Patent Translate Powered by EPO and Google Notice This translation is machine-generated. It cannot be guaranteed that it is intelligible, accurate, complete, reliable or fit for specific purposes. Critical decisions, such as commercially relevant or financial decisions, should not be based on machine-translation output. DESCRIPTION JP2006050342 An object of the present invention is to improve the estimation accuracy of feedback gain and further increase the opportunity to realize two-way simultaneous call by performing total loss control more optimally than before so that closed loop loop gain does not exceed 1 time. . A maximum feedback gain estimation unit (55) selects the largest of the feedback gains obtained for each frequency band as an estimated value of the feedback gain. Therefore, the estimation accuracy of the feedback gain is improved compared to the conventional example in which the feedback gain is estimated from the time average power of the reference signal, and as a result, the total loss control such that the closed loop gain does not exceed 1 It is possible to optimize the above, and it is possible to further increase the opportunity to realize two-way simultaneous call. In addition, even in an environment where it is usually difficult to make a call because of the generation of ambient noise such as outdoor road noise and indoor television noise, a more comfortable call can be realized. [Selected figure] Figure 1 Loudspeaker talker [0001] The present invention relates to a loudspeaker communication apparatus such as an intercom used in a house or an office. [0002] In this type of loudspeaker communication apparatus, an unpleasant echo (sound echo or echo) is generated by the feedback path on the acoustic side formed by the acoustic coupling of the microphone and the speaker, or the feedback path on the line side formed with the other communication terminal. If a closed loop in which the loop gain in an arbitrary frequency 08-05-2019 1 component exceeds 1 times is formed in a speech system due to the feedback path or the like, feedback may occur at the frequency. In some cases, an echo canceller and a voice switch are provided to prevent the occurrence of unpleasant echo and howling as described above. [0003] The voice switch constantly estimates the call state (transmission state, reception state), and inserts a loss into the signal path of the transmission side and the reception side with appropriate allocation based on the estimation result. Also, the echo canceller adaptively identifies the impulse response of the feedback path and estimates the pseudo echo component of the feedback path from the input signal to the feedback path, and the pseudo echo component estimated by the adaptive filter as a feedback path And a subtractor for subtracting from the output signal from. Here, since the adaptive filter of the echo canceller usually requires a learning time of several seconds to identify the impulse response of the feedback path, the echo suppression effect of the echo canceller can not be sufficiently expected for several seconds immediately after the start of the call. The closed loop is formed in the speech system, which may cause unpleasant echo or howling. [0004] Therefore, the present applicant has already proposed a loudspeaker communication apparatus capable of suppressing unpleasant echo and howling immediately after the start of a call (see Patent Document 1). [0005] In this conventional example, in a state where the echo canceller immediately after the start of a call does not converge, the voice switch operates in a fixed mode in which the total amount of loss (total loss amount) inserted in the signal path is fixed to a sufficiently large initial value. This suppresses unpleasant echoes and howlings, and in a state where the echo canceller has converged sufficiently, the voice switch operates in the update mode in which the total loss amount is updated as needed, thereby realizing a two-way simultaneous call. 08-05-2019 2 [0006] The voice switch of the prior art disclosed in Patent Document 1 includes a transmitter attenuator for inserting a loss in a signal path on the transmitter side, a receiver attenuator for inserting a loss in a signal path on the receiver side, and a transmitter. And an insertion loss amount control unit for controlling the amount of loss inserted from each of the talker and the receiver. Also, the insertion loss amount control unit is an acoustic feedback gain of a path (hereinafter, referred to as “acoustic side feedback path”) that returns from the output point of the receiving side attenuator to the input point of the transmitting side attenuator via the acoustic echo path. To estimate the line-side feedback gain of the path (hereinafter referred to as the “line-side feedback path”) to be fed back from the output point of the transmitting side attenuator to the input point of the receiving side attenuator via the line echo path. Calculate the sum of losses to be inserted in the closed loop (sum of insertion loss of transmitting attenuator and insertion loss of receiving attenuator) on the basis of estimated values of feedback gains on the acoustic side and the line side A total loss amount calculation unit, a transmission signal and a reception signal are monitored to estimate a call state, and each of a transmission side attenuator and a reception side attenuator is calculated according to the estimation result and the calculation value of the total loss amount calculation unit. And an insertion loss amount distribution processing unit that determines the distribution of the insertion loss amount. [0007] The total loss amount calculation unit estimates the time average power of the input signal of the transmission side attenuator in a short time using a rectifying and smoothing device, a low pass filter, and the like, and similarly uses the rectifying and smoothing device, a low pass filter, and the like. Estimate the time-averaged power of the output signal of the receiver attenuator in a short time, and estimate the minimum value of the estimated time-averaged power of the output signal of the receiver attenuator at the maximum delay time assumed in the acoustic feedback path. A value obtained by dividing the estimated value of the time-averaged power of the input signal of the transmitting side attenuator by this minimum value is used as the estimated value of the acoustic side feedback gain, and reception is performed using a rectifier / smoothing device, a low pass filter, etc. The time average power of the input signal of the side attenuator in a short time is estimated, and the time average power of the output signal of the transmitting side attenuator in a short time is also estimated using a rectifying / smoothing unit, low pass filter, etc. Feedback path Minimum value of estimated value of time average power of output signal of 08-05-2019 3 transmitting side attenuator at maximum delay time assumed, and the estimated value of time average power of input signal of receiving side attenuator is divided by this minimum value Let the value be an estimate of the line-side feedback gain. Then, the total loss amount calculation unit calculates the total loss amount necessary for obtaining a desired gain margin from each estimated value of the acoustic side feedback gain and the line side feedback gain, and outputs the value to the insertion loss amount distribution processing unit Do. [0008] The insertion loss amount distribution processing unit monitors the input / output signals of the transmitting side attenuator and the input / output signals of the receiving side attenuator, and calls from the information such as the magnitude relationship of the power levels of these signals and the presence or absence of the audio signal. Insertion loss of each attenuator so as to determine the state (reception state, transmission state, etc.) and distribute the total loss amount to the transmission side attenuator and the reception side attenuator at a rate according to the determined communication state Adjust the amount. Japanese Patent Application Laid-Open No. 2002-359580 [0009] By the way, since the feedback path generally has frequency dependency, howling occurs at a specific frequency. Therefore, it is necessary to investigate the frequency dependency and use the maximum value (peak value) for estimation of the gain margin and the like. However, in the feedback gain estimation process of the above-described conventional example, since the estimated value is obtained by the ratio of time average power without considering the frequency dependency, the estimation accuracy is degraded as the frequency dependency of the gain of the feedback path is higher. It was Therefore, in order to prevent the howling that occurs when a closed loop in which the loop gain of the feedback path is more than 1 is formed in the communication system, the gain margin must be designed on the safe side, and therefore the loop gain In some cases, the two-way simultaneous call can not be realized even if the value of 1 does not exceed one time, resulting in a one-way call. [0010] 08-05-2019 4 The present invention has been made in view of the above circumstances, and its object is to improve the estimation accuracy of the feedback gain and to optimize the total loss control such that the loop gain of the closed loop does not exceed 1 time. It is an object of the present invention to provide a speech communication apparatus which can further increase the opportunity to realize two-way simultaneous calls by carrying out. [0011] In order to achieve the above object, the invention according to claim 1 provides a microphone and a speaker, a receiver-side signal path for transmitting a reception signal sent from a call terminal on the other side to the speaker, and a transmission collected by the microphone. A voice switch that switches the call state to reception and transmission by inserting a loss in the transmission side signal path that transmits a signal and sends it to the other party's call terminal, and suppresses an acoustic echo caused by acoustic coupling between a microphone and a speaker The voice switch includes a transmission loss insertion unit for inserting a loss in the transmission signal path, a reception loss insertion unit for inserting a loss in the reception signal path, and a transmission switch. And an insertion loss amount control unit for controlling the amount of loss inserted from each loss insertion unit on the receiving side and the receiving side, and the insertion loss amount control unit controls an acoustic echo path from an output point of the receiving loss insertion unit. Estimate the acoustic feedback gain of the path to the input point of the transmitting side loss insertion unit via the input side of the receiving side insertion loss section from the output point of the transmitting side loss insertion unit via the line echo path Feedback gain estimation unit for estimating the channel-side feedback gain of the path to be fed back, and a total loss amount calculation unit for calculating the sum of loss amounts to be inserted in the closed loop based on estimated values of the feedback gains on the acoustic side and the line side Monitoring the transmission signal and the reception signal to estimate the call state, and according to the estimation result and the calculated value of the total loss calculation unit, the insertion loss amounts of the transmission loss insertion unit and the reception loss insertion unit And an insertion loss amount distribution processing unit that determines the distribution of the total loss amount calculation unit calculates the sum of loss amounts to be inserted into the closed loop based on the estimated value of each feedback gain and adaptively updates the update mode; And fixed the total loss amount to a predetermined initial value Operates in the fixed mode in the period from the start of the call with the other party's call terminal to when the echo canceler fully converges, and the period after the echo canceler sufficiently converges In the speech communication system that operates in the update mode, the feedback gain estimation unit includes a transmission time difference correction unit that corrects the difference in signal transmission time inherent to each of the acoustic and line feedback paths, and a transmission time difference correction unit. The feedback gain for each frequency band is calculated from the signal level calculated by the frequency band-specific signal level calculation unit for obtaining the signal level for each frequency band by subjecting the corrected signal to Fourier transform processing, and And a maximum feedback gain 08-05-2019 5 selection unit that estimates and selects the largest feedback gain among the feedback gains for each frequency band. [0012] The invention of claim 2 comprises, in the invention of claim 1, a reference signal storage unit for storing data obtained by sampling reference signals of a transmission signal and a reception signal in a feedback gain estimation unit, and a signal level calculation unit for each frequency band. Is characterized in that the data stored in the reference signal storage unit is read and the discrete Fourier transform process is performed. [0013] The invention of claim 3 relates to the invention according to claim 1 or 2, wherein an envelope unit for obtaining a time series average for each frequency band with respect to the signal level for each frequency band calculated by the signal level calculation unit for each frequency band is feedback gain. The estimation unit is provided, and the maximum feedback gain selection unit is characterized by selecting the largest time-series average as the feedback gain among the timeseries averages of the signal levels obtained by the envelope unit. [0014] The invention according to claim 4 is the invention according to claim 1 or 2 or 3, wherein smoothing is performed on the maximum feedback gain selected by the maximum feedback gain selection unit in time series and the smoothed value is variable according to parameter setting. A feedback filter unit is provided in the feedback gain estimation unit. [0015] According to the invention of claim 5, in the invention of any one of claims 1 to 4, the feedback gain estimating unit alternately estimates the acoustic side feedback gain and the line side feedback gain at constant time intervals, and estimates them at different times. The present invention is characterized in that the feedback gain estimation is performed by combining the acoustic side feedback gain and the line side feedback gain. [0016] The invention according to claim 6 is the invention according to any one of claims 1 to 5, wherein the frequency-band-specific signal level calculation unit performs a process for obtaining a square root when calculating the signal level, It is characterized in that it has a list in which square roots are associated with each other in a one-to-one manner, and the square root is 08-05-2019 6 calculated by referring to the list. [0017] According to a seventh aspect of the present invention, in the second aspect of the present invention, the frequency band-specific signal level calculation portion may be configured so that all data read from the reference signal storage portion for performing discrete Fourier transform processing is less than a predetermined threshold. For example, the present invention is characterized in that the discrete Fourier transform processing is stopped and the signal level obtained in the previous discrete Fourier transform processing is substituted. [0018] The invention according to claim 8 relates to the invention according to any one of claims 1 to 7, wherein the frequency band-specific signal level calculation unit is configured to apply a double talk state in which an audio component is applied in the middle of a path where the echo canceller estimates a feedback gain. When detected, the discrete Fourier transform processing is stopped and the signal level obtained in the previous discrete Fourier transform processing is substituted. [0019] The invention according to claim 9 is the invention according to any one of claims 1 to 8, wherein the maximum feedback gain selection unit is the maximum feedback with respect to the closed loop feedback gain represented by the sum of the acoustic side feedback gain and the line side feedback gain. In selecting a signal level in a frequency band having a gain, a denominator of each frequency band in a feedback gain calculation formula expressed by a fractional formula is divided. [0020] According to the invention of claim 1, since the largest feedback gain among the feedback gains obtained for each frequency band is selected as the estimated value of the feedback gain, the feedback gain is conventionally estimated from the time average power of the reference signal. Since the estimation accuracy of the feedback gain is improved compared to the example, as a result, total loss control can be optimally performed such that the loop gain of the closed loop does not exceed 1 time. It is possible to further increase the opportunity to realize the call, and also to make the call more comfortable even in an environment where it is usually difficult to make a call, due to the generation of ambient noise such as outdoor road noise and indoor television noise. There is an effect that a speech communication system can be provided. 08-05-2019 7 [0021] According to the invention of claim 2, in addition to the effect of the invention of claim 1, the execution waiting time of the discrete Fourier transform processing until the data to be the target of the discrete Fourier transform processing is accumulated in the reference signal storage unit Is eliminated, and continuous discrete Fourier transform processing synchronized with the sampling period of the reference signal can be performed, so that the estimation accuracy of the feedback gain can be further improved. [0022] According to the invention of claim 3, in addition to the effect of the invention of claim 1 or 2, since the time-series average of the signal level for each frequency band is obtained by the envelope unit, the amount of operation in discrete Fourier transform processing is suppressed. Therefore, even when the number of data to be simultaneously subjected to the discrete Fourier transform processing is small, it is possible to prevent the degradation of the estimation accuracy of the feedback gain due to the application of noise. [0023] According to the invention of claim 4, in addition to the effect of the invention of claim 1 or 2 or 3, the selected maximum feedback gain is smoothed in time series and the smoothed value is variable by parameter setting. This has the effect of facilitating design that prevents the occurrence of howling due to an erroneous estimation. [0024] According to the invention of claim 5, in addition to the effect of the invention according to any one of claims 1 to 4, the amount of calculation is reduced as compared with the case of simultaneously estimating the acoustic feedback gain and the line feedback gain. There is an effect that implementation with a low specification computer becomes possible. [0025] According to the invention of claim 6, in addition to the effect of the invention according to any of claims 1 to 5, the amount of calculation is reduced as compared with the case where square root calculation is performed using a general Newton-Raphson method or the like. Therefore, there is an effect that implementation with a low specification computer becomes possible. [0026] 08-05-2019 8 According to the invention of claim 7, in addition to the effect of the invention according to any one of claims 1 to 6, the deterioration of the feedback gain estimation accuracy due to the estimation of the feedback gain by the ratio of the stationary noise signal is prevented. It has the effect of being able to [0027] According to the invention of claim 8, in addition to the effect of the invention according to any one of claims 1 to 7, the feedback gain is estimated in the double talk state in which the voice component is applied in the middle of the path for estimating the feedback gain. There is an effect that it is possible to prevent the deterioration of the feedback gain estimation accuracy due to the problem. [0028] According to the invention of claim 9, in addition to the effect of the invention according to any one of claims 1 to 8, since the division operation can be avoided by common division, the amount of operation required for selecting the maximum value can be reduced. It has the effect of [0029] Hereinafter, an embodiment in which the present invention is applied to a loudspeaker communication device (interphone terminal) will be described in detail with reference to the drawings. However, the present invention is not limited to this, as long as it is a loudspeaker communication device installed in the entire living space. [0030] In this embodiment, as shown in FIG. 2, a line output amplifier G2 for amplifying a transmission signal to a microphone 1, a speaker 2, a two-line to four-line conversion circuit 3, a microphone amplifier G1, and a line (transmission line of two lines). A line input amplifier G3 for amplifying a received signal from the line, a speaker amplifier G4, a transmission volume adjustment amplifier G5, a reception volume adjustment amplifier G6, the voice switch 10, and the first and second echo cancelers 30A and 30B. Configured 08-05-2019 9 [0031] The first echo canceller 30A has a conventionally known configuration of an adaptive filter 31A and a subtractor 32A, and adapts the impulse response of the feedback path (acoustic echo path) HAC formed by the acoustic coupling between the speaker 2 and the microphone 1 Acoustic echo is suppressed by subtracting a pseudo echo component (acoustic echo) estimated adaptively by the filter 31A and estimated from the reference signal (input signal to the speaker amplifier G4) from the output signal of the microphone amplifier G1 by the subtractor 32A. It is The second echo canceller 30B also has a conventional well-known configuration of an adaptive filter 31B and a subtractor 32B, and reflections due to impedance mismatch between the twowire to four-wire conversion circuit 3 and the transmission line The adaptive filter 31B adaptively identifies the impulse response of the feedback path (line echo path) HLIN formed by the acoustic coupling between the speaker and the microphone in the call terminal (for example, a door phone slave of an intercom system), and the reference signal By subtracting the pseudo echo component (line echo) estimated from (the input signal to the line output amplifier G2, that is, the transmission signal) from the reception signal by the subtractor 32B, the line echo is suppressed. [0032] The voice switch 10 includes a transmit side attenuator 11 for inserting a loss in a signal path on the transmit side, a receiver side attenuator 12 for inserting a loss in a signal path on the receive side, and attenuations on the transmit side and the receive side. And an insertion loss amount control unit 13 that controls the amount of loss inserted from the devices 11 and 12. The insertion loss amount control unit 13 returns a path from the output point Rout of the receiving side attenuator 12 to the input point Tin of the transmitting side attenuator 11 via the acoustic echo path HAC (hereinafter referred to as the “acoustic side feedback path”) Path from the output point Tout of the transmitting side attenuator 11 to the input point Rin of the receiving side attenuator 12 via the line echo path HLIN (hereinafter referred to as “the line side feedback path And the estimated values α ′ and β ′ of the feedback gains α and β on the acoustic side and the line side, respectively. Total loss amount calculation unit 15 for calculating the sum (sum of insertion loss amount of transmitting side attenuator and insertion loss amount of receiving side attenuator), and monitoring the transmitting signal and the receiving signal to estimate the call state, According to the estimation result and the value 08-05-2019 10 calculated by the total loss amount calculation unit 15, the transmitting side attenuator 1 And it consists of the insertion loss distribution processing unit 16 for determining the distribution of the insertion loss of the reception side attenuator 12. The first and second echo cancellers 30A and 30B and the voice switch 10 in this embodiment control hardware of a DSP (Digital Signal Processor) with software (program) for echo canceller and voice switch. Is realized by Therefore, the input and output signals (received and transmitted signals) of the voice switch 10 and the first and second echo cancelers 30A and 30B in the following description are sampled at a predetermined sampling period, and are quantized by the A / D converter. It has been [0033] The total loss amount calculation unit 15 calculates the total loss amount Lt necessary for obtaining the desired gain margin MG from the estimated values α ′ and β ′ of the acoustic side feedback gain α and the line side feedback gain β, and the value Lt Are output to the insertion loss amount distribution processing unit 16. [0034] The total loss amount distribution processing unit 16 monitors the input / output signal of the transmitting side attenuator 11 and the input / output signal of the receiving side attenuator 12, and information such as the magnitude relation of the power levels of these signals and the presence or absence of the audio signal To determine the call state (reception state, transmission state, etc.), and distribute the total loss amount Lt to the transmission side attenuator 11 and the reception side attenuator 12 at a rate corresponding to the determined communication state. Adjust the insertion loss amount of the attenuators 11 and 12. [0035] By the way, the total loss amount calculation unit 15 in the present embodiment is the amount of loss to be inserted into the closed loop based on the estimated values α ′ and β ′ of the feedback gains α and β estimated by the feedback gain estimation unit 14 as described above. There are two operation modes, an update mode for calculating the sum and adaptively updating, and a fixed mode for fixing the total loss amount to a predetermined initial value, and the first and second operation modes from the call start with the other party's call terminal The system 08-05-2019 11 operates in the fixed mode in a period until the echo cancellers 30A and 30B fully converge, and operates in the update mode in a period after the first and second echo cancellers 30A and 30B sufficiently converge. That is, in the total loss amount calculation unit 15, both the acoustic feedback gain α and the estimated values α ′ and β ′ of the channel feedback gain β continue for a predetermined time (several hundreds of milliseconds) or more from the start of the call for a predetermined threshold ε For example, it is considered that the first and second echo cancellers 30A and 30B have sufficiently converged when falling below each estimated value α 'and β' by 10 to 15 dB) at the start of the call, The operation mode is switched to the update mode in which the total loss amount is adaptively updated based on the estimated values α 'and β' after the above time point. The initial value of the total loss amount in the fixed mode is set to a value sufficiently larger than the total loss amount updated as needed in the update mode. [0036] Thus, in a state where the first and second echo cancellers 30A and 30B immediately after the start of a call do not converge sufficiently, the total loss amount calculation unit 15 operating in the fixed mode is set to a sufficiently large value. Since the initial total loss amount is inserted into the closed loop, it is possible to realize stable half-duplex communication by suppressing the occurrence of unpleasant echoes (acoustic echo and line echo) and howling. In addition, in a state where the first and second echo cancellers 30A and 30B have converged sufficiently after a lapse of time from the start of the call, the operation mode of the total loss amount 15 is switched from the fixed mode to the update mode and inserted in the closed loop. Simultaneous loss can be realized because the total amount of loss is reduced to a value sufficiently lower than the initial value. [0037] Here, the specific operation of the total loss amount calculation unit 14 in the update mode will be described with reference to the flowchart of FIG. 08-05-2019 12 [0038] Total loss amount calculation unit 15 estimates acoustic side feedback gain α and channel side feedback gain β estimated value α ′ (n ), The product α ′ (n) · β ′ (n) of β ′ (n) is read (step 1), and the gain of the closed loop is obtained from the product α ′ (n) · β ′ (n) and the gain margin MG The total loss amount desired value Lr (n) required to keep the margin at MG [dB] is calculated by the following equation (step 2). [0039] Lr (n) = 20 log | α '(n) · β' (n) | + MG [dB] Note that α '(n), β' (n) and Lr (n) are n times from the update mode transition time point respectively The estimated value of the feedback gain calculated by the sampling of and the total loss amount desired value are shown. Furthermore, the total loss amount calculation unit 15 calculates the nth total loss amount desired value Lr (n) calculated from the above equation and the previous (n−1th) total loss amount value Lt (n−1), ie, the previous one. When the total loss amount desired value Lr (n) calculated this time is larger than the total loss amount determined in the process and actually inserted, the previous total loss amount Lt (n-1) is increased slightly by Δi [ The total loss amount Lt (n) = Lt (n-1) + Δi is obtained by adding the dB) to the present total loss amount Lt (n1) + Δi (steps 3 and 4). When the total loss amount desired value Lr (n) is small, a value obtained by subtracting the slight decrease amount Δd [dB] from the previous total loss amount Lt (n-1) is the total loss amount Lt (n) = Lt (the current loss amount). It is set as n−1) −Δd (steps 5 and 6). [0040] As described above, by suppressing the increase / decrease of the total loss amount by the total loss amount calculation unit 15 to a small value of Δi or Δd, the first or second echo canceller 30A as immediately after the start of the call with the other party's call terminal Since the coefficient 30B is actively updating the coefficients toward convergence, even in a state where the change of the acoustic feedback gain α and the line side feedback gain β is severe, the sense of incongruity can be eliminated. [0041] Next, the feedback gain estimation unit 14 that is the subject matter of the present invention will 08-05-2019 13 be described with reference to FIGS. 1 and 4 to 9. FIG. 1 is a block diagram showing in detail the configuration of the feedback gain estimation unit 14 in the insertion loss control unit 13 in particular. The feedback gain estimation unit 14 is a path (hereinafter, referred to as “acoustic side feedback path”) that returns from the output point Rout of the receiving side attenuator 12 to the input point Tin of the transmitting side attenuator 11 via the acoustic echo path HAC. An acoustic side feedback gain estimation unit 14A for estimating the acoustic side feedback gain α, and a path from the output point Tout of the transmitting side attenuator 11 to the input point Rin of the receiving side attenuator 12 via the line echo path HLIN , And the line side feedback gain estimation unit 14 B for estimating the line side feedback gain β of the “line side feedback path”. [0042] The acoustic side feedback gain estimation unit 14A and the channel side feedback gain estimation unit 14B are configured such that at a certain time n, the input point Tin and the output point Tout of the transmit side attenuator 11 on the transmit path and the receive side attenuator 12 on the receive path. The transmitting signal and receiving signal Tin (n), Tout (n), Rin (n), Rout (n) are taken in from the input point Rin and the output point Rout, respectively, and the acoustic side feedback gain α (n) and the line side feedback gain The estimated values α ′ (n) and β ′ (n) with respect to β (n) are output and delivered to the total loss amount calculation unit 15. [0043] The sound side feedback gain estimation unit 14A processes the reference signal (reception signal) referred from the output point Rout of the reception side attenuator 12 and the reference signal referred from the input point Tin of the transmission side attenuator 11. And a transmitter block for processing (transmitter signal). Reference signal storage unit 52C for storing the reference signal referred to in the transmission side block, and a signal according to frequency band for obtaining a signal level for each frequency band by reading the reference signal from the reference signal storage unit 52C and performing discrete Fourier transform processing The level calculating unit 53C includes an 08-05-2019 14 envelope unit 54C that obtains a time-series average for each frequency band with respect to the signal level for each frequency band calculated by the signal level calculating unit for each frequency band 53C. In the reception side block, a transmission time difference correction unit 51A for correcting the difference in signal transmission time inherent to the acoustic side feedback path including the acoustic side echo path HAC, and a reference signal after the difference in signal transmission time is corrected. A reference signal storage unit 52A for storing (reception signal), and a frequency band signal level calculation unit 53A for obtaining a signal level for each frequency band by reading a reference signal from the reference signal storage unit 52A and performing discrete Fourier transform processing; An envelope unit 54A for obtaining a time-series average for each frequency band with respect to the signal level for each frequency band calculated by the frequency band-specific signal level calculation unit 53A is included. Further, the acoustic-side feedback gain estimation unit 14A includes a maximum feedback gain selection unit 55 and a smoothing filter unit 56A. The maximum feedback gain selection unit 55 estimates the feedback gain for each frequency band from the output signals of the receive side block and the transmit side block of the envelope portions 54A and 54C, and the maximum feedback among the feedback gains for each frequency band Perform processing to select gains. Further, the smoothing filter unit 56A performs a process of smoothing the maximum feedback gain selected by the maximum feedback gain selection unit 55 in time series, and varies the smoothed value by parameter setting as described later. . [0044] On the other hand, the channel-side feedback gain estimation unit 14 B is configured to process a reference signal (transmission signal) referenced from the output point Tout of the transmissionside attenuator 11 and input point Rin of the reception-side attenuator 12. And a receiver block that processes the reference signal (listening signal) referred to. Reference signal storage unit 52D for storing the reference signal referred to in the receiver side 08-05-2019 15 block, and a signal level for each frequency band to obtain a signal level for each frequency band by reading the reference signal from the reference signal storage unit 52D and performing discrete Fourier transform processing A calculating unit 53D and an envelope unit 54D for obtaining a time-series average for each frequency band with respect to the signal level for each frequency band calculated by the signal level calculating unit for each frequency band 53D are included. In the transmission side block, a transmission time difference correction unit 51B for correcting the difference in signal transmission time inherent to the line side feedback path including the line side echo path HLIN, and a reference after the difference in signal transmission time is corrected. Reference signal storage unit 52B for storing a signal (transmission signal) and a frequency level signal level calculation unit 53B for obtaining a signal level for each frequency band by reading a reference signal from the reference signal storage unit 52B and performing discrete Fourier transform processing And an envelope unit 54B for obtaining a time-series average for each frequency band with respect to the signal level for each frequency band calculated by the signal level calculation unit for each frequency band 53B. The line side feedback gain estimating unit 14B further includes a maximum feedback gain selecting unit 55 and a smoothing filter unit 56B. The maximum feedback gain selection unit 55 is shared with the acoustic side feedback gain estimation unit 14A, and estimates feedback gain for each frequency band from the output signals of the envelope units 54B and 54D of the transmission side block and the reception side block. And processing for selecting the largest feedback gain among the feedback gains for each frequency band. In addition, the smoothing filter unit 56B performs a process of smoothing the maximum feedback gain selected by the maximum feedback gain selection unit 55 in time series, and varies the smoothed value by parameter setting as described later. . The above-described units included in the voice switch 10 are realized by controlling the DSP hardware with a dedicated program, and the signals processed by the feedback gain estimating units 14A and 14B on the acoustic side and the line side Are all treated as sampled and quantized digital data of analog transmission and reception signals. 08-05-2019 16 [0045] Transmission specific to the system until the reception signal output from the output point Rout of the reception side attenuator 12 reaches the input point Tin of the transmission side attenuator 11 via the acoustic side feedback path including the acoustic echo path HAC It takes time. Therefore, in the transmission time difference correction unit 51A, the time at which the single pulse generated from the output point Rout of the receiver-side attenuator 12 reaches the input point Tin of the transmitter-side attenuator 11 is measured and set in advance. The reference signal from the output point Rout of the receiving side attenuator 12 is delayed by the transmission time Dα of the system. For example, when the sampling frequency of the reference signal is 8 [kHz] and the measured delay time is 12 [msec], the signal storage unit for delay processing of 8 × 12 = 96 data (separate from the reference signal storage units 52A to 52D) (FIFO (First In First Out) type signal storage unit) is prepared, and at time n, the oldest 12 [msec] previous data in the delay processing signal storage unit (not shown) is taken as DRout (n) A signal delay is realized by storing the reference signal Rout (n), which is passed to the reference signal storage unit 52A and referred to from the output point Rout of the receiving side attenuator 12 in the delay processing signal storage unit. [0046] By the way, the frequency band signal level calculators 53A to 53D combine the reference signal data DRout (n), DRout (n-1),..., DRout (n-7) of a fixed number Nf (for example, Nf = 8). Assuming that discrete Fourier transform processing is performed by reading from the reference signal storage unit 52A (see FIG. 4A), discrete Fourier transform processing can be performed only at a rate of once per Nf of interrupt processing that occurs every sampling time. There is a possibility that the estimation accuracy of the feedback gain may be reduced. Therefore, in the present embodiment, the reference signal storage units 52A to 52D are first-in first-out (FIFO) type, and the data of the reference signal storage units 52A to 52D are generated each time an interrupt for sampling occurs as shown in FIG. The reference signal data DRout (n), DRout (n-1),..., DRout (n-7) after being shifted one by one (shifted into blocks) and replaced are 08-05-2019 17 output from the reference signal storage units 52A to 52D. By performing discrete Fourier transform processing by reading into the band-specific signal level calculation units 53A to 53D, a delay is prevented from occurring and the estimation accuracy of the feedback gain is improved (see FIG. 6). [0047] Next, the method of calculating the signal level in the frequency band signal level calculation unit 53A will be described with reference to the flowchart in FIG. 5 with the number of data to be processed at one time (the length of discrete Fourier transform processing) Nf being eight. The frequency band signal level calculation unit 53A calculates a plurality of (eight in the present embodiment) reference signal data DRout (n), DRout (n-1),..., DRout (n-7) referenced in time series. Reading from the reference signal storage unit 52A (step 1), discrete Fourier transform processing is performed on the read data (step 2). In this discrete Fourier transform processing, the real part and the imaginary part are individually calculated for each of the four divided frequency bands [f0, f1, f2, f3] (step 3), and the square sum of the real part and the imaginary part is further calculated. The signal levels | FRout_f0 (n) |, | FRout_f1 (n) |, | FRout_f2 (n) | and | FRout_f3 (n) | are calculated by calculating the square root (step 4). That is, since the discrete Fourier transform is an even function with respect to frequency components, it is sufficient to calculate half of the four components if the length Nf is 8, so the sampling frequency of the reference signal is 8 kHz. In the case of Nf = 8, four frequency band components f0: 0 to 0.5 [kHz], f1: 0.5 to 1.5 [kHz], f2: 2.5 to 2.5 [kHz], f3: 2.5 to 3.5 [kHz] are calculated It is like that. In step 2 of FIG. 5, a matrix of reference signal data DRout (n), DRout (n-1),..., DRout (n-7) and signal levels for each frequency band | FRout_f0 (n) |, | FRout_f1. The relational expression of transformation with the matrix of (n) .vertline., .vertline.FRout.sub .-- f2 (n) .vertline., and .vertline.FRout.sub.f3 (n) .vertline. Then, after performing the matrix operation process of step 2, for each of the band components of [f0, f1, f2, f3], the one obtained by squaring the real part component and the imaginary part component (step 3) is added, and further By performing square root calculation processing (step 4), the magnitudes of signal levels for each band component [| FRout_f0 (n) |, | FRout_f1 (n) |, | FRout_f2 (n) |, | FRout_f3 (n) | It calculates | requires and outputs to the envelope part 54A. [0048] In an algorithm in which the value gradually approaches the true value each time the loop 08-05-2019 18 operation such as the Newton-Rapson method that is generally performed is repeated for the square root operation processing in step 4, for example, discrete Fourier transform processing Even if the method of alternately executing feedback path estimation gains α 'and β' to be described later with a length Nf = 8, loop operation is executed in each of 2 inputs × 4 bands = 8 square root operations The load on the arithmetic processing of the DSP that implements the feedback gain estimation unit 14 is increased. In the present embodiment, therefore, a list in which the results before the square root operation and the results after the square root operation are stored in advance in the non-volatile storage area, and DSP operation processing is performed by referring to the list and obtaining the square root value. The load is reduced. [0049] Specifically, in the case of finding the square root of X (0 ≦ X <1) Y = 0X (0 ≦ Y <1), after decomposing X = a × 2-b, the evenness of b is taken into consideration Y = √a × 2-n (b = 2 n, n is a positive integer) Y = √ (0.5 × a) × 2-n (b = 2 n + 1, n is a positive integer) For the remaining part of the variable a, a list in which the variable a and its square root aa are made to correspond is stored in advance in a storage area, and the list is referred to. If a square root 得 a is obtained, a desired square root Y can be obtained. However, if there is enough processing capacity of the DSP to be used, a general square root calculation algorithm such as NewtonRaphson method may be used. [0050] Here, in a situation where a call is not being made, only stationary noise components are included in the reference signal data DRout (n), DRout (n-1),..., DRout (n-7). Estimating the feedback gain by the ratio will reduce the estimation accuracy. Therefore, in the frequency band signal level calculation unit 53A, all signal levels of the reference signal data DRout (n), DRout (n1),..., DRout (n-7) read from the reference signal storage unit 52A are predetermined. If it is less than or equal to the threshold value, the discrete Fourier transform processing is stopped, and the signal level obtained in the previous discrete Fourier transform processing is substituted, and the substituted signal level is output to the envelope unit 54A. As a result, the estimation accuracy is prevented from being reduced due to the estimation of the feedback gain by the ratio of the stationary noise signal. [0051] 08-05-2019 19 Also, in the so-called double-talk state in which a voice component is applied in the acoustic echo path HAC by the speaker talking almost simultaneously between the other party's call terminal and the projector, in the so-called double talk state, as in the case where a steady noise signal exists The estimation accuracy of the feedback gain will be reduced. Therefore, in the present embodiment, when the double talk detection function is detected by the double talk detection function of the first echo canceller 30A, the discrete Fourier transform processing is stopped in the frequency band-classified signal level calculation unit 53A, and the previous discrete Fourier transform is performed. The signal level obtained by the processing is substituted, and the substituted signal level is output to the envelope unit 54A, so that the deterioration of the estimation accuracy caused by the estimation of the feedback gain in the double talk state as described above is prevented. There is. [0052] By the way, in the present embodiment, the length Nf of the discrete Fourier transform processing is set to a value (Nf = 8) considerably smaller than the general value (256 to 1024) in consideration of the load reduction of the arithmetic processing by the DSP. Therefore, the time length representing the length Nf of the discrete Fourier transform process in time units becomes very short at 8/8 [kHz] = 1 [msec] at the sampling frequency (= 8 [kH]), and noise The signal level by the frequency band is calculated by the frequency band signal level calculation unit 53A in order to improve the feedback gain estimation accuracy by the application, so that the signal level [| FRout_f0 (n) |, The envelope unit 54A performs an operation (envelope operation) for obtaining a time-series average for each frequency band with respect to | FRout_f1 (n) |, | FRout_f2 (n) |, | FRout_f3 (n) | reference). This envelope calculation is an operation represented by the following equation for signal levels by frequency band [| FRout_f0 (n) |, | FRout_f1 (n) |, | FRout_f2 (n) |, | FRout_f3 (n) |]. The time series average is obtained by performing However, the envelope coefficient ρ in the following equation is the measured value [| FRout_f0 (n) |, | FRout_f1 (n) |, | FRout_f2 (n) |, | FRout_f3 (n) |] and the average value [env_RO_f0 (n), env_RO_f1 (n), env_RO_f2 (n), env_RO_f3 (n)] is a numerical value that determines the weighting. [0053] env_RO_f0 (n) = (1-ρ) env_RO_f0 (n-1) + ρ | FRout_f0 (n) | env_RO_f1 (n) = (1-ρ) env_RO_f1 (n-1) + ρ | FRout_f1 (n) | env_RO_f2 (n) = (1-RO) env_RO_f2 (n-1) + ρ | FRout_f2 (n) | env_RO_f3 (n) = (1-ρ) env_RO_f3 (n-1) + ρ | FRout_f3 (n) | For example, 1 [kHz for reference 08-05-2019 20 signal In the case of a sine wave signal], as shown in FIG. 7A, the 8-point Fourier transform processing (time length 1 [msec]) with a length Nf of discrete Fourier transform of 8 is shown in FIG. As shown, the frequency spectrum values after conversion match (= 0.5) with the 16-point Fourier transform processing (time length 2 [msec]) which doubles the length Nf of the discrete Fourier transform to 16 and both accuracy There is no difference in On the other hand, when a noise component is applied to a sine wave signal, the 8-point Fourier transform processing (see FIG. 8 (a)) having a shorter time length has a noise component compared to the 16-point Fourier transform processing (see FIG. 8 (b)). Since the contribution of is large, the accuracy drop becomes large. However, when envelope processing is used in the eight-point Fourier transform processing, since the periodicity of the data in the past can be utilized from the data of the reference signal storage unit 52A processed as shown in FIG. Even in the worst case immediately before leaving the reference signal storage unit 52A, it can be seen that the accuracy is improved as compared with the doubled 16-point Fourier transform processing (however, in FIG. 8C, the envelope coefficient を is ρ = 1/16). [0054] Next, the operation of the maximum feedback gain selection unit 55 will be described in the case where the length Nf of the discrete Fourier transform processing is eight. The maximum feedback gain selection unit 55 sets each frequency band component at time n calculated by the transmission time difference correction unit 51A to the envelope unit 54A using the output point Rout of the receiving side attenuator 12 as a reference point in the sound side feedback gain estimation unit 14A. The output values [env_RO_f0 (n), env_RO_f1 (n), env_RO_f2 (n), env_RO_f3 (n)] of the envelope section 54A of the reference signal storage section 52C from the reference signal storage section 52C with the input point Tin of the transmitting side attenuator 11 as a reference point. By dividing the output value of the envelope unit 54C for each frequency band component calculated by the envelope unit 54C [env_TI_f0 (n), env_TI_f1 (n), env_TI_f2 (n), env_TI_f3 (n)] for each frequency band component ( The acoustic side feedback gain estimated values [env_α'f0 (n), env_α'f1 (n), env_α'f2 (n), env_α'f3 (n)] are calculated (see the following equation) (FIG. 9 (a) reference). [0055] env_TI_f0 (n) / env_RO_f0 (n) ≡env_α'f0 (n) env_TI_f1 (n) / env_RO_f1 (n) ≡env_α'f1 (n) env_TI_f2 (n) / env_RO_f2 (n) ≡env_α'f2 (n) env_TI_f3 (n) n) / env_RO_f3 (n) ≡env_α'f3 (n) In addition, in the line side feedback gain estimating unit 14B, the transmission time difference correction unit 51B is calculated by the envelope unit 54B using the output point Tout of the transmission side attenuator 11 as a reference point. The output value [env_TO_f0 (n), env_TO_f1 (n), env_TO_f2 (n), env_TO_f3 (n)] of the envelope unit 54B for each frequency band 08-05-2019 21 component at time n is used as the input point Rin of the receiving side attenuator 12 as a reference point. The output value [env_RI_f0 (n), env_RI_f1 (n), env_RI_f2 (n), env_RI_f3 (n)] of the envelope unit 54D for each frequency band component calculated by the envelope unit 54D from the reference signal storage unit 52D (Refer to the equation below) to obtain the estimated values of the feedback gain on the network side [env_β'f0 (n), env_β'f1 (n), env_β'f2 (n), env_β'f3 (n)]. And are (see FIG. 9 (b)). [0056] env_RI_f0 (n) / env_TO_f0 (n) ≡env_β'f0 (n) env_RI_f1 (n) / env_TO_f1 (n) ≡env_β'f1 (n) env_RI_f2 (n) / env_TO_f2 (n) ≡env_β 'f2 (n) env_RI_f3 (n) n) / env_TO_f3 (n) ≡env_β'f3 (n) Furthermore, acoustic side gain estimated values [env_α'f0 (n), env_α'f1 (n), env_α'f2 (n), env_α'f3 (n)] and Product for each frequency band component of the line side gain estimated value [env_β'f0 (n), env_β'f1 (n), env_β'f2 (n), env_β'f3 (n)] env_α'f0 (n) × env_β ' f0 (n) env_α'f1 (n) × env_β'f1 (n) env_α'f2 (n) × env_β'f2 (n) env_α'f3 (n) × env_β'f3 (n) has the largest value Select frequency band components. This corresponds to identifying the frequency component with the least margin among the frequency dependencies of the howling margin of the closed loop loop gain in the interphone system, and it is possible to accurately estimate the frequency component with the most margin and the level thereof. Become. For example, as shown in FIG. 9C, if the frequency band f1 is a band having the maximum value of the feedback gain estimated value, the maximum feedback is achieved by the components env_α'f1 (n) and env_β'f1 (n) of the frequency band f1. Let the maximum feedback gains [α1 (n), β1 (n)] at time n selected by the gain selection unit 55 be α1 (n) = env_α′f1 (n) β1 (n) = env_β′f1 (n) This is output to the acoustic side smoothing filter unit 56A and the line side smoothing filter unit 56B. However, when obtaining the maximum value of the product for each frequency band component, the amount of computation can be reduced if the product is shared. That is, the maximum value can be obtained without dividing by dividing the product for each frequency band component, and the load on arithmetic processing can be reduced by avoiding the division processing which is considered to be poor by the DSP. Can. Of course, division processing may be performed if there is a margin in the arithmetic processing capability of the DSP. [0057] Finally, the operation of the smoothing filter units 56A and 56B will be described with reference 08-05-2019 22 to FIG. However, since the smoothing processing of the sound side smoothing filter unit 56A and the line side smoothing filter unit 56B is common, the sound side smoothing filter unit 56A will be described below. [0058] First, before starting the smoothing process, variables UsCounter and DsCounter are respectively initialized to preset parameters Us_Init and Ds_Init. Also, the acoustic-side feedback gain estimated value α ′ (n) after the averaging process is also initialized to the initial value (= 0) (step 1). Subsequently, the maximum feedback gain α1 (n) output from the maximum feedback gain selection unit 55 is read, and the acoustic-side feedback gain estimated value α1 (n) at time n is averaged after time n−1. Compared to the feedback gain estimated value α '(n-1), if α1 (n) is larger (α1 (n)> α' (n-1)), the variable UsCounter is decremented and the variable DsCounter is initialized On the other hand, if α1 (n) is smaller (α1 (n) <α '(n-1)), the variable DsCounter is decremented and the variable UsCounter is initialized. If both are equal (α1 (n) = α '(n), continue the process (step 2). Then, the variable UsCounter is 0, that is, the acoustic feedback gain estimated value α1 (n) at time n in step 2 is the acoustic feedback gain estimated value α ′ (n− after averaging processing at time n−1. 1) If the larger case continues continuously, the processing of setting α '(n-1) plus the parameter ε to α' (n) is performed, and the variable DsCounter is 0, that is, in step 2. If the acoustic side feedback gain estimated value α1 (n) at time n is smaller than the acoustic side feedback gain estimated value α ′ (n−1) after the averaging process at time n−1, then α continues as A process in which the value obtained by subtracting the parameter ε from '(n−1) is taken as α ′ (n) is performed. Otherwise, the previous value is held as α ′ (n) = α ′ (n−1) (step 3). [0059] Thus, in the smoothing filter unit 56A, the acoustic-side feedback gain estimated value α '(n) after averaging processing is increased or decreased according to the magnitude relationship between the parameters Us_Init and Ds_Init representing the initial values of the variables UsCounter and DsCounter. It is possible, and if Us_Init> Ds_Init, the acoustic-side feedback gain estimated value α ′ (n) can be induced in a direction smaller than the time-series average calculated by the envelope unit 54A, conversely, Us_Init If <Ds_Init, then the acoustic-side feedback gain estimated value α ′ (n) can be derived in a direction in which the value becomes larger than the time-series average calculated by the envelope unit 54A. That is, if the feedback gain is estimated to be an excessively large value by time-series averaging by the envelope unit 54A, there is a risk that a howling may occur due to an erroneous estimation. It is desirable to 08-05-2019 23 derive the acoustic side feedback gain estimated value α ′ (n) in a direction in which the value becomes smaller than the time-series average calculated by the envelope unit 54A. For example, in the present embodiment, (Us_Init, Ds_Init) = (64, 1). [0060] Although the processing flow using the output point Rout of the receiver-side attenuator 12 of the acoustic-side feedback gain estimation unit 14A as a reference point has been mainly described, the input point Tin of the transmitter-side attenuator 11 and the transmitter-side attenuator 11 The same processing is performed with the exception of the presence or absence of the transmission time difference correction unit 51, with regard to the processing flow in which the output point Tout of and the input point Rin of the receiver-side attenuator 12 are used as reference points. [0061] By the way, in the case where the feedback gain estimation unit 14 is configured by a DSP, the DSP performs interrupt processing for obtaining feedback gain estimated values on the acoustic side and the line side at every sampling time when A / D converts a reference signal. The burden of arithmetic processing becomes quite large. Therefore, as opposed to the interrupt processing generated every A / D sampling time in the DSP, in the present embodiment, the acoustic side feedback gain unit 14A and the line side feedback gain unit 14B are alternately operated to process the processing load on the DSP. It is reduced. Specifically, the transmission time difference correction unit 51, the reference signal storage unit 52, the maximum feedback gain selection unit 55, and the smoothing filter unit 56 are always operated, and the frequency band signal level calculation unit 54 and the envelope unit 55 are operated. Operate / stop exclusively every time an interrupt occurs. At that time, in the maximum feedback gain selection unit 55, for example, the acoustic side feedback gain estimated value α ′ (2k + 1) is updated at time 2k + 1, and the acoustic side feedback gain estimated value β ′ (2k) is calculated at time 2k. In the case of updating, at time 2k + 1, acoustic side feedback gain estimated values [env_α'f0 (2k + 1), env_α'f1 (2k + 1), env_α'f2 (2k + 1), env_α'f3 (2k Product of each frequency band component of the frequency band component of the channel side feedback gain estimated value [env_β 'f0 (2 k), env_β' f1 (2 k), env_β 'f2 (2 k), env_β' f3 (2 k)] (2k + 1) × env_β′f0 (2k) env_α′f1 (2k + 1) × env_β′f1 (2k) env_α′f2 (2k + 1) × env_β′f2 (2k) env_α′f3 (2k + 1) At time 2k, from among × env_β'f3 (2k), the acoustic side feedback gain estimated value [env_α'f0 (2k-1), env_α'f1 (2k1), env_α'f2 (2k-1), For each frequency band component of env_α 'f3 (2k-1)] and estimated 08-05-2019 24 value of feedback gain on the network side [env_β' f0 (2 k), env_β 'f1 (2 k), env_β' f2 (2 k), env_β 'f3 (2 k)] Product of env_α'f0 (2k-1) x env_β'f0 (2k) env_α'f1 (2k-1) x env_β'f1 (2k) env_α'f2 (2k-1) x env_β'f2 (2k) env_α'f3 (2k-1) × env_β 'f Select the frequency band component having the maximum value from 3 (2 k). Determine maximum feedback gains [α 1 (n), β 1 (n)] at time n = 2k + 1 or n = 2 k, and output them to acoustic side smoothing filter unit 56A and line side smoothing filter unit 56B, respectively. . However, if there is enough processing capacity of the DSP, the feedback gains on the acoustic side and the line side may be simultaneously updated in one interrupt process. [0062] It is a block diagram which shows the insertion loss amount control part in one Embodiment of this invention. It is a block diagram which shows the same as the above. It is a flowchart for operation | movement description of the audio | voice switch in said same. It is operation | movement explanatory drawing of the signal level calculation part according to frequency band in the same as the above. It is a flowchart for operation | movement description of the signal level calculation part classified by frequency band in the same as the above. It is operation | movement explanatory drawing of the envelope part in the same as the above. It is operation | movement explanatory drawing of the envelope part in the same as the above. It is operation | movement explanatory drawing of the envelope part in the same as the above. It is operation | movement explanatory drawing of the largest feedback gain selection part in same as the above. It is operation | movement explanatory drawing of the smoothing filter part in the same as the above. Explanation of sign [0063] 13 Insertion loss control unit 14A Acoustic side feedback gain estimation unit 14B Line side feedback gain estimation unit 51A, 51B Transmission time difference correction unit 52A to 52D Reference signal storage unit 53A to 53D Frequency band signal level calculation unit 54A to 54D Envelope unit 55 Maximum feedback gain selector 56A, 56B smoothing filter 08-05-2019 25

1/--страниц