting signal without performing echo removal, and the variable coefficient multiplier 531 is used.
The coefficient updating operation for .about.53n and 541 to 54n is stopped.

(D) Silence state In the silence state in which neither the transmission signal nor the reception signal exists,
The received voice detector 10 gives a non-significant detection signal to the call state determiner 11, and the transmitted voice detector 8 also gives a non-significant detection signal to the call state determiner 11.

In this way, the call state judging device 11 judges that there is no sound, and gives the judgment result to the coefficient control unit 6. At this time, the coefficient control unit 6 closes only the switch 61 and opens the other switches 60, 62, 63 and the switch group 56. That is, the same control as in the transmitting state is performed. Therefore, the captured signal by the microphone 1 (only noise)
P directly becomes a transmission signal via the adder 58 and the switch 61. Therefore, the transmitted voice detector 8 continuously supplies the insignificant detection signal to the call state determiner 11. Even in the silent state, there is no echo component of the received signal, so there is no need to perform echo removal. Also in this case, the switch 61
Since all the switches other than the above are open, the coefficient updating operation is not performed, and the variable coefficient multipliers 531 to 53n, 54 are opened by opening the other switches.
The coefficients held in 1 to 54n are protected without being affected by noise or a received signal.

In the above-mentioned conventional method, the silent state and the receiving state are handled in the same manner (although the silent state is also determined as the receiving state), the silent state is processed in the same manner as the transmitting state in this embodiment. There is. Both the silent state and the speaking state have the common point that the captured signal P by the microphone 1 does not have an echo component, and therefore can be treated in the same manner.
When the silent state is handled as in the receiving state, the inconvenience that the coefficient is updated occurs, and it is preferable to carry out as in this embodiment.

As described above, in the silent state, the captured signal P of the microphone 1 is directly used as the transmission signal without performing echo removal, and the variable coefficient multiplier 531 is used.
The coefficient updating operation for .about.53n and 541 to 54n is stopped.

According to the above embodiment, the received voice detector 10
Since the transmitted voice detector 8 detects the presence of a received signal or a transmitted signal using the autocorrelation function, and the call state determination unit 11 determines the call state based on these detection signals, the presence of noise is detected. However, the call state can be determined with high accuracy, the echo removing operation can be appropriately controlled, and the echo removing characteristic can be improved.

Further, according to the above embodiment, in the double talk state, the transmission signal is determined by selecting from the two types of post-removal signals that have been echo-removed using the two types of pseudo echo signals. Even when the state changes from the receiving state or the transmitting state to the double talk state or the coefficient is disturbed during the double talk state, the echo can be removed by the pseudo echo signal using the stable past coefficient. The echo removal characteristic can be improved.

Further, according to the above-described embodiment, the microphone used only for determining the call state is not required, the hands-free telephone can be downsized, and the positional flexibility of the provided microphone 1 can be increased.

Furthermore, according to the above-described embodiment, the silent state is discriminated from other call states, and the coefficient update of the adaptive filter is stopped in the silent state. It is possible to prevent unnecessary updating of the coefficient due to noise in the silent state.

In the above embodiment, the adaptive filter has a double structure. However, the present invention can be applied to a single structure and further has a triple structure or more. The present invention can also be applied to.

Further, in the above embodiment, the receiving voice detector 10 and the transmitting voice detector 8 both use the autocorrelation function to detect the presence of the receiving voice signal or the transmitting voice signal. As the received voice detector 10 in which the combination of the components of (3) is simple, one that does not use an autocorrelation function may be applied. For example, the power of the input signal may be obtained and the power may be compared with a detection threshold value to be detected.

Further, in the above embodiment, the coefficient of the variable coefficient multipliers 541 to 54n may be updated in the double talk state, but the coefficient update of the variable coefficient multipliers 541 to 54n may be completed in the double talk state. It may be prohibited.

Furthermore, from the variable coefficient multipliers 541 to 54n in the receiving state to the variable coefficient multipliers 531 to 53n.
The coefficient transfer period to the variable coefficient multiplier is not limited to that in the above embodiment as long as it is possible to ensure the stabilization of the coefficients set in the variable coefficient multipliers 531 to 53n. Alternatively, a large-capacity register may be provided in the coefficient control unit 6 so that the coefficient control unit 6 transfers data. In this case, the transfer cycle is one sampling cycle (however, the transferred coefficient is considerably before). You can also do

[0081]

As described above, according to the present invention, a detection signal which is provided on a transmission signal line and which is significant in the double talk state and the transmission state is based on the autocorrelation of the signal on the signal line. And a sending voice detector that outputs a non-significant detection signal in a receiving state and a silent state, and a double talk state and a receiving state provided on the receiving signal line based on the signal on the signal line. Output a significant detection signal at the same time, and outputs a non-significant detection signal at the time of the transmitting state and the silent state, the receiving voice detector, and the detection signals from the transmitting voice detector and the receiving voice detector. Based on the call state determiner for determining the call state and the coefficient control unit for controlling the coefficient update of the adaptive filter based on the determined call state, various call states including the double talk state are provided. The Can be determined with high accuracy regardless of the presence of the figure can be realized compact echo canceler can be improved echo cancellation accuracy.

[Brief description of drawings]

FIG. 1 is a block diagram showing a conceptual overall configuration of an embodiment.

FIG. 2 is a block diagram showing a conventional configuration.

FIG. 3 is a diagram for explaining a coefficient updating method according to an embodiment.

FIG. 4 is a block diagram showing a detailed configuration around an adaptive filter and an adder according to an embodiment.

FIG. 5 is an explanatory diagram of a detection response time of a transmission signal and a time difference between both coefficients according to the embodiment.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Microphone, 2 ... Speaker, 3, 57, 58 ... Adder, 4 ... Adaptive filter, 6 ... Coefficient control part, 8 ... Transmitted voice detector, 10 ... Received voice detector, 11 ... Call state determination device, 51, 52 ... Adder group, 53, 54 ... Variable coefficient multiplier group, 55 ... Delay element group, 56 ... Switch group, 59 ... Power comparator, 60-63 ... Switches.

Claims (3)

[Claims] 1. An echo canceller provided in a hands-free telephone, in which an adder removes an echo signal sneaking into a microphone from a speaker using a pseudo echo signal formed by an adaptive filter, is provided on a transmission signal line. , Based on the autocorrelation of the signal on the signal line, a significant detection signal is output in the double talk state and the transmitting state, and a non-significant detection signal is output in the receiving state and the silent state. It is provided on the voice detector and the receiving signal line, and outputs a significant detection signal in the double talk state and the receiving state based on the signal on the signal line, and it is insignificant in the transmitting state and the silent state. In order to output a different detection signal, the reception voice detector, the transmission voice detector, and the detection signal from the reception voice detector are used to determine the call state. An echo canceller, comprising: a talk state determiner; and a coefficient control unit for controlling coefficient update of the adaptive filter based on the determined call state. 2. The echo canceling configuration section comprising the adaptive filter and the adder is multiplexed, and the time points at which the coefficient control section creates the coefficients in the adaptive filter of each echo removing configuration section are different, and A power comparator that determines the post-removal signal to be the transmission signal in the double-talk state by comparing the powers of the post-removal signals output from the echo removal components, and the post-removal signal determined by the power comparator in the double-talk state Transmission signal selection, which is selected as a transmission signal, is selected as a transmission signal after being removed by an echo canceling component that is predetermined in the reception state, and is selected as a transmission signal by the microphone in the transmission state and the silent state. The echo canceller according to claim 1, further comprising: 3. The coefficient of the adaptive filter in each echo cancellation component is fixed in the double talk condition except for the predetermined echo cancellation component whose output signal is selected as the transmission signal in the listening state. The echo canceller according to claim 2, wherein