JP2012105115A - Echo canceller, echo cancellation program, and telephone apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an echo canceller, echo cancellation program, and a telephone apparatus for removing an echo generated on a telephone apparatus stably and continuously.SOLUTION: An echo canceller removes an echo component from near-end input signals. The echo canceller includes: echo cancelling means for generating a signal cancelling an echo component of near-end input signals using a pseudo echo generated by pseudo echo generation means; sound switch means for attenuating the near-end input signals using an attenuator; switching means for selecting a post-cancellation signal generated by the echo cancellation means or a signal generated by the sound switch means, and sending the signal to a far-end side; call state determination means for determining the call state between a far-end speaker and a near-end speaker using the far-end input signal and the near-end input signal; and switch controlling means for controlling the switching means in response to the determination result of the call state determination means.

Description

  The present invention relates to an echo canceller, an echo cancellation program, and a telephone device, and can be applied to, for example, a telephone device that supports a hands-free function.

  In recent years, with the spread of VoIP, call charges have become cheaper and communication speeds have become faster, so telephone conference devices and videophone devices, which have not been so common in the past, are rapidly spreading.

  As such telephone devices become widespread, new technical demands are generated as the number of users increases. Conventionally, since teleconference was an expensive and special method of use, devices used by users who can use such devices emphasize the convenience of users, rather than those used by general users like today. However, much larger and more expensive conference systems have been used.

  One of the reasons for making these conference systems expensive is the implementation of a hands-free function. This is because, as described in Non-Patent Document 1, it is necessary to introduce a function for preventing the sound of the speaker from being re-input to the microphone and generating howling and making the call impossible. It is.

  There are also users who want to make a conference call but attach importance to cost rather than usability. For example, there is a case where the cost for the meeting place is given first priority and the user's usability is lowered. In this case, instead of using an expensive apparatus as described above, a user who wants to realize a hands-free state has been holding a conference with a headset during the conference. However, wearing a headset puts a considerable burden on the user, such as fatigue due to a sense of incongruity, and damage to the skin caused by rubbing the earpad and other parts that touch the ear.

  Users during the meeting may take notes during the meeting or write down some of the distributed materials, so the need for a hands-free function that frees both hands is natural and essential It's a feature, and the demand will never stop.

  Furthermore, in recent years, so-called telework, which does not go to the workplace and works at satellite offices near homes, has become widespread from the viewpoints of diversity and efficiency of work styles and ensuring business continuity of companies in the event of various disasters. I'm starting. The spread of telework can be expected to spread rapidly in the future. In the case of such telework, it is fully expected that the main form of the conference will be telephone conference and video conference via a communication device rather than direct dialogue.

  In addition, systems that always connect offices with video and audio to eliminate the sense of geographical division in the office have begun to be used, and there is an increasing demand for remote users to communicate naturally. is expected. On the other hand, for example, in the extension of the prior art, a user who only needs to wear a headset only during a meeting is required to wear the headset all the time.

  If a hands-free state is maintained with the above-described constant connection, all users are forced to wear a headset at all times or use an expensive device. Conventionally, a voice switch as described in Non-Patent Document 2 has been used as a countermeasure to such a problem.

  The voice switch described in Non-Patent Document 2 compares the power of each signal line for transmission and reception, transmits the signal with the higher power, and blocks the signal with the lower power. The method using the voice switch of Non-Patent Document 2 can create a device at a relatively low cost, and is currently useful for realizing a wide and inexpensive hands-free device.

  However, the voice switch technology described in Non-Patent Document 2 inevitably involves switching of a one-way call, so that one of the calls is blocked until the detection switching reaction time. Since normal conversations are alternated, the previous call direction is maintained until the detection time of reversal detection switching until the call direction is reversed, so the first part of the utterance is often lost. . In order to remedy this drawback, Patent Document 1 discloses a technique for providing an intermediate state between the transmission state and the reception state in addition to the transmission state and the reception state so as to reduce the first and last missing speech.

  However, in the technique described in Patent Document 1, attenuation is applied to both the transmission path and the reception path in the intermediate state described above. In other words, since the signal that should be transmitted originally is attenuated as a price to reduce the first and last deficits in the speech, the volume of the call voice increases and decreases sharply, and the user feels tired with difficulty in hearing. Met.

  On the other hand, an echo canceller technique that can eliminate the defect that the first part of the speech of the voice switch is lost is already known. The echo canceller has an adaptive filter composed of a signal processor, etc., creates a pseudo echo by calculation processing using the received signal and the echo re-input to the microphone, cancels the echo and the pseudo echo, and transmits from the speaker to the microphone. This removes the re-input signal and removes howling.

  However, the echo canceller presupposes that the acoustic path (echo path) from the adaptive filter to the microphone is fixed. Therefore, the conventional echo canceller has a drawback in that if the echo path is changed due to the speaker's comforting or moving, the echo cancellation performance is severely deteriorated and the echo cannot be erased until re-learning. Furthermore, since the conventional echo canceller mixes echo and the voice of the speaker, even the voice is degraded. Therefore, it is necessary to have a highly accurate call state determination function. There was a need to stop the coefficient update.

  Furthermore, since the conventional echo canceller can create a pseudo echo only when both the echo and the received signal can be obtained with high accuracy, there is a problem that the effect is greatly deteriorated when the background noise on the microphone side (near end side) is large. .

  As a technique for solving the above problems, there are conventional techniques described in Patent Documents 2 to 4.

  Japanese Patent Application Laid-Open No. 2004-228561 discloses a technique for utilizing the advantages of both a voice switch and an echo canceller.

  Patent Document 3 discloses a technique in which an echo canceller and a voice switch are used together to compensate for the performance deterioration of the echo canceller by a voice switch provided at the output stage of the echo canceller.

  Further, Patent Document 4 discloses a technique that uses only the voice switch but improves the switching characteristics of the voice switch by using the nature of the conversational voice to ensure the naturalness of the call.

Japanese Patent Laid-Open No. 11-88497 JP 2006-270147 A JP-A-7-170337 Japanese Patent Laid-Open No. 10-308816

Edited by ARL, Inc., “Acoustic technology and audio / AV terminology and capture”, [Online], INTERNET, [October 26, 2010 search], <URL: http: //www.ari-web. com / sound / measurement / note.htm> ITU (INTERNATIONAL TELECOMMUNICATION UNION), “ECHO SUPPRESSORS ITU-T Recommendation G.164”, [Online], INTERNET, [October 26, 2010 search], <URL: http://www.itu.int/rec /T-REC-G.164-198811-I/en>

  However, in the method of Patent Document 2, since the voice switch is provided on the echo path as seen from the echo canceller, the echo path is severely non-linearly changed for the echo canceller. Consequently, the effect of the echo canceller combined with the voice switch is effective. It was difficult to realize a two-way call.

  In the technique described in Patent Document 3, the echo canceller and the voice switch are cascade-connected except during double talk. Therefore, in the technique described in Patent Document 3, the echo cancellation amount in the echo path variation is deteriorated in the preceding echo canceller, and the cancellation amount is deteriorated due to the real double talk voice (the microphone side speaker voice in the opposite direction). Therefore, the echo remaining power generated by the change in the echo path is regarded as the echo canceller output power as it is. As a result, the technique described in Patent Document 3 has a problem that an echo generated due to a change in the echo path is transmitted to the far end side without being subjected to any suppression processing without being distinguished from the microphone-side speaker voice. The problem is that acoustic echo path movement always occurs as a measure for suppressing acoustic echo and howling in always-on communication, which is a fatal defect in the stability of the communication system.

  Furthermore, although the technique described in Patent Document 4 can reduce switching flutter during voice, it has not been possible to avoid the fatal drawback of the voice switch that simultaneous calls are still impossible.

  Therefore, in view of the above-described problems, an echo canceller, an echo cancellation program, and a telephone device that can stably and continuously remove echoes generated in the telephone device are desired.

  The first aspect of the present invention is an echo canceller for removing an echo component from a near-end input signal. (1) A signal obtained by canceling the echo component of the near-end input signal by using a pseudo echo generated by a pseudo-echo generating means. (2) a voice switch means for generating a signal obtained by attenuating the near-end input signal by an attenuator, and (3) a post-cancellation signal generated by the echo cancellation means, or A switch unit that selects one of the signals generated by the voice switch unit and transmits the signal to the far end side; and (4) a far-end speaker using the far-end input signal and the near-end input signal. A call state determination unit that determines a call state with a near-end speaker, and (5) a switch control unit that controls the switch unit according to a determination result of the call state determination unit. To do.

  The echo cancellation program according to the second aspect of the present invention provides a computer installed in an echo canceller that removes an echo component from a near-end input signal by using (1) a pseudo echo generated by a pseudo-echo generating means, and Echo canceling means for generating a signal obtained by canceling the echo component of the signal; (2) voice switch means for generating a signal obtained by attenuating the near-end input signal by an attenuator; and (3) the echo canceling means. Switch means for selecting one of the post-cancellation signal generated by the voice signal or the signal generated by the voice switch means and sending it to the far end side; and (4) the far end input signal and the near end input signal. A call state determination unit that determines a call state between a far-end speaker and a near-end speaker, and (5) controls the switch unit according to a determination result of the call state determination unit. Characterized in that to function as switch control means.

  A telephone device according to a third aspect of the present invention includes the echo canceller according to the first aspect of the present invention.

  According to the present invention, echo generated in the telephone device can be stably and continuously removed.

It is a block diagram which shows the functional structure of the telephone apparatus which concerns on 1st Embodiment. It is a block diagram which shows the functional structure of the adaptive filter which concerns on 1st Embodiment. It is a block diagram which shows the functional structure of the telephone apparatus which concerns on 2nd Embodiment. It is the graph shown about the example of the calculation result of the echo loss estimation part which concerns on 2nd Embodiment. It is explanatory drawing (1) shown about operation | movement of the echo loss estimation part which concerns on 2nd Embodiment. It is explanatory drawing (2) shown about operation | movement of the echo loss estimation part which concerns on 2nd Embodiment. It is a block diagram which shows the functional structure of the telephone apparatus which concerns on 3rd Embodiment. It is a block diagram which shows the functional structure of the telephone apparatus which concerns on 4th Embodiment. It is a block diagram which shows the functional structure of the signal mixing part which concerns on 4th Embodiment.

(A) First Embodiment Hereinafter, an echo canceller, an echo cancellation program, and a telephone device according to a first embodiment of the present invention will be described in detail with reference to the drawings.

(A-1) Configuration of the First Embodiment FIG. 1 is a block diagram showing the overall configuration of the telephone device 10 of the first embodiment.

  The telephone device 1 includes an echo canceller 10, a digital / analog converter (hereinafter referred to as “D / A”) 20, an analog / digital converter (hereinafter referred to as “A / D”) 30, a speaker 40, and a microphone 50. ing.

  In FIG. 1, the left side represents the far end side, and the right side represents the near end side. Although not shown in FIG. 1, the telephone device 1 is assumed to have other components necessary for the telephone device, such as a push button for making a call. In FIG. 1, the telephone device 1 is described as a hands-free telephone device, but the type of telephone device applied to the telephone device 1 is not limited.

  The echo canceller 10 is implemented in a program execution configuration such as a CPU, a ROM, a RAM, an EEPROM, and a hard disk (not limited to one, but may be configured so that a plurality of units can be distributedly processed). 1 may be constructed by installing an echo canceling program or the like, and even in that case, it can be functionally shown as in FIG. For example, when the telephone device 1 is constructed as a so-called soft phone on an information processing device such as a personal computer or a mobile phone terminal, the echo cancellation program of the embodiment is installed in the information processing device. Thus, the echo canceller 10 may be constructed.

  The far-end input signal x (n) received by the telephone device 1 from the far-end side is input to the D / A converter 20 via the reception input terminal Rin and the reception output terminal Rout, and the D / A converter 20 Converted to an analog audio signal. And the speaker 40 converts the electric signal given from the D / A converter 20 into an acoustic signal, and provides it to a near-end speaker. Note that “n” in x (n) represents the sample order. In the following, “n” used when expressing a signal in the same manner also represents the order of samples.

  When an acoustic signal is input to the microphone 50 by the near-end speaker uttering, the microphone 50 converts the acoustic signal into an electrical signal, and the electrical signal is given to the A / D converter 30. In the A / D converter 30, the electrical signal (analog signal) given from the microphone 50 is digitally converted and input to the echo canceller 10 via the transmission input terminal Sin as a near-end input signal. In the echo canceller 10, the near-end input signal is input to the echo canceller 10 via the transmission input terminal Sin, and after echo suppression processing is performed by the echo canceller 10, the far-end side signal is transmitted via the transmission output terminal Sout. Is output towards

  The acoustic signal output from the speaker 40 is input (mixed) into the microphone 50 together with the voice uttered by the near-end speaker via a path (echo path) that is reflected or propagates through space.

  The echo mixed in the microphone 50 is converted into an analog electric signal by acoustoelectric conversion, and is digitally converted by the A / D converter 30 to become an echo y (n), which is input to the echo canceller 10. The echo canceller 10 performs processing for suppressing the component of this echo y (n) mixed in the near-end input signal.

  Next, the internal configuration of the echo canceller 10 will be described.

  The echo canceller 10 includes an adder 101, an adaptive filter (hereinafter also referred to as “ADF”) 102, a bidirectional communication detector (hereinafter also referred to as “DTD”) 103, a control unit 104, a delay unit 105, a delay unit 106, An attenuator 107 and a switch 108 are included.

  The adder 101 receives the pseudo echo y ′ (n) generated by the adaptive filter 102 and the echo y (n) mixed in the near-end input signal output from the A / D converter 30 by a process described later. The signal e (n) that has been canceled by addition and the echo y (n) component is suppressed is provided to the bidirectional communication detector 103, the delay unit 106, and the adaptive filter 102.

  The delay unit 106 holds and delays the signal e (n) given from the adder 101 for a predetermined time and then gives it to the switch 108.

  The delay unit 105 supplies, to the attenuator 107, a signal yd (n) that is supplied from the A / D converter 30 and delayed by holding the near-end input signal mixed with the echo y (n) for a predetermined time. give.

  The attenuator 107 generates a signal att (n) obtained by attenuating yd (n) given from the delay unit 105 by a predetermined attenuation amount, and supplies the signal att (n) to the switch 108.

  The switch 108 switches either the signal given from the delay unit 106 or the signal given from the attenuator 107 and sends it to the far end side (transmission output terminal Sout).

  In the switch 108, the terminal a1 is connected to the output of the delay unit 106. In the switch 108, the terminal b1 is connected to the output of the attenuator 107. Further, in the switch 108, the terminal c1 is connected to the transmission output terminal Sout. That is, in the switch 108, the connection to the terminal c1 is switched to either the terminal a1 or the terminal b1. The switch 108 performs the above-described switching by the control unit 104.

  The control unit 104 controls switching of the switch 108 in accordance with a determination result given from a two-way call detector 103 described later. Specific control contents will be described in detail in an operation description described later. .

  Next, details of the adaptive filter 102 will be described.

  FIG. 2 is a block diagram showing a functional configuration inside the adaptive filter 102.

  The adaptive filter 102 includes a pseudo echo creation unit 201, a coefficient update amount calculation unit 202, and a step gain calculation unit 203.

  The pseudo echo creation unit 201 generates a pseudo echo y ′ (n) using the far-end input signal x (n) and the filter coefficient obtained by the coefficient update amount calculation unit 202. The pseudo echo creation unit 201 201 includes a holding register 2011, a product-sum operation unit 2012, and a coefficient unit 2013. Note that the processing of the pseudo echo creating unit 201 will be described in detail in an operation description to be described later.

  The coefficient update amount calculation unit 202 uses the content fed back from the pseudo echo creation unit 201 (holding register 2011) and the step gain obtained by the step gain calculation unit 203 to supply to the pseudo echo creation unit 201. The filter coefficient of is obtained. Note that the processing of the coefficient update amount calculation unit 202 will be described in detail in the description of operations described later.

  The step gain calculation unit 203 obtains a step gain to be supplied to the coefficient update amount calculation unit 202 by using a determination result by the bidirectional call detector 103 described later and the signal e (n). Note that the processing of the pseudo echo creating unit 201 will be described in detail in an operation description to be described later.

  Next, the bidirectional call detector 103 will be described.

  The two-way call detector 103 uses the far-end input signal x (n) and the signal e (n) from the adder 101 to talk on the telephone device 1 (between the far-end talker and the near-end talker). ) And supplies the determination result to the adaptive filter 102 and the control unit 104.

  The two-way call detector 103 is in a state where only the voice of the far-end speaker is present (only x (n) is voiced; hereinafter referred to as “far-end single talk state” or “ST”), State where both near-end speakers are speaking (x (n) and e (n) are voiced; hereinafter referred to as “double talk state” or “DT”), state where only near-end speakers are speaking (When only e (n) is in a sound state; hereinafter referred to as “near-end talk state” or “NT”), neither is speaking (when x (n) and e (n) are in a silent state; Hereinafter, either “silent state” or “IDL” is detected.

  The two-way call detector 103 also detects a state in which the echo path has changed (hereinafter referred to as “echo path change state” or “EPC”).

  The detection of these states by the two-way call detector 103 can use, for example, the technology described in “JP 2009-33344 A” (reference document 1), but is not limited to this. Of course, other methods may be used as long as they can be used correctly.

  In general, if the output of the adder is used as the input on the transmission side of the two-way call detector, it becomes possible to detect the transmission signal (ie, the near-end speaker signal) after roughly removing the echo component. It is known that the accuracy of call state detection is dramatically improved as compared with a call state detection method in which signal powers in a transmission path and a reception path are compared. Considering that both the echo and the near-end speaker signal are human speech signals, it is advantageous to detect the near-end signal after distinguishing and removing the echo. It is natural.

(A-2) Operation of the First Embodiment Next, the operation of the echo canceller 10 in the telephone device 1 of the first embodiment having the above configuration will be described.

(A-2-1) Operation of Adaptive Filter 102 First, the far-end input signal x (n) input from the receiving side input terminal Rin is stored in the holding register 2011 of the adaptive filter 102, and the following (1) It is treated as a data vector as shown in the equation (hereinafter also referred to as “received signal vector”). In the following equation (1), L is the number of taps of the adaptive filter 102. (L = 1024 here, but is not limited). In the following formula (1), t represents transposition of a matrix. In the following equation (1), a collection of received signal x (n) data having the current time n as a starting point and holding up to past L samples constitutes a received signal vector of equation (1). Represents.

X (n) = [x (n), x (n−1),..., X (n−L + 1)] ^t (1)
On the other hand, the coefficient unit 2013 stores data of filter coefficients (consisting of L coefficient values) supplied from the coefficient update amount calculation unit 202. As will be described later, the coefficient value constituting the filter coefficient is updated from time to time by the coefficient update amount calculation unit 202 using, for example, an NLMS (Normalized Last Mean Square) algorithm. The L coefficient values of the coefficient unit 2013 are also handled as one coefficient vector as in the above expression (1), and are expressed as in the following expression (2).

H (n) = [h (0), h (1),..., H (L−1)] ^t (2)
In the above equation (2), L is the number of filter taps. The product-sum operation unit 2012 uses the coefficient vector of the coefficient unit 2013 and the data vector of the holding register 2011 to calculate a vector product y ′ (n) as in the following equation (3). This y ′ ( n) is the sample data of the pseudo echo. Note that y ′ (n) is scalar value data having one value.

y ′ (n) = H ^t (n) × (n) (3)
The pseudo echo y ′ (n) created as described above is supplied to the adder 101. The adder 101 receives the echo y (n) digitally converted from the speaker 40 via the microphone 50 as described above.

  Then, the adder 101 outputs a residual signal e (n) obtained by adding and canceling signals as shown in the following equation (4) to the transmission output terminal Sout.

e (n) = y (n) −y ′ (n) (4)
Then, e (n) is sent out toward the far end via the transmission output terminal Sout.

On the other hand, e (n) that is the output of the adder 101 is supplied to the coefficient update amount calculation unit 202 via the step gain calculation unit 203. Then, the update amount calculation unit 202 uses e (n) and the signal x (n) output from the holding register 2011, for example, by a known NLMS algorithm such as the following equation (5). The filter system number of the filter 102 is updated.

  The above equation (5) indicates that an adaptive filter 102 coefficient vector to be used for the next pseudo echo creation is created from the samples e (n) and x (n). Here, α is a step gain and is a constant of 0 ≦ α ≦ 2. When α is large, the convergence time is short, but the steady echo cancellation performance is poor. If α is small, steady echo cancellation performance is large, but the convergence time is long.

  However, when a signal other than the far-end single talk state (ST) and the echo path fluctuation state (EPC) is input from the bidirectional call detector 103 to the step gain calculation unit 203, the step gain calculation unit 203 Is output to the coefficient update amount calculation unit 202 as α = 0, and the coefficient update is stopped.

  Here, in the coefficient update amount calculation unit 202, the coefficient update is stopped when α = 0, but a switch may be provided to cut off e (n) and stop the coefficient update. The coefficient update stop method is not limited to α = 0.

  The coefficient vector (filter coefficient) created as described above is output from the coefficient update amount calculation unit 202 to the coefficient unit 2013, and is used together with the input vector X (n) to create a pseudo echo according to the above equation (3).

  As the time advances, the calculation of the above equation (5) is executed several times and the coefficient update proceeds, and e (n) obtained by the above equation (4) gradually decreases. When e (n) becomes sufficiently small in the future, the coefficient update in the coefficient update of the above equation (5) hardly occurs. After the convergence of the adaptive filter 102, the echo can be canceled out with high accuracy by the adder 101 thereafter.

  Here, the NLMS algorithm as described above is used as the coefficient update algorithm of the adaptive filter 102, but the echo estimation algorithm may be any algorithm as long as it can estimate the echo, and is not limited to NLMS.

(A-2-2) Operation of bidirectional call detector 103 As described above, the bidirectional call detector 103 includes the voice signal x (n) from the far end and the signal e (n) from the adder 101. Is used to detect the far-end single talk state (ST), double-talk state (DT), near-end talk state (NT), silence state (IDL), and echo path variation state (EPC).

  In the first embodiment, detection of the far-end single talk state (ST) by the two-way call detector 103 is particularly important. In the first embodiment, paying attention to the fact that human conversation is mainly performed alternately, unlike conventional ones, a configuration in which echoes are canceled using a pseudo echo composed of an adaptive filter 102 and an adder 101 (hereinafter referred to as “echo”). , And a configuration for attenuating the near-end input signal composed of the delay unit 105 and the attenuator 107 (hereinafter also referred to as “voice switch”), but not connected in cascade. Connected in parallel.

  As a result, in the first embodiment, in the far-end single talk state (ST), howling is prevented by strongly attenuating the echo mainly using the voice switch part for the removal of the echo. On the other hand, bidirectionality is ensured by using an echo canceller, such as for a moment when a two-way call is necessary (in the case of a double talk state (DT)). In this way, the time for the echo canceller to intervene in the circulation path of the acoustic signal can be very short, so even if the echo canceller falls into a somewhat unstable operation, the echo circulates indefinitely along the call path. You can prevent it from growing for the first time.

(A-2-3) Operations of Control Unit 104, Switch 108, and Attenuator 107 Next, operations of the control unit 104, switch 108, and attenuator 107 will be described.

  The two-way call detector 103 supplies a detection result of the call state of the telephone device 1 to the control unit 104.

  When the far-end single talk state (ST) is given from the two-way call detector 103, the control unit 104 gives a control signal to the switch 108 so as to connect the terminal c1 to the terminal b1 side. .

  In the first embodiment, the control unit 104 outputs a control signal to the switch 108 so as to connect the terminal c1 to the terminal a1 side in a case other than the far-end single talk state (ST). In particular, the control unit 104 always controls the switch 108 to connect the terminal c1 to the terminal a1 side when the far-end single talk state (ST) ends.

  However, for example, in an environment where there is no near-end background noise, the control processing of the switch 108 by the control unit 104 is not limited to this. In such a case, for example, the control unit 104 may connect the terminal c1 to the terminal b1 side with respect to the switch 108 even in the silent state (IDL) and the echo path variation state (EPC). good. This is because the speech act itself is harmless even if the attenuator 107 is inserted in the silent state (IDL), and the adaptive filter 102 cancels out due to the follow-up deterioration of the adaptive filter 102 in the echo path fluctuation state (EPC). This is because echo suppression by the attenuator 107 may be preferable to the effect.

  However, as described above, in the far-end single talk state (ST), it is desirable that the control unit 104 always connects the terminal c1 to the terminal b1 side with respect to the switch 108. As an initial state of the switch 108, it is desirable that the terminal c1 is connected to the terminal a1 side.

  As already described, the delay unit 105 is supplied with the signal y (n), and the delay unit 105 adds a delay of the same amount as the time required for the state determination of the bidirectional call detector 103 to add yd ( n) and supplied to the attenuator 107.

  The attenuator 107 adds attenuation necessary for preventing howling and outputs the signal att (n) to the terminal b1 of the switch 108.

  Here, the attenuation amount of the attenuator 107 is 10 dB, but is not limited to this. However, in order to obtain a howling prevention effect, it is desirable to set an attenuation that at least cancels the gain of the echo path.

  As described above, in particular, the control unit 104 controls the switch 108 so as to connect the terminal to a1 when the far-end single talk state (ST) ends.

  Here, the delay unit 106 is described. It is desirable that the delay unit 106 is provided at the subsequent stage of the output of the echo canceller unit (adder 101) as viewed from the near end side. Further, the delay unit 105 is desirably provided in front of the attenuator 107 when viewed from the near end side.

  The reason why the delay unit 106 is provided after the output of the echo canceller unit (adder 101), that is, after the input point of the bidirectional call detector 103, is to prevent the determination of the bidirectional call detector 103 from being delayed. . The reason why the delay unit 105 is provided in the preceding stage of the attenuator 107 is to add attenuation to the signal reflecting the call state according to the determination result of the preceding two-way call detector 103.

(A-2-4) Overall Operation of the Echo Canceller 10 As described in detail so far, the echo canceller 10 for the purpose of suppressing echo and howling is basically a voice in a conversation between people. It pays attention to the fact that these exchanges are performed alternately.

  In the conventional echo canceller, it is assumed that the signal power is simply detected physically, and basically the presence of the signal occurs evenly in either case. For this reason, the conventional echo canceller needs to set the detection accuracy to a very high sensitivity even if it is aware of the erroneous detection, or sets a neutral state that is neither the call state on the transmission / reception side. However, instead of always waiting for the voice to be generated next to the neutral state, in this neutral state, the voice of the two-way call is attenuated and the sound quality is sacrificed instead of preventing howling.

  On the other hand, in the echo canceller 10 of the first embodiment, as described above, paying attention to the basic nature of human conversation, in which human conversation is often performed alternately, After the state where only the signal from the far end (ST) is in use and the near end speaker will respond (NT), the switch 108 operation is reflected. This is repeated alternately.

  That is, after the far-end single talk state (ST), the near-end speaker will respond to the words of the far-end talker. The terminal c1 of the switch 108 is connected to the a1 side, and processing is performed so as to select the signal e (n) on the echo canceller unit side without sound interruption even when the near-end speaker's voice is generated. At this time, even if the far-end speaker speaks again and goes into the far-end single talk state (ST) contrary to the assumption, the echo by the echo canceller (adder 101 and adaptive filter 102) is more or less. Since the cancellation is performed by the adder 101, no problem occurs.

  On the other hand, as expected, even if the near-end talk state (NT) occurs in the telephone device 1, the switch 108 selects the signal e (n) on the echo canceller side (transmits to the far end side). No problem. Furthermore, even if a double talk state (DT) that is a momentary conversation collision occurs, the double talk state (DT) is input from the bidirectional call detector 103 to the control unit 104, and the switch 108 connects the terminal c1. Control is performed so as to connect to the terminal a1 side.

  Therefore, the voice from the far end (far-end input signal) is transmitted among the two-way call signals, and the signal (signal e (n)) on the echo canceller side is used as the voice signal (transmission signal) of the near-end speaker. ) Is selected. Since the filter coefficient of the adaptive filter 102 is frozen (stops updating) in the double talk state (DT) as described above, the signal whose echo is reduced by the adder 101 is transmitted to the far end side. The Thereby, the echo canceller 10 can prevent the sound quality of the audio signal transmitted to the far end side from being deteriorated so much.

(A-3) Effect of First Embodiment (A-3-1) In the echo canceller 10, when the two-way call detector 103 detects the far-end single talk state (ST), the switch 108 And the signal att (n) from the voice switch unit (delay unit 105 and attenuator 107) is output to the far end side. On the other hand, when a state other than the far-end single talk state (ST) is detected by the two-way call detector 103, the switch 108 is controlled so that the echo canceller unit (adder 101, adaptive filter 102) The signal e (n) is transmitted to the far end side.

  In particular, in the echo canceller 10, the terminal c1 of the switch 108 is always connected to the terminal a1 when the far-end single talk state (ST) ends. As a result, in the alternating exchange of conversations between people, in the far-end single talk state (ST), the attenuation is surely inserted into the call path to prevent the feeling of echo, and the quasi-howling signal component that is growing is assumed to be Even if it exists, it can be attenuated at each far-end single talk state (ST). In other speech states, for example, near-end talk state (NT), the output of the echo canceller (adder 101) is selected, so that the first and last interruptions in speech are prevented and echoes are stably and continuously removed. be able to.

  The echo canceller 10 also selects the output of the cancellation adder 101 on the adaptive filter 102 side in the transmission state even in the double talk state (DT) that occurs momentarily during a call, so the first and last utterances This makes it possible to make a stable and natural call without interruptions or extreme volume levels.

(A-3-2) What is important in the echo canceller 10 is that the main body of echo cancellation in the single talk state (ST) is the voice switch unit (the delay unit 105 and the attenuator 107), and the echo y (n) It is to ensure that a certain amount of attenuation is applied. This is because howling that makes a call impossible is a phenomenon that occurs when the level distribution setting of a call path is poor. Specifically, as described in Non-Patent Document 1, this decrease occurs when the gain of the round-trip path of the acoustic signal is 1.0 or more. However, in reality, it is not so much that the cyclic gain of the communication path greatly exceeds 1.0 from the beginning. In such a case, howling does not occur suddenly, but it often ends up in howling as a result of many rounds of the signal. When the echo canceller has a variation in the echo path, the echo cancellation performance is degraded, and even in the single talk state, the background noise level is the limit of echo cancellation. It may not be able to hold down to 1.0 or less.

  However, in the echo canceller 10 of the first embodiment, there is always a time for forcibly incorporating the effect of the attenuator 107 on the communication path at least during detection of the far-end single talk state (ST). In other words, the howling cause signal that gradually grows, that is, the “quasi-howling signal”, can be pulled back to the state before the growth every time the far-end single talk state (ST), and the frequency of howling can be reduced.

(B) Second Embodiment Hereinafter, an echo canceller, an echo cancellation program, and a telephone device according to a second embodiment of the present invention will be described in detail with reference to the drawings.

(B-1) Configuration of the Second Embodiment FIG. 3 is a block diagram showing the overall configuration of the telephone device 1A of the second embodiment, which is the same as FIG. A reference numeral is attached.

  Hereinafter, the difference between the second embodiment and the first embodiment will be described.

  The telephone device 1A is different from the first embodiment in that the echo canceller 10 is replaced with an echo canceller 10A. The echo canceller 10A is different from the first embodiment in that an attenuation calculation unit 109, an echo loss estimation unit (hereinafter also referred to as “ELE”) 110, and an attenuator 111 are added. In addition, the determination result of the call state by the two-way call detector 103 is also supplied to the echo loss estimation unit 110.

  Note that details of the attenuation calculation unit 109, the echo loss estimation unit 110, and the attenuator 111 will be described in the description of operations described later.

(B-2) Operation | movement of 2nd Embodiment Next, operation | movement of 10 A of echo cancellers in 1 A of telephone apparatuses of 2nd Embodiment which has the above structures is demonstrated.

(B-2-1) Operation of Echo Loss Estimator 110 (ELE) The echo loss estimator 110 includes the far-end input signal x (n) and the component of the echo y (n) from the A / D converter 30. Near-end input signal is supplied. The echo loss estimation unit 110 calculates an echo loss, that is, an echo path attenuation amount (or an amplification amount) when the signal passes through an echo path from the reception output terminal Rout to the transmission input terminal Sin by a method described below.

The echo loss estimation unit 110 operates as follows only when the determination result of the far-end single talk state (ST) is supplied from the bidirectional call detector 103. The echo loss estimation unit 110 does nothing when other determination results are input from the two-way call detector 103, so the echo loss estimation unit 110 is in the same state as when there is no echo loss. The echo loss estimation unit 110 is far from the two-way call detector 103. When the determination result of the single talk state (ST) is input, the power smoothing value P_av (n) of the far-end input signal x (n) is calculated as in the following equation (6). When the determination result of the far-end single talk state (ST) is input from the bidirectional call detector 103, the calculation using the following equation (6) is the most accurate calculation of the signal increase / decrease in the echo path. This is desirable. In the following expression (6), δpav is a constant of 0 <δpav <1.0 that determines the degree of smoothness.

P_av (n) = δpav × x (n) ²
+ (1.0−δpav) × P_av (n) (6)
Further, the echo loss estimation unit 110 calculates the power smooth value EC_av (n) of the echo y (n) as in the following equation (7).

EC_av (n) = δpav × y (n) ²
+ (1.0−δpav) × EC_av (n) (7)
In the above equation (7), if δpav is large, it corresponds to a sudden change in the signal, but it is easily affected by noise, and conversely, if it is small, it corresponds to a rough change in the signal and is hardly affected by noise. Here, δpav = 0.01 is used, but it is not limited to this value.

  The echo loss estimation unit 110 obtains the maximum values P_MAX (k) and EC_MAX (k) of P_av (n) and EC_av (n) in a predetermined period T_ELE (for example, a period of 10 seconds). Note that the length of the period applied to T_ELE is not limited. In this case, k is an integer representing the order every 10 seconds.

  FIG. 4 is an explanatory diagram showing examples of P_MAX (k) and EC_MAX (k).

  As described above, the echo loss estimation unit 110 calculates the power transition of the far-end input signal x (n) output from the speaker 40, that is, P_av (n) by the above equation (6). Here, it is assumed that P_av (n) is represented as a graph as shown in FIG. Then, the echo loss estimation unit 110 obtains the maximum value P_MAX (k) of P_av (n) within the period of T_ELE.

  Originally, the echo y (n) is a signal that the far-end input signal x (n) output from the speaker 40 returns to the microphone 50 via the echo path and is observed again. Therefore, if there is no background noise in the echo path, the echo y (n) is a signal that is radiated from the speaker 40 and whose power decreases due to spatial propagation and is input to the microphone 50. That is, the value of EC_av (n) is usually a value obtained by slightly reducing P_av (n) as shown in FIG.

  That is, the echo path attenuation desired by the echo loss estimation unit 110 is simply the difference diff_EL (n) between P_av (n) and EC_av (n) obtained by the following equation (8), or (9 ) Expression P_av (n) and EC_av (n) ratio ratio_EL (n).

diff_EL (n) = P_av (n) −EC_av (n) (8)
ratio_EL (n) = P_av (n) / EC_av (n) (9)
However, in reality, since the echo path is a real environment where the speaker 40 and the microphone 50 of the telephone device 1A on the near-end speaker side are placed, there is almost no background noise. Therefore, as a background noise, as shown in FIG. 4B, consider a case where there is a background noise having a uniform power on the time axis (a component represented by black paint; hereinafter also referred to as “BGN”). Then, the relationship between EC_av (n), background noise BGN, and echo loss estimation value ELE_EST (k) shown in FIG. 4A is as shown in FIG. That is, as can be seen from FIG. 4C, since the power of the echo y (n) decreases correspondingly in the portion where the power of the far-end input signal x (n) is small, the echo y (n) It becomes smaller than the power of the background noise BGN and is buried in the background noise BGN.

  In such a case, even if the above formulas (8) and (9) are simply calculated, the EC_av (n) should actually be observed, but the power of the background noise BGN that has overcome the power is calculated. Therefore, the loss of the true echo path cannot be obtained.

  Therefore, as will be described later, the echo loss estimation unit 110 obtains an echo loss estimated value ELE_EST (k) and reflects it in the operations of the two attenuators 107 and 111 responsible for preventing howling.

  The echo loss estimation unit 110 calculates BGN ′ (k) as the minimum value of a predetermined period T_BG (for example, the same 10 seconds as T_ELE) of EC_av (n) shown in the above equation (7). k is an integer representing the order (frame) of period breaks as in the above equation (7). Here, the period for obtaining BGN '(k) is 10 seconds, but the present invention is not limited to this.

  Note that a predetermined initial value may be used as long as 10 seconds have not elapsed from the initial state (for example, after signal processing is started by the echo canceller 10). Here, description will be made assuming that −45 dBm0 (0 dBm0 may use the 0 dBm0 reference signal of the international standard ITU-TG.711) as the initial value described above, but is not limited thereto. Further, as BGN '(k), the minimum value within the period from the initial state to the present time may be used.

  FIG. 4D shows an example of the relationship between EC_av (n) and BGN ′ (k). FIG. 4 (e) shows an example in which P_av (n) is further superimposed on the graph of FIG. 4 (d).

  FIG. 4D shows a state in which many parts of EC_av (n) representing the echo component are buried in the background noise BGN component, and BGN '(k) matches the BGN level. That is, in FIG. 4D, EC_av (n) can represent the true echo level only in the portion exceeding BGN '(k).

  Among them, the echo loss that is most likely to be estimated is likely to be estimated in the vicinity of the peak value of the signal on the output side of the speaker 40 (far-end input signal x (n)), that is, P_MAX in FIG. It can be seen that this is a portion representing the difference between the point indicating (k) and EC_MAX (k). This is natural considering that the maximum portion of the envelope of the audio signal radiated from the speaker 40 (far-end input signal x (n)) becomes the maximum portion even as the echo y (n). is there.

  As described above, when the telephone device 1A is applied to, for example, a hands-free call device, in a sound environment where the hands-free call device is placed, whether or not howling occurs depends on whether echo loss is an amplification system. The In this case, in order to estimate and calculate the echo loss in the hands-free call device, the relationship between the power of the background noise BGN and the power of the echo is important in the sound environment where the hands-free call device is installed. .

  Basically, the echo loss includes the level P_av (n) (the maximum value is P_MAX (k)) related to the signal output from the speaker 40 and the level EC_av (n) related to the echo input to the microphone 50 via the echo path. ) (The maximum value is EC_MAX (k)) and can be calculated.

  However, in practice, it is necessary to consider the influence of the level of the background noise BGN ′ (k) that is confusing with the echo level (the same is true for power). The echo loss estimation unit 110 compares the magnitude relationships of P_MAX (k), EC_MAX (k), and BGN ′ (k), makes a determination related to the detected sound environment state, and determines the determination result, etc., as described later. This is given to the quantity calculation unit 109.

  FIGS. 5 and 6 are explanatory diagrams showing the determination operation related to the sound environment by the echo loss estimation unit 110.

  5 and 6, the vertical axis represents the power magnitude, and the horizontal axis represents the state type and state number (hereinafter also referred to as “EP_STAT”) of the sound environment state used in the echo loss estimation unit 110.

  In the echo loss estimation unit 110, the sound environment state is set to state 1 (EP_STAT: 1, state type: good state), state 2 (EP_STAT: 2, state type: noise state), state 3 (EP_STAT: 3, state type: unsuitable). State) and state 4 (EP_STAT: 4, state type: amplification state). Hereinafter, each of the states 1 to 4 will be described.

[In the case of state 1 (state type: good state, state number (EP_STAT): 1)]
In the echo loss estimation unit 110, the power maximum value P_MAX (k) related to the signal radiated from the speaker 40, the power maximum value EC_MAX (k) related to the echo, and the estimated background noise power BGN ′ (k) are sequentially decreased. In this case, the sound environment state is determined as state 1 (good state). The relationship among P_MAX (k), EC_MAX (k), and BGN ′ (k) in state 1 (good state) is illustrated in FIG. That is, state 1 (good state) can be said to be a state in which the echo is smaller than the speaker output signal and there is background noise below it.

  At this time, as shown in the following equation (10), the difference diff_POW (k) between P_MAX (k) and EC_MAX (k) accurately represents the amount of attenuation of the echo path. The diff_POW (k) correctly represents the echo loss AECHO (k) as shown in the following equation (10).

AECHO (k) = diff_POW (k)
= P_MAX (k) -EC_MAX (k) [dB] (10)
In the above equation (10), diff_POW (k) is obtained by using the difference between P_MAX (k) and EC_MAX (k). However, as shown in the following equation (10) ′, it is obtained by a linear ratio. May be. Even if the following expression (10) ′ is used, the same meaning as the above expression (10) is the same. Here, in order to simplify the description, AECHO (k) will be described using a differential expression such as the above equation (10), but it is natural that the present invention is not limited to this.

AECHO (k) = diff_POW (k)
= EC_MAX (k) / P_MAX (k) (10) '
That is, in the state 1 (good state), the sound signal radiated from the speaker 40 is attenuated by passing through the echo path. As described above, in such a system configuration of the echo path, since the echo does not go to divergence by the communication path circulation, the echo grows and no howling occurs.

  Then, when detecting the state 1 (good state), the echo loss estimation unit 110 gives EP_STAT = 1 and diff_POW (k) representing the state 1 (good state) to the attenuation amount calculation unit 109.

[In the case of state 2 (state type: noise state, state number (EP_STAT): 2)]
When the estimated background noise power BGN ′ (k) has a strength exceeding the power maximum value EC_MAX (k) related to the echo, the echo loss estimation unit 110 determines the sound environment state as state 2 (noise state). The relationship among P_MAX (k), EC_MAX (k), and BGN ′ (k) in state 2 (noise state) is illustrated in FIG. 5B. In state 2 (noise state), EC_MAX (k) is buried in BGN ′ (k), and real echo loss cannot be observed due to background noise.

  On the other hand, however, in the case of state 2 (noise state), it reflects the fact that EC_MAX (k) is considerably smaller than P_MAX (k), and it is impossible to calculate true AECHO (k) by chance. This is a safe state where no howling occurs.

  In the case of state 2 (noise state), the meaning of BGN ′ (k) is the result of tracking the minimum value of EC_av (n), so EC_MAX (k) and BGN ′ (k) look the same. Actually, EC_MAX (k) = BGN ′ (k). Therefore, in the case of state 2 (noise state), calculating the above equation (10) is actually the same as obtaining P_MAX (k) −BGN ′ (k), but the echo loss value is attenuated. A direction value is calculated.

  When detecting the state 2 (noise state), the echo loss estimation unit 110 gives EP_STAT = 2 representing the state 2 (noise state) and diff_POW (k) to the attenuation amount calculation unit 109.

[In the case of state 3 (state type: inappropriate state, state number (EP_STAT): 3)]
When the estimated background noise power BGN ′ (k) further increases from state 2 (noise state) and exceeds P_MAX (k), the echo loss estimation unit 110 determines the sound environment state to be state 3 (unsuitable state). The relationship among P_MAX (k), EC_MAX (k), and BGN ′ (k) in the state 3 (unsuitable state) is illustrated in FIG.

  In state 3 (unsuitable state), it means that BGN ′ (k) input to the microphone 50 is larger than P_MAX (k). State 3 (inappropriate state), in other words, as a state where the near-end speaker is placed, should be a state where the other party's voice radiated from the speaker 40 in the surrounding strong background noise can be heard hard. is there. Accordingly, it can be said that the use of hands-free calling is originally unsuitable as a calling means in the telephone device 1A. Naturally, in such a state, it is no longer possible to estimate AECHO (k) correctly.

  The echo loss estimation unit 110 performs calculation to obtain AECHO (k) using the above equation (10) even in the state 3 (unsuitable state), but the relationship between the echo and the background noise is the state 2 ( Noise state).

  Then, when detecting the state 3 (unsuitable state), the echo loss estimation unit 110 gives EP_STAT = 3 and diff_POW (k) representing the state 3 (unsuitable state) to the attenuation amount calculation unit 109. However, in the case of state 3 (unsuitable state), diff_POW (k) is not calculated as an amount of attenuation but rather as a value representing the amount of amplification. That is, in the state 3 (unsuitable state), it can be said that there is a high risk of howling.

[In the case of state 4 (state type: amplification state, state number (EP_STAT): 4)]
The echo loss estimation unit 110 has a power maximum value EC_MAX (k) related to the echo larger than the power maximum value P_MAX (k) related to the signal radiated from the speaker 40, that is, the power of the signal radiated from the speaker 40 is higher. When the sound path is amplified by the echo path, the sound environment state is determined as state 4 (amplified state).

  Usually, the sound power does not increase naturally as a result of air propagation. In many cases, an artificial amplification function not intended by the designer is mixed in somewhere in the echo path from the microphone 50 to the echo loss estimation unit 110. If this state is left unattended, in the telephone device 1A, the power continues to increase while the audio signal circulates, and howling is surely generated. That is, in the state 4, the risk of howling occurring in the telephone device 1A is extremely high.

  In FIG. 6A, the case where BGN ′ (k) is smaller than P_MAX (k) is illustrated as “state 4”, and in FIG. 6B, BGN ′ (k) is derived from P_MAX (k). Is also shown as “state 4 ′”. However, in the state where EC_MAX (k) is larger than P_MAX (k), the distinction is not so important. Therefore, hereinafter, “state 4” and “state 4 ′” are collectively referred to as “state 4”. ".

  Then, when detecting the state 4 (amplification state), the echo loss estimation unit 110 gives EP_STAT = 4 representing the state 4 (amplification state) and diff_POW (k) to the attenuation amount calculation unit 109.

(B-2-2) Operation of attenuation amount calculation unit 109 The attenuation amount calculation unit 109 sets the attenuation amount to be set in the attenuators 107 and 111 according to the state number EP_STAT of the sound environment state given from the echo loss estimation unit 110. To decide.

  Hereinafter, the operation of the attenuation amount calculation unit 109 will be described for each given EP_STAT value.

[In the case of state 1 (good state) or state 2 (noise state)]
First, the operation of the attenuation calculation unit 109 when the determination result given from the echo loss estimation unit 110 is EP_STAT = 1 (state 1, good state) or EP_STAT = 2 (state 2, noise state) will be described. To do. As described above, in the case of the state 1 or the state 2, the risk of howling occurring in the telephone device 1A is small.

  In the case of the state 1 or the state 2, the attenuation amount calculation unit 109 performs the attenuation applied to the attenuator 107 in the first embodiment in order to reduce the feeling of echo (the degree to which the near-end speaker perceives discomfort due to the echo). The amount of attenuation similar to the amount is notified. The attenuator 107 gives the attenuation amount notified from the attenuation amount calculation unit 109 to the signal yd (n) and outputs att (n).

  Further, in the case of the state 1 or the state 2, the attenuation amount calculation unit 109 may not notify the attenuator 111 of anything, and may indicate a value indicating that no attenuation is performed (for example, gain = 0). You may make it notify. At this time, the attenuator 111 transmits the near-end speaker signal including the echo y (n) (does not attenuate).

[In case of state 3 (unsuitable state)]
Next, the operation of the attenuation amount calculation unit 109 when the determination result given from the echo loss estimation unit 110 is EP_STAT = 3 (state 3, unsuitable state) will be described.

  As described above, when the telephone device 1A is applied to a hands-free calling device, the state 3 (unsuitable state) is a state in which use is inappropriate. That is, in the case of the state 3 (unsuitable state), there is a high possibility that the voice communication between the far-end speaker and the near-end speaker is difficult due to background noise on the near-end side. Furthermore, as described above, diff_POW (k) calculated by the echo loss estimation unit 110 does not necessarily represent AECHO (k).

  The state 3 (unsuitable state) can be considered as a special case where EC_MAX (k) = BGN '(k) in the state 4' described above. That is, the operation of the attenuation amount calculation unit 109 in the state 3 (unsuitable state) is in accordance with the operation in the state 4 (amplified state). The operation of the attenuation amount calculation unit 109 in the state 4 (amplification state) will be described later.

[In case of state 4 (amplification state)]
Next, the operation of the attenuation amount calculation unit 109 when the determination result given from the echo loss estimation unit 110 is EP_STAT = 4 (state 4, amplification state) will be described.

  As described above, when the telephone device 1A is applied to a hands-free calling device, the state 4 (amplified state) is a state in which the risk of occurrence of howling is extremely high.

  In the state 4 (amplification state), the attenuation calculation unit 109 notifies the attenuator 107 of the same attenuation as in the first embodiment. In the state 4 (amplification state), the attenuation amount calculation unit 109 notifies the attenuator 111 of an attenuation amount that cancels the amplification (attenuation) notified to the attenuator 107. As described above, in the case of state 4 (amplified state), the value of diff_POW (k) is not actually the amount of attenuation, but is a value representing the amount of amplification because of the order of difference calculation. Therefore, the attenuation amount calculation unit 109 inverts the sign of diff_POW (k), converts it to gain = −diff_POW (k), and notifies the attenuator 111 of the result. Furthermore, the attenuation amount calculation unit 109 may notify the attenuator 111 of an attenuation amount obtained by adding an attenuation amount of δatt20 (for example, 6 dB) as an attenuation margin of the attenuator 111. The margin attenuation δatt20 is not limited to 6 dB.

  In the state 4 (amplification state), when the attenuation calculation unit 109 and the echo loss estimation unit 110 are operated as described above, the power of BGN ′ (k) is at least P_MAX (k) −δatt20 due to the attenuation of the attenuator 111. The echo path from the reception output terminal Rout to the adder 101 is always an attenuation path.

  Further, by adding the above-mentioned margin attenuation amount δatt20 (6 dB) to the attenuation amount notified from the attenuation amount calculation unit 109 to the attenuator 111, the level of the signal in the transmission direction (near-end input signal) is further increased to P_MAX ( k) It can be limited to the following. This can prevent the bidirectional call detector 103 from erroneously determining the echo y (n) and the transmitted speaker signal until the adaptive filter 102 converges.

  That is, in the echo canceller 10A, occurrence of howling can be prevented unless there is an extreme signal amplification feedback on the far end side.

  Further, in the echo canceller 10A, as in the first embodiment, the echo y (n) is generated when the switch 108 selects the path on the voice switch unit side (the delay unit 105 and the attenuator 107). Attenuator 107 attenuates. Therefore, even when the state of the echo path cannot be known in advance, the occurrence of howling can be prevented more reliably.

(B-3) Effects of Second Embodiment According to the second embodiment, the following effects can be achieved.

  In the echo canceller 10A, the echo loss estimation unit 110 receives the maximum power value P_MAX (k) of the received signal output via the speaker 40, the maximum power value EC_MAX (k) of the echo, and the background noise on the microphone 50 side. An estimated value BGN ′ (k) is calculated.

  The echo loss estimation unit 110 attenuates the output (near-end input signal) of the microphone 50 by the attenuator 111 in accordance with the surrounding environment such as signal increase / decrease and background noise in the echo path. As a result, the echo canceller 10A automatically attenuates the amplification of the echo without the need for special expert designer settings even when there is an amplification factor in the echo path that cannot be known in advance. Howling can be prevented by canceling out by the device 111.

(C) Third Embodiment Hereinafter, an echo canceller, an echo cancellation program, and a telephone device according to a third embodiment of the present invention will be described in detail with reference to the drawings.

  In the echo cancellers 10 and 10A of the first and second embodiments, as described above, attention is paid to the fact that conversations between people basically consist of phrases being exchanged alternately. When there is a signal from the far end (when the far end input signal x (n) is in a voiced state), reception is performed so as not to degrade the far end side voice (far end input signal x (n)). Side signal is transmitted and output to the speaker 40, and on the transmitting side (signal transmission in the direction from the near end to the far end), the voice switch unit selects a path for inserting attenuation, thereby performing howling. It is surely prevented. In the echo cancellers 10 and 10A according to the first and second embodiments, the signal from the far end ends (the voiced state of the far end input signal x (n) ends), and at the same time, the transmission side path automatically. As described above, once a route without attenuation (a route that does not pass through the voice switch unit) is selected, even if a near-end speaker utters, the first and last loss of utterance does not occur.

  And in the echo cancellers 10 and 10A of the first and second embodiments, only in the bi-directional communication state (double talk state (DT)) or the transmission state (near end talk state (NT)) that occurs for a moment, The output (signal e (n)) of the echo canceller (adaptive filter and adder) is selected and transmitted to the far end side. Thereby, in the echo cancellers 10 and 10A of the first and second embodiments, priority is given to prevention of loss at the beginning and end of the utterance rather than ensuring the echo path attenuation.

  This is unique to the voice switch section and prevents the voice of the near-end speaker from being strongly attenuated, or the degree of attenuation drastically changing, making it difficult for the far-end speaker to hear and disrupting the conversation. is there.

  Therefore, in the third embodiment, in order to further improve the call quality, the frequency of use of the voice switch unit is slightly lowered, the opportunity to use the output of the echo canceller unit is increased, and the call quality is improved. ing.

(C-1) Configuration of the Third Embodiment FIG. 7 is a block diagram showing the overall configuration of the telephone device 1B of the third embodiment, which is the same as FIG. A reference numeral is attached.

  Hereinafter, the difference between the third embodiment and the second embodiment will be described.

  The telephone device 1B is different from the second embodiment in that an echo canceller 10A is replaced with an echo canceller 10B.

  In the echo canceller 10B, an echo attenuation amount calculation unit (hereinafter also referred to as “ACOM”) 112, a howling risk determination unit (hereinafter also referred to as “HWM determination unit”) 113, and an echo canceller effect determination unit (hereinafter, referred to as “AWM”). The second embodiment is different from the second embodiment in that an “EC determination unit” 114 is added. The echo canceller 10B is different from the second embodiment in that the control unit 104, the switch 108, and the echo loss estimation unit 110 are replaced with the control unit 104B, the switch 108B, and the echo loss estimation unit 110B.

(C-2) Operation of the Third Embodiment Next, the difference between the operation of the echo canceller 10B in the telephone device 1B of the third embodiment having the above configuration and the second embodiment will be described. .

  First, the difference between the operation of the echo loss estimation unit 110B and the second embodiment will be described.

  The operation of the echo loss estimation unit 110B is substantially the same as that of the second embodiment, except that the estimated background noise estimated value BGN '(k) is also given to the EC determination unit 114.

  Next, the operation of the EC determination unit 114 will be described.

  The EC determination unit 114 is supplied with the signal e (n) from the adder 101. Then, as will be described later, the EC determination unit 114 determines whether or not the performance of the echo canceller unit is performing up to an appropriate value corresponding to the background noise. An echo cancellation success signal conv indicating good performance is output to the determination unit 113, and nothing is output at other times.

  The EC determination unit 114 is supplied with a signal e (n) from the adder 101 in addition to the background noise estimation value BGN ′ (k) from the echo loss estimation unit 110B. In the EC determination unit 114, as in the echo loss estimation unit 110 in the second embodiment, the call state determination result from the two-way call detector 103 is the far-end single talk state (ST) for e (n). Only when the smoothing value Err_av (n) of e (n) is calculated between BGN ′ (k) and the synchronization (for example, between the above-described T_ELE and the synchronization) as in the following equation (11). Then, the EC determination unit 114 further calculates the maximum value ERR_MAX (k) of Err_av (n) during the predetermined period T_ELE. In the following formula (11), the meanings of n, k, and δpav are the same as those in the second embodiment.

Err_av (n) = δpav × e (n) ²
+ (1.0−δpav) × Err_av (n) (11)
Then, the EC determination unit 114 compares the calculated ERR_MAX (k) and BGN ′ (k) given from the echo loss estimation unit 110B as in the following equation (12). Here, in the following equation (12), it is assumed that δec_m = 2 dB is set, but the value of δec_m is not limited to this.

ERR_MAX (k) ≦ BGN ′ (k) + δec_m (12)
Next, the meaning of the above equation (12) will be described.

  It is known that the echo cancellation performance of an echo canceller using an adaptive filter and an adder is generally limited to the background noise power (level) input to the microphone simultaneously with the echo. In order for the normal echo canceller to be able to cancel the echo with high accuracy, it is necessary to repeat the update of the filter coefficient of the adaptive filter as described in the first embodiment. The relationship as in equation 12) does not hold. Therefore, when the state detection from the two-way call detector 103 is in the far-end single talk state (ST), the above equation (12) is established. This means that the coefficient update of the adaptive filter 102 proceeds, and the adder 101 Is reduced to near the background noise power BGN ′ (k), and it can be understood that δec_m is a margin threshold value for effect confirmation.

  Therefore, the EC determination unit 114 supplies the echo cancellation success signal conv to the HWM determination unit 113 when the above expression (12) is satisfied, and does not supply anything else.

  Next, the operation of the ACOM 112 will be described.

  The ACOM 112 is supplied with the far-end input signal x (n) from the reception input terminal Rin and the output signal e (n) of the adder 101. Then, similarly to the echo loss estimation unit 110, the ACOM 112 obtains P_av (n) by the above equation (6) and obtains the maximum value P_MAX (k) of the period T_ELE in the obtained P_av (n). Further, the ACOM 112 obtains the smoothed value Err_av (n) of the signal e (n) by the above equation (11) as in the EC determination unit 114 described above, and further calculates the maximum value of the predetermined period T_ELE in the obtained Err_av (n). The value ERR_MAX (k) is obtained.

  As described above, the ACOM 112 calculates the above formulas (6) and (11), but the configuration (processing) for calculating ERR_MAX (k) and P_MAX (k) is the EC determination unit. 114 and ELE23 may be shared, and only the calculation result may be acquired from the EC determination unit 114 and ELE23.

  Then, the ACOM 112 calculates the echo total attenuation amount ACOM_E (k) as shown in the following equation (13), and gives it to the HWM determination unit 113.

ACOM_E (k) = P_MAX (k)
-ERR_MAX (k) [dB] (13)
Next, the operation of the HWM determination unit 113 will be described.

  The HWM determination unit 113 compares ACOM_E (k) given from the ACOM 112 with the threshold value δacom. Here, δacom is described as 6 dB, but is not limited to this.

  Then, the HWM determination unit 113 satisfies ACOM_E (k) ≧ δacom, and while the signal conv from the EC determination unit 114 is supplied, the howling risk low signal NO_HW (indicating that the howling risk is low). Is supplied to the control unit 104B. On the other hand, HWM determination unit 113 outputs nothing to control unit 104B at other times.

  When NO_HW is given from the HWM determination unit 113, the control unit 104B controls the switch 108B to connect the terminal c1 to the terminal a1 side.

  When NO_HW is given from the HWM determination unit 113, the control unit 104B gives priority to the signal from the HWM determination unit 113 even if the two-way call detector 103 is notified of the far-end single talk state (ST). Then, the state of the switch 108B is left as it is (the state in which the terminal c1 is connected to the terminal a1 side). In other states, the control unit 104B performs the same control as the first embodiment on the switch 108B.

(C-3) Effects of Third Embodiment According to the third embodiment, the following effects can be achieved.

  As described above, in the third embodiment, the ACOM 112, the HWM determination unit 113, and the EC determination unit 114 are provided to control the switch 108B and use the output on the echo canceller unit side (the output of the adder 101). Increased opportunities. Details of the effects of the third embodiment will be described below.

  In the first and second embodiments, when the far-end single talk state (ST) in which only the far-end speaker speaks as a result of the state detection of the two-way call detector 103, a signal that is always transmitted to the far-end side by the switch unit When the far end speaker finishes speaking, the attenuation disappears (the output of the echo canceller is transmitted to the far end side). In the first and second embodiments, howling can be surely prevented, and even at the double talk state (DT), the first and last loss of speech can be prevented.

  On the other hand, however, from the viewpoint of the far-end speaker of the third embodiment, an unpleasant phenomenon may be caused in some cases as described below. That is, the far-end speaker may sound that the background noise on the near-end side increases or decreases significantly in accordance with the timing when he / she speaks. When this phenomenon is noticed during a conversation, some users at the far end may complain of great discomfort.

  In the third embodiment, in view of that, when the performance of the echo canceller unit side can be sufficiently exhibited and howling can be prevented, the echo canceller unit is responsible for echo cancellation, and otherwise As in the first and second embodiments, the strong attenuator of the switch unit is made to play the main role in echo cancellation.

  That is, the EC determination unit 114 mainly detects whether or not the far-end speaker has received echoes by detecting whether or not the echo cancellation by the echo canceller unit is successful up to the level of background noise. is doing. On the other hand, the ACOM 112 and the HWM determination unit 113 mainly detect whether or not a state that does not cause howling is secured as a result of cancellation of the echo canceller.

  For example, when the telephone device 1B is started up, the filter coefficient does not grow (converge) in the adaptive filter 102 of the echo canceller unit. In this case, the echo canceller cannot exert an effect on howling. However, if the coefficient of the adaptive filter 102 grows (converges), the echo can be canceled and the howling can be prevented while the audibility is more natural than the voice switch. become. In the third embodiment, when such a state is automatically detected and the use of the echo canceler unit is preferable to the voice switch unit, the output of the echo canceler unit is automatically selected, as long as the internal state of the apparatus permits. Natural speech without a sudden change in the transmission signal level can be provided as a call signal to the far-end speaker.

(D) Fourth Embodiment Hereinafter, an echo canceller, an echo cancellation program, and a telephone device according to a fourth embodiment of the present invention will be described in detail with reference to the drawings.

  In the echo cancellers 10, 10 </ b> A, and 10 </ b> B of the first to third embodiments, the echo generated mainly in the near-end side echo path is prevented from leaking into the signal transmitted to the far-end side. At the same time, the echo leaks into the signal transmitted to the far end side, and prevents howling which occurs while growing around the far end side and the near end side.

  As described so far, generally, an audio switch unit (a configuration in which echo is suppressed by a delay unit and an attenuator) and an echo canceller unit (a pseudo echo generated by an adaptive filter is used to cancel the echo by an adder unit) Configuration) is a technique used for the purpose of suppressing echo signals. However, in practice, the echo canceller is often preferred over the voice switch as an echo countermeasure. The echo canceller unit calculates the echo leaking into the microphone by calculation with an advanced algorithm, creates a pseudo echo, and reverses and cancels it with an adder. If such a determination process is successful, only the echo can be removed well and the call quality is good. On the other hand, the voice switch unit inserts and removes attenuation of an attenuator provided on the transmission or reception signal line to change the volume of the signal. That is, the voice switch unit eliminates the feeling of echo by adding a large attenuation to the attenuator on the transmission signal line in the far-end single talk state (ST) referred to in the present invention.

  However, since the voice switch section simply attenuates the signal line in principle, it is assumed that how far the far end single is accurately determined by a call state determination process (for example, a determination process such as the bidirectional call detector 103). Even if the talk state (ST) state is successfully detected, signals other than echoes (such as background noise on the near end side) are also attenuated together in the processing of the voice switch unit.

  Furthermore, also in the switching between the echo canceller unit and the voice switch unit described in the first to third embodiments, the output of the voice switch unit (output from the attenuator) depends on the state of background noise on the near end side. When switching to the output on the echo canceller side, a phenomenon that the magnitude of the background noise component suddenly changes may occur in the signal transmitted to the far end due to the added attenuation switching.

  This phenomenon is very unnatural for the far-end speaker, and every time the far-end speaker speaks, the other party's background noise sounds like suddenly intermittent. As a result, the far-end speaker is forced to talk with the near-end speaker while always having an unnatural feeling.

  This unnatural feeling can be caused by a hands-free telephone device that can be used temporarily to communicate requirements between the far-end speaker and the near-end speaker, or a conference phone device that can be used only during a conference. In many cases, it is acceptable.

  However, in a telework environment and a SOHO environment that have become widespread in recent years, the number of forms in which hands-free communication devices are always connected and used on both the near end side and the far end side is increasing, and the circumstances are different. That is, in such a new usage, the sudden change in the other party background noise that occurs simultaneously with the utterance as described above strongly appeals to the far-end speaker. This is because, in the case of hands-free communication devices that are always connected, the “background noise” that has been explained so far is not just extra noise in some cases, but “background sound” that is a clue to the other party's state change, environmental change, and atmosphere. Alternatively, it may be heard by the far-end speaker as “background environmental sound”.

  On the other hand, the echo canceller has a known property that the echo can be removed only to the near-end background noise level because of the structure in which the output feedback of the adder is used to update the coefficient of the adaptive filter. Therefore, when the background noise (sound) on the near end side is large, there arises a problem that the echo cannot be sufficiently removed during the conversation. In particular, humans can often hear human voices even if they are mixed with noise, and the echo at the end of the far-end speaker's utterance is often harsh to the far-end speaker.

  The conventional “Japanese Patent Laid-Open No. 3-254530” (reference 2) discloses a technique that uses band-limited Gaussian noise as a transmission substitute signal when the performance of the echo canceller cannot be exhibited. When the signal is replaced, the discomfort of the far-end speaker described above may rather increase.

  Therefore, in the fourth embodiment, when the background noise is large, the voice switch unit that can reliably suppress the echo is used, but the background noise is not unnatural and suddenly increases or decreases. Thus, in the fourth embodiment, even if the echo canceller unit and the voice switch unit are used together to switch the signal to be transmitted to the far end side, the background noise component in the signal to be transmitted to the far end side is rapidly increased or decreased. Can be eliminated. Further, in the fourth embodiment, as in the third embodiment, when the background noise is small enough to remove the echo by using only the echo canceller without using the voice switch, the echo canceller Echoes can be removed mainly to prevent howling.

(D-1) Configuration of the Fourth Embodiment FIG. 8 is a block diagram showing the overall configuration of the telephone device 1B according to the third embodiment, which is the same as FIG. A reference numeral is attached.

  Hereinafter, the difference between the third embodiment and the second embodiment will be described.

  The telephone device 1C is different from the third embodiment in that an echo canceller 10B is replaced with an echo canceller 10C.

  The echo canceller 10C is different from the third embodiment in that the attenuator 111 is omitted.

  Further, in the echo canceller 10C, the delay unit 106, the attenuator 107, and the attenuation amount calculation unit 109 are replaced with the delay unit 106C, the attenuator 107C, and the attenuation amount calculation unit 109C. Is different. Furthermore, the echo canceller 10C is different from the third embodiment in that the switch 108B is replaced with a time-varying weighted average unit 116. In the echo canceller 10 </ b> C, a signal mixing unit 115 (hereinafter also referred to as “N_MIX”) is provided between the attenuator 107 </ b> C and the time-varying weighted average unit 116. The details of the processing by the signal mixing unit 115 will be described in the operation description to be described later.

  FIG. 9 is a block diagram illustrating a functional configuration inside the signal mixing unit 115.

  The signal mixing unit 115 includes a switch control unit 301, a switch 302, a switch 303, a pseudo background noise holding unit 304, a mixing unit 305, a pseudo background noise power calculation unit 306, an actual background noise power calculation unit 307, and a fluctuation noise power calculation unit. 308 and a mixture ratio calculation unit 309. Note that the configuration (processing) of each part of the signal mixing unit 115 will be described in detail in the description of operations described later.

(D-2) Operation of the Fourth Embodiment Next, the difference between the operation of the echo canceller 10C in the telephone device 1C of the fourth embodiment having the above-described configuration and the third embodiment will be described. .

  First, the operation of the attenuation calculation unit 109C will be described.

  The attenuation amount calculation unit 109C calculates the attenuation amount once as in the third embodiment.

  Then, the attenuation amount calculation unit 109C compares a predetermined desired echo attenuation amount acom_4 (for example, 50 dB attenuation) with gain, and adds an attenuation amount that is insufficient for the attenuation to reach acom_4. The amount is notified to the signal mixing unit 115 and the attenuator 107C as a new attenuation amount gain4. Note that the value set in acom_4 is not limited.

  That is, the attenuation applied to the attenuator 107C is considerably larger than the attenuation set in the first to third embodiments. For example, if the desired echo attenuation amount acom_4 is set to 50 dB attenuation, and further attenuation of gain = 10 dB is calculated in the attenuation amount calculation unit 109C, gain4 = 40 dB (attenuation). In other words, in the fourth embodiment, when the increase / decrease in the acoustic echo path between the speaker 40 and the microphone 50 is combined with the attenuation by the attenuator 107C, the desired echo attenuation amount acom_4 is always calculated. The attenuation amount gain4 is notified to the attenuator 107C.

  Next, the operation of the signal mixing unit 115 will be described.

  The signal mixing unit 115 is notified of gain4 (for example, −40 dB) from the attenuation calculation unit 109C. The signal mixing unit 115 is supplied with the signal att (n) from the attenuator 107C. Further, the result of the call state determination from the two-way call detector 103 is supplied to the signal mixing unit 115. The signal mix (n) generated by the signal mixing unit 115 is supplied to the time-varying weighted average unit 116.

  Next, the operation of each unit of the signal mixing unit 115 will be described with reference to FIG.

  When the call state determination result notified from the two-way call detector 103 is a silent state (IDL), the switch control unit 301 closes the switch 302 and sends the output signal of the delay unit 106C to the pseudo background noise holding unit 304. Control to supply.

  The pseudo background noise holding unit 304 is configured by a ring buffer, holds data for a certain period (for example, 10 seconds) of the signal supplied from the delay unit 106C, and when the signal exceeds that, the pseudo background noise holding unit 304 Old data may be overwritten with new data. Here, the data holding period of the pseudo background noise holding unit 304 is set to 10 seconds, but this period is not limited.

  The switch control unit 301 opens the switch 302 and supplies a signal to the pseudo background noise holding unit 304 when the determination result of the call state notified from the two-way call detector 103 is other than the silence state (IDL). I will not let you.

  Therefore, the signal data held in the pseudo background noise holding unit 304 is a signal input to the microphone 50 in a silent state in which neither a far-end speaker nor a near-end speaker is speaking, that is, background noise on the near-end side. This is signal data indicating.

  The signal data held by the pseudo background noise holding unit 304 is supplied to the pseudo background noise power calculation unit 306.

  The pseudo background noise power calculation unit 306 calculates power p_pad_n for each predetermined period T_bn (for example, 1 second) from the supplied signal data, and supplies the fluctuation noise power calculation unit 308 to the fluctuation noise power calculation unit 308. give. The pseudo background noise power calculation unit 306 may calculate the power of the held signal when the data of the held signal is less than one second.

  On the other hand, when the far-end single talk state (ST) is input from the bidirectional call detector 103 as the call state determination result from the switch control unit 301, the switch 303 is closed and the mixing unit 305 receives the pseudo background. The signal data held by the noise holding unit 304 is supplied. Note that the switch 303 is opened under the control of the switch control unit 301 except in the far-end single talk state (ST).

  As described above, the mixing unit 305 is supplied with signal data from the pseudo background noise holding unit 304 via the switch 303.

  Next, the operation of the actual background noise power calculation unit 307 will be described.

  The actual background noise power calculation unit 307 receives the signal e_d (n) supplied from the adder 101 only when the far-end single talk state (ST) is input from the two-way call detector 103 as the determination result of the call state. ) To calculate the actual noise power p_real_n for a certain period T_bn, and supplies the calculation result to the fluctuation noise power calculation unit 308.

  The fluctuation noise power calculation unit 308 calculates a ratio dif_pad_n between the real noise power p_real_n and the pseudo background noise power p_pad_n. Although details will be described later, this is because the background noise level at the time when the data held by the pseudo background noise power calculation unit 306 is acquired and the background when the far-end single talk state (ST) actually occurs. This is because the difference from the output of the adder 101 of the echo canceller unit close to the noise level is reflected in the processing of the mixing unit 305.

  The mixing ratio calculation unit 309 calculates the mixing ratio between the signal supplied from the attenuator 107C and the signal supplied from the pseudo background noise holding unit 304 as follows, for example.

  For example, when gain 4 is supplied from the attenuation amount calculation unit 109C to the mixing ratio calculation unit 309, it is possible to calculate how much yd (n) has been attenuated in the attenuator 107C to create att (n). become. For example, in the attenuation calculation unit 109C, when gain4 = 40 dB is calculated and supplied to the mixture ratio calculation unit 309, when gain4 is considered as a linear ratio, this corresponds to a magnification of 0.01.

  The mixing ratio calculation unit 309 converts the supplied gain 4 into a linear ratio value δs, and then determines the mixing ratio δn with the output of the pseudo background noise holding unit 304 by the following equation (14).

δn = (1.0−δs) (14)
In this case, the gain 4 is obtained by reducing the signal that was originally 1 time to δs = 0.01 times, and if the pseudo background noise holding unit 304 output signal is mixed with the signal that is multiplied by 0.99, the signal power As a result, a signal having a small echo component ratio can be created.

  Further, the output ratio dif_pad_n of the fluctuation noise power calculation unit 308 is supplied to the mixture ratio calculation unit 309, and the background noise at the time when the data is held in the pseudo background noise holding unit 304 and the current echo canceller unit (adder) 101) Correct the discrepancy in power with the residual signal that is output.

  That is, as shown in the following equation (15), the amount of pseudo background noise to be mixed is set in accordance with the current near-end background noise amount in which the far-end single talk state (ST) occurs. Thus, in the far-end single talk state (ST), the amount of the pseudo background noise bn (n) mixed by the mixing unit 305 in view of the fact that the echo canceller removes the echo to the background noise level. Can be adjusted.

  In the case of dif_pad_n = 1.0 (the power of the adder output at the time when the far-end single talk state (ST) is generated is the same as when data is held), δn = (1.0−δs) ).

δn = (1.0−δs) × dif_pad_n (15)
Then, the mixing unit 305 generates mix (n) according to the following equation (16), and supplies the generated mix (n) to the time-varying weighted average unit 116.

mix (n) = δn × bn (n) + δs × att (n) (16)
Finally, the time-varying weighted average unit 116 will be described.

  The time-varying weighted average unit 116 has a function corresponding to the switch 108 (108B) in the first to third embodiments. Specifically, the time-varying weighted average unit 116 receives the signal e_d (n) supplied from the delay unit 106C through the terminal a2, or the signal mix (n) supplied from the signal mixing unit 115 through the terminal b2. ) Is selected based on the determination result of the call state supplied from the bidirectional call detector 103, and the selected signal is transmitted to the far end side (transmission output terminal Sout) via the terminal c2. The state is transitioned to Although not shown in FIG. 8, the time-variant weighted average unit 116 is supplied with the determination result of the call state from the two-way call detector 103.

  The time-varying weighted average unit 116 temporally combines the signal e_d (n) supplied from the delay unit 106C and the signal mix (n) supplied from the signal mixing unit 115 as shown in the following equation (17). A weight-distributed signal sout (n) is generated and output to the transmission output terminal Sout (transmitted to the far end side). In the following equation (17), δecho is a temporal weight distribution coefficient.

  When the time-varying weighted average unit 116 receives a control signal for selecting the terminal b2 from the delay unit 105, δecho = 0.0. In that case, the result of the following equation (17) is as shown in the following equation (18). In the first to third embodiments, the terminal c1 of the switch 108 (108B) is connected to the terminal b1. The same result.

  When the time-varying weighted average unit 116 receives a control signal for selecting the terminal a2 from the delay unit 105, δecbo = 1.0. In that case, the result of the following formula (17) is as shown in the following formula (19). In the first to third embodiments, the terminal c1 of the switch 108 (108B) is connected to the terminal a1. The same result.

sout (n) = δecho × e_d (n)
+ (1.0−δecho) × mix (n) (17)
sout (n) = mix (n) (18)
sout (n) = e_d (n) (19)
When the far-end single talk state (ST) ends, the time-varying weighted average unit 116 operates so that the weight of the terminal a2 is always increased. However, the time-varying weighted average unit 116 is required in advance to shift from δecho = 0.0 to δecho = 1.0. The operation may be performed so that the weight changes smoothly over a predetermined period, for example, 0.5 seconds.

  For example, when the background noise is large, as described above, the echo canceller can only remove the echo up to the background noise level. For this reason, when the output signal of the adder 101 is shifted to e_d (n), which is a signal abruptly subjected to strong signal attenuation, to e_d (n), the far-end talk that has remained unerased by the echo canceller unit. The echo of the last part of the person's utterance stands out. Therefore, by providing the time-varying weighted average unit 116 as described above and switching signals smoothly (gradually), the echo of the last part of the far-end speaker's utterance becomes conspicuous. The phenomenon can be prevented.

(D-3) Effects of the fourth embodiment According to the fourth embodiment, the following effects can be achieved.

(D-3-1) In the echo canceller 10C, the signal mixing unit 115 is provided in the voice switch unit. The signal mixing unit 115 holds the near-end input signal (noise not mixed with echo) in the silent state (IDL) as pseudo background noise, and from the attenuator 107C in the far-end single talk state (ST). A signal mix (n) obtained by mixing the pseudo background noise held in the output att (n) is output. In the signal mixing unit 115, the mixing ratio at the time of signal mixing is set to a value such that the power of mix (n) maintains the near-end input signal power before attenuation by the voice switch unit.

  As a result, in the fourth embodiment, the background noise level can suppress only the echo signal while simulating the current state of the near-end speaker ambient noise in terms of the audibility of the far-end speaker. That is, the echo canceller 10C can improve the quality of the signal transmitted to the far end side.

(D-3-2) In the echo canceller 10C, a time-varying weighted average unit 116 is provided, and an output of the voice switch unit (output of the signal mixing unit 115) and an output of the echo canceller unit (output of the adder 101) However, it is designed to switch smoothly in time. As a result, in the fourth embodiment, after the far-end single talk state (ST), even if the selection of the signal to be sent to the far-end side is always switched to the echo canceller side, the output signal of the voice switch unit is immediately Therefore, it is possible to prevent the echo of the last part of the utterance due to the utterance of the far-end speaker from returning to the far-end side.

  Further, in the echo canceller 10C, by providing the time-varying weighted average unit 116, it is possible to prevent a sudden change in the background noise on the near end side felt by the far end speaker due to the switching of the outputs of the voice switch unit and the echo canceller unit. It becomes like this. Accordingly, in the fourth embodiment, desired attenuation can be surely inserted into the communication path, so that it is possible to provide an echo canceller capable of a natural call without howling.

(E) Other Embodiments The present invention is not limited to the above-described embodiments, and may include modified embodiments as exemplified below.

(E-1) In the third embodiment, the attenuation calculation unit 109, the echo loss estimation unit 110B, and the attenuator 111 may be omitted. That is, a configuration (ACOM 112, HWM determination unit 113, EC determination unit 114, etc.) that is a feature of the third embodiment may be added to the first embodiment. However, in that case, it is necessary that the EC determination unit 114 can obtain the content acquired by the EC determination unit 114 from the echo loss estimation unit 110B in the third embodiment.

(E-2) In the fourth embodiment, the time-varying weighted average unit 116 may be replaced with the same switch as in the first to third embodiments. Conversely, the time-varying weighted average unit 116 of the fourth embodiment may be applied to the first to third embodiments.

  Furthermore, the signal mixing unit 115 of the fourth embodiment may be applied to the first to third embodiments. However, in that case, in the fourth embodiment, it is necessary that the signal mixing unit 115 can obtain the content acquired by the signal mixing unit 115 from the attenuation amount calculation unit 109C.

(E-3) In the second embodiment, the echo loss estimation unit 110 operates only when the far-end single talk state (ST) is input from the two-way call detector 103. However, the silence state (IDL) Since no echo is generated in the near-end talk state (NT), the background noise BGN ′ (k) around the microphone 50 may be calculated during this state.

(E-4) In the third embodiment, the echo attenuation amount is calculated from the received input (far-end input signal x (n)) and the output of the adder 101 to calculate the total attenuation amount of the echo path. However, the power cancellation amount is calculated by inputting the signals before and after the adder 101, and the entire ACOM 112 is calculated by the amount of attenuation combined with the attenuation amount of the echo path calculated by the echo loss estimation unit 110B. You may make it do.

(E-5) In the third embodiment, the HWM determination unit 113 outputs both the establishment of “ACOM_E (k) ≧ δacom” and the presence of the echo cancellation success signal conv, and outputs the small howling risk signal NO_HW. However, if there is no need to worry about the feeling of echo, the small howling risk signal NO_HW may be output only by the establishment of the above inequality.

(E-6) In the third embodiment, addition of attenuation according to the state of the echo path is applied to the attenuator 111, while addition of attenuation according to the call state is divided and applied to the attenuator 107. However, the attenuation amount may be added by one attenuator 107.

  Further, in the third embodiment, when the detection delay of the bidirectional call detector 103 is small enough to be ignored, it may be attenuated by one attenuator 111 in the same manner.

(E-7) In the fourth embodiment, a pseudo background noise power calculation unit 306, an actual background noise power calculation unit 307, and a fluctuation noise power calculation unit 308 are provided in the signal mixing unit 115 so that data at the time of data retention is provided. The divergence between the amount of noise and the amount of noise during the far-end single talk state (ST) is reduced. However, when the fluctuation of the background noise is not so significant, the pseudo background noise power calculation unit 306, the actual background noise power calculation unit 307, and the fluctuation noise power calculation unit 308 are omitted, and the pseudo background noise bn (n) and the attenuation are reduced. The output signal att (n) of the device 107C may be mixed.

  DESCRIPTION OF SYMBOLS 1 ... Telephone apparatus, 20 ... D / A converter, 30 ... A / D converter, 40 ... Speaker, 50 ... Microphone, 10 ... Echo canceller, 101 ... Adder, 102 ... Adaptive filter, 201 ... Pseudo echo creation part , 2011 ... holding register, 2012 ... product-sum operation unit, 2013 ... coefficient unit, 202 ... coefficient update amount calculation unit, 203 ... step gain calculation unit, 103 ... bidirectional call detector, 104 ... control unit, 105 ... delay unit 106 delay unit 107 attenuator 108 switch a1, b1, c1 terminal.

Claims (13)

In an echo canceller that removes echo components from the near-end input signal,
Echo canceling means for generating a signal obtained by canceling the echo component of the near-end input signal using the pseudo echo generated by the pseudo echo generating means;
Voice switch means for generating a signal obtained by attenuating the near-end input signal by an attenuator;
Switch means for selecting either the signal after cancellation generated by the echo cancellation means or the signal generated by the voice switch means and sending it to the far end side;
Call state determination means for determining a call state between a far end speaker and a near end speaker using the far end input signal and the near end input signal;
An echo canceller comprising: switch control means for controlling the switch means in accordance with a determination result of the call state determination means.   The call state determination means detects a voiced state or a silent state for each of the far-end input signal and the near-end input signal, and based on the result, the far-end speaker and the near-end speaker The echo canceller according to claim 1, wherein a call state between the two is determined.   When the call state determination unit determines that the far-end single talk state in which only the far-end speaker is speaking, the switch control unit sends a signal generated by the voice switch unit to the switch unit. The echo canceller according to claim 2, wherein the echo canceller is controlled so as to select.   When it is determined that the call state determining unit has transitioned to a state other than the far-end single talk state after the determination, the switch control unit generates the echo canceling unit for the switch unit. The echo canceller according to claim 3, wherein control is performed so as to select a signal. Echo path characteristic calculating means for calculating an attenuation amount or an amplification amount of an echo path related to the echo component;
Power comparison means for comparing the power of the entire near-end input signal, the power of the echo component, and the power of the background noise component on the near-end side included in the near-end input signal;
A signal attenuating means for generating a signal giving a set amount of attenuation to the near-end input signal, and supplying the generated signal to the echo canceling means and the voice switch means;
Attenuation amount calculating means for obtaining an attenuation amount to be set in the signal attenuating means using the calculation result of the echo path characteristic calculating means and the comparison result of the power comparing means,
The echo canceller according to claim 1, further comprising: In the near-end input signal and the far-end input signal, further comprising howling occurrence risk determination means for determining a risk that howling based on the echo component occurs,
The switch control means selects the signal generated by the voice switch means for the switch means when the howling occurrence risk judgment means determines that the risk of howling is greater than or equal to a predetermined value. The echo canceller according to any one of claims 1 to 5, wherein The above-mentioned howling occurrence risk determination means is:
The far-end input signal power and the cancellation generated by the echo cancellation means while the speech state determination means determines that only the far-end speaker is speaking a far-end single talk state. Using the power of the post-signal, an echo attenuation amount calculation unit for calculating the power of the echo component remaining in the post-cancellation signal generated by the echo cancellation means;
The echo canceller according to claim 6, further comprising: a howling occurrence risk level determination unit that determines a risk level of howling using the calculation result of the echo attenuation amount calculation unit. The above-mentioned howling occurrence risk determination means is:
The signal after cancellation generated by the echo cancellation means is compared with the power of the background noise component on the near end side, and based on the comparison result, the cancellation of the echo component by the echo cancellation means is successful. It further has an echo cancellation effect determination unit that determines whether or not
The echo canceller according to claim 7, wherein the howling occurrence risk determination unit determines a risk level of howling using the determination result of the echo cancellation effect determination unit. The switch means includes
A time-variant weighted average unit that generates a signal weighted and distributed to the far end side of the signal generated by the voice switch unit and the signal after cancellation generated by the echo cancellation unit;
For the time-varying weighted average unit, the weighting for the post-cancellation signal generated by the echo cancellation unit and the weighting for the signal generated by the voice switch unit are updated according to the control of the switch control unit. The echo canceller according to claim 1, further comprising: a weight update unit. The voice switch part
Pseudo background noise holding means for holding the near end input signal as a pseudo background noise signal when both the far end input signal and the near end input signal are determined to be silent in the call state determination means;
While the speech state determination means determines that only the far-end input signal is in a sound state, the signal output from the attenuator and the pseudo background noise held by the pseudo background noise holding means Signal mixing means for generating a mixed signal as a signal output by the voice switch unit;
The mixing ratio to be set in the signal mixing means is determined, and the mixing ratio is obtained by using the power of the pseudo background noise held by the pseudo background noise holding means and the power of the near-end input signal. The echo canceller according to claim 1, further comprising: a mixture ratio determining unit that determines the ratio.   11. The echo canceller according to claim 10, wherein the mixing ratio determining means determines a mixing ratio such that the power of the signal output from the signal mixing means does not deviate from the power of the near-end input signal. A computer installed in an echo canceller that removes the echo component from the near-end input signal.
Echo canceling means for generating a signal obtained by canceling the echo component of the near-end input signal using the pseudo echo generated by the pseudo echo generating means;
Voice switch means for generating a signal obtained by attenuating the near-end input signal by an attenuator;
Switch means for selecting either the signal after cancellation generated by the echo cancellation means or the signal generated by the voice switch means and sending it to the far end side;
Call state determination means for determining a call state between a far end speaker and a near end speaker using the far end input signal and the near end input signal;
An echo canceling program that functions as a switch control unit that controls the switch unit according to a determination result of the call state determination unit.   A telephone apparatus comprising the echo canceller according to any one of claims 1 to 11.

Images