JP2007053473A - Acoustic echo canceling device, telephone using the same, and acoustic echo canceling method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an echo canceler capable of executing learning by highly accurately detecting a double talk state, and to provide a telephone set employing the same and an echo cancellation method. <P>SOLUTION: Power processors 21, 23 operate call reception power indicating power of a call reception signal from a far-end caller, and residual power indicating power of a residual signal as a residue of canceled acoustic echo. An execution propriety determining section 22 determines whether update execution processing for updating a filter coefficient should be executed depending on the call reception power and the residual power. When the determining section 22 determines that the update processing should be executed, an update processing executer 25 executes update processing. As described above, since the residual signal as an output signal of a subtractor 26 for an index to be used for propriety of update processing, the propriety of update processing can be prevented from being influenced by noises such as an acoustic echo or an external sound included in an input signal of the subtractor 26. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  INDUSTRIAL APPLICABILITY The present invention is used in a voice hands-free device such as a speakerphone telephone, and a speaker output signal is input to a microphone via an acoustic echo path. More particularly, the present invention relates to an acoustic echo canceling apparatus, a telephone using the acoustic echo canceling method, and an acoustic echo canceling method capable of detecting a double talk state with high accuracy and performing learning normally.

Conventionally, in an audio hands-free device such as a speakerphone telephone, an acoustic echo canceling device is used to prevent howling and echo. In such an acoustic echo canceling device, a reception signal from the far-end speaker and a transmission signal from the near-end speaker are detected, respectively, and the far-end speaker and the near-end speaker speak simultaneously. An acoustic echo canceling device that can detect a talk state is known (see, for example, Patent Document 1). By doing so, learning can be stopped during double talk, and erroneous learning during double talk can be prevented.
JP 11-289282 A

  However, since the transmission signal from the near-end speaker includes noises such as echoes and external sounds, the double talk state may not be detected with high accuracy. As a result, even during double talk, the filter coefficient of the adaptive filter is updated, learning is not performed normally, and there is an inconvenience that hinders conversation.

  Therefore, in view of the above problems, the present invention provides an acoustic echo canceling apparatus, a telephone using the acoustic echo canceling apparatus, and an acoustic echo canceling method capable of accurately detecting a double talk state and performing learning normally. For the purpose.

  An acoustic echo canceling apparatus according to the present invention is an acoustic echo canceling apparatus that estimates an acoustic echo using an adaptive filter, removes the acoustic echo using the estimated pseudo echo, and updates a filter coefficient of the adaptive filter. Power calculation unit for calculating the reception power indicating the power of the reception signal from the user and the residual power indicating the power of the residual signal that is the residual of the acoustic echo, and the update processing for updating the filter coefficient An update is performed when the execution determination unit and the execution determination unit determine that the update process should be executed, in accordance with the received power and residual power calculated by the calculation unit. And an update process execution unit that executes the process.

  The acoustic echo canceling apparatus according to the present invention uses a residual signal that is an output signal of a subtractor as an index used for whether or not update processing is possible. Noise such as external sounds can be prevented from being affected. Thereby, since the double talk state can be detected with high accuracy, erroneous learning during double talk can be suppressed.

  A first invention made to solve the above problem is an acoustic echo canceling apparatus that estimates an acoustic echo using an adaptive filter, removes the acoustic echo using the estimated pseudo echo, and updates a filter coefficient of the adaptive filter. The power calculation unit for calculating the reception power indicating the power of the reception signal from the far-end speaker and the residual power indicating the power of the residual signal, which is the unerased acoustic echo, and the filter coefficient is updated. The execution determination unit and the execution determination unit determine whether or not to execute the update process according to the received power and the residual power calculated by the power calculation unit. In this case, an update process execution unit is provided for executing the update process.

  According to this configuration, since the residual signal, which is the output signal of the subtractor, is used as the index used for whether or not the update process is possible, the acoustic echo or the external sound included in the input signal of the subtractor is determined whether or not the update process is possible. It is possible to prevent the noise from being affected. Thereby, since the double talk state can be detected with high accuracy, erroneous learning during double talk can be suppressed.

  2nd invention made | formed in order to solve the said subject is 1st acoustic echo cancellation apparatus, Comprising: An execution availability determination part is set to the power ratio of reception power and residual power calculated by the electric power calculation part. Accordingly, it is determined whether or not to execute the update process.

  According to this configuration, since it is determined whether or not the update process can be performed with only one parameter indicated by the power ratio between the received power and the residual power, the update process can be performed easily. Thereby, the mislearning at the time of double talk can be suppressed with a simple process.

  3rd invention made | formed in order to solve the said subject is 1st or 2nd acoustic echo cancellation apparatus, Comprising: The feasibility determination part further has the received power calculated by the electric power calculation part from predetermined value Only when the value is larger, it is determined that the update process should be executed.

  According to this configuration, when the received power is smaller than the predetermined value, the update process is not executed. Therefore, when the received power is relatively small, the update process is affected by the noise included in the input signal of the subtractor. What can not be done properly can be suppressed. Thereby, even when the received power is relatively small, the double talk state can be detected with high accuracy.

  A fourth invention made to solve the above-described problem is any one of the first to third acoustic echo canceling apparatuses, and further according to the received power and the residual power calculated by the power calculation unit. The filter section control unit controls the section of the adaptive filter.

  According to this configuration, it is possible to balance the convergence and stability in the acoustic echo cancellation process according to the learning situation indicated by the received power and the residual power. Thereby, mislearning at the time of double talk can be suppressed, suppressing acoustic echo efficiently.

  A fifth invention made to solve the above-mentioned problems comprises any one of the first to fourth acoustic echo canceling apparatuses, a speaker that outputs a received signal as sound, and a microphone that receives sound input. Is a telephone in which the sound output from is input to the microphone as an acoustic echo.

  According to this configuration, mislearning during double talk can be suppressed, so that conversation can be performed without hindrance.

  A sixth invention made to solve the above problem is an acoustic echo cancellation method for estimating an acoustic echo by an adaptive filter, removing the acoustic echo by the estimated pseudo echo, and updating a filter coefficient of the adaptive filter. The update processing for calculating the filter coefficient is calculated by calculating the received power indicating the power of the received signal from the far-end speaker and the residual power indicating the power of the residual signal that is the residual of the acoustic echo. In accordance with the received power and the residual power, it is determined whether or not to execute, and when it is determined that the update process should be executed, the update process is executed.

(Embodiment 1)
FIG. 1 is a hardware block diagram illustrating an example of a speakerphone telephone. A telephone to which the present invention is applied is, for example, a speakerphone type telephone 1, as indicated by a broken line in FIG. The speakerphone system telephone 1 is an example of a so-called hands-free device that uses a speaker and a microphone built in the telephone without using a handset. The telephone is not limited to a fixed telephone and may be mobile communication such as a mobile phone.

  Specifically, the speakerphone telephone 1 includes a main IC (Integrated Circuit) 2, a speaker 3, a microphone 5, a ROM (Read Only Memory) 6, a RAM (Random Access Memory) 7, a communication interface 8, and an A / D. Converters 9a and 9c, D / A converters 9b and 9d, and amplifiers 10a and 10b are included. The main IC 2 includes a DSP (Digital Signal Processor) 20 that functions as an acoustic echo canceling device, peripheral buses 30a and 30b, and a program data bus 40. The ROM 6 stores a program for operating the DSP 20, and the RAM 7 is used as a data area for reading the program. In addition, although DSP20 was shown as an example of an acoustic echo cancellation apparatus, if it functions as an acoustic echo cancellation apparatus, it does not need to restrict to this, For example, CPU (Central Processing Unit) may be sufficient.

  A telephone line such as a PSTN (Public Switched Telephone Networks) is connected to the communication interface 8 as shown in FIG. The communication interface 8 is connected to the speaker 3 via the A / D converter 9a, the peripheral bus 30a, the DSP 20, the peripheral bus 30b, the D / A converter 9b, and the amplifier 10a. A reception path RP is constituted by a path from the communication interface 8 to the speaker 3. On the other hand, the microphone 5 is connected to the communication interface 8 via the amplifier 10b, the A / D converter 9c, the peripheral bus 30b, the DSP 20, the peripheral bus 30a, and the D / A converter 9d. A transmission path SP is constituted by a path from the microphone 5 to the communication interface 8. The DSP 20 is connected to the ROM 6 and the RAM 7 via the program data bus 40.

  FIG. 2 is a functional block diagram illustrating an example of the DSP 20. The DSP 20 includes a received signal integration unit 21, a learning availability determination unit 22, a residual signal integration unit 23, a filter 25, and a subtraction unit 26, as indicated by a broken line in FIG. The filter 25 includes an adaptive filter such as an FIR (Finite Impulse Response) filter. The reception signal integration unit 21 is connected to the reception route RP, and the residual signal integration unit 23 is connected to the transmission route SP. The received signal integration unit 21 and the residual signal integration unit 23 are connected to the filter 25 via the learning availability determination unit 22. The filter 25 is connected to the reception route RP. The subtractor 26 is provided in the transmission path SP, and the filter 25 is connected to the transmission path SP via the subtractor 26.

  The received signal integration unit 21 and the residual signal integration unit 23 function as a power calculation unit. The learning availability determination unit 22 functions as an execution availability determination unit. The filter 25 functions as an update process execution unit.

  In the above configuration, the operation of the speakerphone telephone 1 according to the first embodiment will be described with reference to FIGS. FIG. 3 is an explanatory diagram of acoustic echo cancellation processing in the first embodiment based on the correlation between the power ratio ERR and the received signal integrated value Pr.

  When the near-end speaker speaks, the voice uttered by the near-end speaker is input to the microphone 5 as a transmission signal s (n) as shown in FIG. The transmission signal s (n) is output to the telephone line via the transmission path SP and transmitted to a telephone (not shown) on the far end speaker side.

  On the other hand, when the far-end speaker speaks, the voice uttered by the far-end speaker is transmitted to the speakerphone type telephone 1 through the telephone line as a reception signal x (n) as shown in FIG. . The received reception signal x (n) is output as sound from the speaker 3 through the reception path RP.

  At this time, since the speaker 3 radiates the received signal x (n) as a sound to the space, the emitted sound is a direct route to the microphone 5 where the sound is not reflected, or the sound is within the space. The signal is input to the microphone 5 via an echo path EP (a one-dot chain line on the right side of FIG. 1) including an indirect path that reflects off an object such as a wall, floor, or ceiling. Hereinafter, a signal in which radiated sound is input to the microphone 5 via the echo path EP is referred to as “acoustic echo d (n)”.

  When receiving the reception signal x (n), the speakerphone telephone 1 executes a known acoustic echo cancellation process. That is, as shown in FIG. 2, the filter 25 performs echo by the steepest descent method such as a normalized LMS (Least Mean Square) algorithm based on the received received signal x (n) and the acoustic echo d (n). A transfer function of the path EP is estimated, and a pseudo echo y (n) is generated based on the transfer function. The subtraction unit 26 subtracts the generated pseudo echo y (n) from the acoustic echo d (n) (that is, removes the acoustic echo d (n)), and uses the difference as a residual signal e (n). Output.

  Furthermore, the speakerphone telephone 1 executes a known filter coefficient update process in the acoustic echo cancellation process described above. That is, when the subtractor 26 subtracts the pseudo echo y (n) and outputs it as the residual signal e (n), the filter 25 receives the residual signal e (n) and receives the received signal x (n). , The filter coefficient correction value is calculated, and the filter coefficient is updated. When the filter coefficient is updated, the filter 25 generates a pseudo echo y (n) based on the updated filter coefficient, and subtracts the pseudo echo y (n) from the acoustic echo d (n) as described above.

  In the first embodiment, the above-described update process is set to be executed when the power ratio ERR exceeds a predetermined threshold. The “power ratio ERR” refers to a received power indicating the power of the received signal x (n) (“received signal integrated value” described below) Pr and a residual power indicating the power of the residual signal e (n) (described later). In this Embodiment 1, the power ratio is expressed as ERR = Pe / Pr. The reception signal integration unit 21 shown in FIG. 2 integrates the received reception signal x (n) and calculates the reception power as the reception signal integration value Pr. On the other hand, the residual power is calculated as a residual signal integrated value Pe by integrating the residual signal e (n) input by the residual signal integrator 23 shown in FIG.

  That is, when only the far-end speaker starts speaking in a state where the acoustic echo cancellation processing is being executed (that is, when the far-end call state is entered), the reception power increases, but the near-end speaker does not speak. The power (residual power) of the transmission signal s (n) decreases. Since the power ratio ERR is represented by ERR = Pe / Pr as described above, it is a relatively small value.

  The reception signal integration unit 21 outputs the calculated reception signal integration value Pr to the learning availability determination unit 22, and similarly, the residual signal integration unit 23 uses the calculated residual signal integration value Pe as a learning availability determination unit. 22 for output. The learning possibility determination unit 22 calculates the power ratio ERR using the received signal integration value Pr and residual signal integration value Pe.

The learning possibility determination unit 22 determines whether or not the power ratio ERR exceeds the power ratio threshold value ERR _SH , and when it is determined that the power ratio ERR exceeds the power ratio threshold value ERR _SH , Instructs not to execute the update process. On the other hand, when it is determined that the power ratio ERR does not exceed the power ratio threshold ERR _SH , the learning availability determination unit 22 instructs the filter 25 to execute the update process.

For example, since the calculated power ratio ERR is a relatively small value, when it corresponds to the operating point P1 in the area AR10 shown in FIG. 3, the power ratio ERR does not exceed the power ratio threshold ERR _SH , so learning is performed. The availability determination unit 22 determines that the power ratio ERR does not exceed the power ratio threshold ERR _SH , and instructs the filter 25 to execute the update process. In response to this instruction, the filter 25 executes the update process described above.

  Next, when the far-end speaker speaks and the near-end speaker starts speaking (that is, when the double-talk state is entered), the acoustic echo d (n) includes the transmission signal s (n). The remaining unerased acoustic echo d (n) output from the subtractor 26 increases. As a result, the residual signal integral value Pe increases. On the other hand, since the far-end speaker continues speaking, it is assumed that the received signal integrated value Pr is substantially maintained.

The power ratio ERR moves from the operating point P1 in FIG. 3 to the right as shown by the arrow. When the power ratio ERR moves from the area AR10 shown on the left side of FIG. 3 to the area AR11 shown on the right side of the figure, the learning possibility determination unit 22 determines that the power ratio ERR exceeds the power ratio threshold value ERR _SH. . Then, the learnability determination unit 22 instructs the filter 25 not to execute the update process. In response to this instruction, the filter 25 stops the update process described above.

  As described above, when only the far-end speaker speaks, the operating point is included in the area AR10 shown in FIG. 3 and the update process is executed. On the other hand, when the far-end speaker and the near-end speaker speak simultaneously, the operating point becomes The update process is stopped in the area AR11 shown in FIG. Therefore, the update process is not executed in the double talk state, so that mislearning during double talk can be suppressed, and the conversation between the far-end speaker and the near-end speaker can be performed without any trouble.

  As described above, in the acoustic echo cancellation apparatus according to the first embodiment, the residual signal e (n) that is the output signal of the subtracting unit 26 is used as an index used to determine whether or not the update process can be performed. Whether or not noise such as an acoustic echo d (n) and an external sound included in the input signal of the subtractor 26 can be prevented from being affected. Thereby, since the double talk state can be detected with high accuracy, erroneous learning during double talk can be suppressed.

  In addition, since it is determined whether or not the update process can be executed with only one parameter indicated by the power ratio ERR between the received power and the residual power, the update process can be easily executed. Thereby, the mislearning at the time of double talk can be suppressed with a simple process.

  In the above-described first embodiment, the case where the update process is executed according to the power ratio ERR has been described. However, it is determined whether the update process should be executed according to the received power and the residual power. If it determines, it will not be restricted to this. For example, the absolute values of the received power and the residual power may be used.

  In the above-described first embodiment, whether or not only the update process can be executed according to the received power and the residual power has been described. Further, during the execution of the update process, according to the received power and the residual power, It is also possible to control the section of the filter coefficient. At this time, the learning possibility determination unit 22 functions as a filter section control unit.

  A filter coefficient section (for example, 1 to M) is set in the filter 25, and the learning possibility determination unit 22 determines a filter coefficient section n from the set section during the update process. To do. In this case, when the power ratio ERR is small, a sufficient amount of erasure of the acoustic echo d (n) can be ensured. Therefore, the section n is shortened giving priority to the high-speed convergence of the acoustic echo cancellation processing. On the other hand, when the power ratio ERR is large, a sufficient amount of erasure of the acoustic echo d (n) cannot be ensured, so the section n is shortened giving priority to the stability of the acoustic echo cancellation process.

  By doing so, it is possible to balance high-speed convergence and stability in the acoustic echo cancellation processing in accordance with the learning situation indicated by the received power and residual power. As a result, it is possible to suppress erroneous learning during double talk while efficiently suppressing the acoustic echo d (n).

(Embodiment 2)
Since the configuration of the speakerphone type telephone 1 in the second embodiment is the same as that shown in FIGS. 1 and 2, the description thereof is omitted.

  With the above configuration, the operation of the speakerphone telephone 1 according to Embodiment 2 will be described with reference to FIGS. 1, 2, and 4. FIG. FIG. 4 is an explanatory diagram of acoustic echo cancellation processing in the second embodiment based on the correlation between the power ratio ERR and the received signal integrated value Pr. Since the acoustic echo cancellation process is the same as that in the first embodiment, only the difference in the update process will be described.

As in the first embodiment, the learning availability determination unit 22 determines whether or not the update process should be executed according to the power ratio ERR. The learning availability determination unit 22 further determines the received power from a predetermined value. It is set to determine that the update process should be executed only when the value is larger. Here, it is assumed that the predetermined value is set as the integration threshold value Pr _SH .

For example, when neither the far-end speaker nor the near-end speaker speaks, the received signal integrated value Pr indicating the received power indicates a relatively small value. At this time, when the received signal integration value Pr becomes smaller than the integration threshold value Pr _SH , the operating point P11 is included in the area AR20 shown in FIG. Therefore, the learning possibility determination unit 22 determines that the received signal integration value Pr is smaller than the integration threshold value Pr _SH and instructs the filter 25 not to execute the update process. That is, in the area AR20, it is regarded as a silent state, and the update process is not executed.

In this state, when only the near-end speaker starts speaking, the transmission signal s (n) is input to the subtracting unit 26, so that the residual signal integrated value Pe increases. When the residual signal integral value Pe increases, the power ratio ERR exceeds the power ratio threshold value ERR _SH (that is, the operating point moves to the right in FIG. 4), but the received signal integral value Pr becomes the integral threshold value Pr _SH. In the smaller state, the operating point P12 is included in the area AR21 shown in FIG. Therefore, the learning possibility determination unit 22 continues to instruct the filter 25 not to execute the update process. That is, in the area AR21, it is regarded as a near-end call, and the update process is not executed as in the silent state.

Next, when the near-end speaker stops speaking and the far-end speaker starts speaking, the received signal integrated value Pr increases, so that the power ratio ERR decreases. When the received signal integral value Pr becomes larger than the integral threshold value Pr _SH and the power ratio ERR becomes smaller than the power ratio threshold value ERR _SH , the operating point moves in the upper left direction in FIG. 4, and the operating point P13 is shown in FIG. It is included in area AR22. Therefore, the learning possibility determination unit 22 instructs the filter 25 to execute the update process. That is, in the area AR22, the update process is executed by regarding the far-end call.

Further, when the near-end speaker also starts speaking, the residual signal integrated value Pe increases again. When the power ratio ERR increases and exceeds the power ratio threshold value ERR _SH , the operating point P14 is included in the area AR23 shown in FIG. Accordingly, the learning possibility determination unit 22 determines that the received signal integration value Pr is larger than the integration threshold value Pr _SH , instructs the filter 25 not to execute the update process, and the update process is stopped. That is, in the area AR23, execution of the update process is stopped by regarding the double talk state.

  Thus, the operating point is included in any of the areas AR20, AR21, AR22, AR23 shown in FIG. 4 and the operating point is included in the area AR22 according to the values of the reception signal integrated value Pr and the power ratio ERR. Only when the update process is executed. Therefore, unlike the first embodiment, the second embodiment considers the area AR20 as a silent state and does not execute the update process, so that the noise included in the input signal of the subtracting unit 26 is relatively low when the received power is relatively small. It is possible to prevent the update process from being performed properly under the influence of the above.

  As described above, in the acoustic echo cancellation apparatus according to the second embodiment, when the received power is relatively small, the update process cannot be appropriately performed due to the influence of noise included in the input signal of the subtracting unit 26. Therefore, even if the received power is relatively small, the double talk state can be detected with high accuracy.

  In the second embodiment, as in the first embodiment, the update process is not limited to this as long as it is determined whether or not the update process should be executed according to the received power and the residual power. . Further, in the first embodiment described above, as in the first embodiment, it is possible to control the section of the filter coefficient according to the received power and the residual power during the update process.

  Since the present invention accurately detects the double talk state, it can suppress mislearning during double talk, and can be applied to a hands-free device that uses a speaker and a microphone built in the telephone without using a handset. I can do it. It can also be applied to mobile communications such as fixed telephones and mobile phones.

Hardware block diagram showing an example of a speakerphone phone Functional block diagram showing an example of a DSP Explanatory drawing of the acoustic echo cancellation processing in Embodiment 1 based on the correlation between the power ratio and the received signal integral value Explanatory drawing of the acoustic echo cancellation process in the second embodiment based on the correlation between the power ratio and the received signal integral value

Explanation of symbols

1 Telephone (Speakerphone type telephone)
3 Speaker 5 Microphone 20 Acoustic echo cancellation device (DSP)
21 Power calculator (received signal integrator)
22 Executability determination unit, filter section control unit (learning determination unit)
23 Power calculator (residual signal integrator)
25 Adaptive filter, update processing execution unit (filter)
d (n) acoustic echo e (n) residual signal ERR power ratio ERR _SH predetermined value (power ratio threshold)
Pe residual power (residual signal integral value)
Pr Received power (received signal integral value)
y (n) Pseudo echo x (n) Received signal

Claims (6)

An acoustic echo canceling apparatus that estimates an acoustic echo with an adaptive filter, removes the acoustic echo with the estimated pseudo echo, and updates a filter coefficient of the adaptive filter,
A power calculator that calculates a received power indicating the power of a received signal from a far-end speaker and a residual power indicating a power of a residual signal that is an unerased residual of the acoustic echo;
An execution determination unit that determines whether or not to update the filter coefficient according to the received power and the residual power calculated by the power calculation unit;
An update process execution unit that executes the update process when the execution determination unit determines that the update process should be executed;
An acoustic echo canceling device. The acoustic echo canceling device according to claim 1,
The execution availability determination unit determines whether to perform the update process according to a power ratio of the received power and the residual power calculated by the power calculation unit.
An acoustic echo canceling device. The acoustic echo canceling device according to claim 1 or 2,
The execution availability determination unit further determines that the update process should be executed only when the received power calculated by the power calculation unit is greater than a predetermined value.
An acoustic echo canceling device. The acoustic echo canceling device according to any one of claims 1 to 3, further comprising:
A filter section control unit that controls a section of the adaptive filter according to the received power and the residual power calculated by the power calculation unit;
An acoustic echo canceling device. The acoustic echo canceling device according to any one of claims 1 to 4,
A speaker for outputting the received signal as voice;
A microphone that accepts voice input,
The sound output from the speaker is input to the microphone as the acoustic echo.
A telephone characterized by that. An acoustic echo cancellation method for estimating an acoustic echo with an adaptive filter, removing the acoustic echo with the estimated pseudo echo, and updating a filter coefficient of the adaptive filter,
Calculating the received power indicating the power of the received signal from the far-end speaker, and the residual power indicating the power of the residual signal that is the residual of the acoustic echo,
It is determined whether or not the update process for updating the filter coefficient should be executed according to the calculated received power and residual power,
If it is determined that the update process should be executed, the update process is executed.
An acoustic echo canceling method characterized by the above.

Images