JP5049256B2 - Echo canceller and echo canceling method - Google Patents

Description

  The present invention relates to an acoustic echo cancellation technique, and more particularly to an acoustic echo cancellation technique suitable for a conference system having a microphone and a speaker.

In Patent Document 1, in a telephone conference system or a video conference system, using a microphone array having a plurality of microphone elements, a band in which only speaker sound exists is determined from a phase difference between microphones, and an acoustic echo canceller is applied only to that band. A technique for adaptively controlling the above is disclosed. According to this technique, the echo canceller can be controlled dynamically according to the situation of the conference room or the like, and the acoustic echo suppression performance can be enhanced.
JP 2008-141718 A

  In the technique described in Patent Document 1, an audio signal input to a microphone array is divided into a plurality of frequency bands, and an acoustic path is learned and estimated for each frequency band.

  By the way, it is difficult to perform complete estimation in an actual use environment, and an estimation error may actually occur and a residual echo that is a residual component of an acoustic echo may occur. For this reason, in the technique described in Patent Document 1, the size of the acoustic echo occupied in the input audio signal is estimated for each frame, and transmission of a frame in which this exceeds a predetermined threshold is blocked by a voice switch (patent) (Refer to paragraph 0034 of literature 1). However, if this is done, the transmission of the voice signal is interrupted instantaneously, but the full-duplex call state cannot be maintained.

  The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a technique capable of further enhancing the echo suppression effect of an acoustic echo canceller.

  In order to solve the above problems, the present invention suppresses residual echo after adaptation to an acoustic echo canceller without interrupting frame transmission. In consideration of the difference between the characteristics of the residual echo component in the speech signal after adaptation to the acoustic echo canceller and the speech component, the residual echo component is estimated and suppressed from the sound pressure spectrum information of the speech signal after adaptation to the acoustic echo canceller. .

For example, the echo canceller of the present invention is
An echo canceler that eliminates an acoustic echo of an output audio signal that is an audio signal output from a speaker from an input audio signal that is an audio signal input to a microphone,
Pseudo echo generating means for estimating the acoustic echo of the output audio signal for each frequency band and generating a pseudo echo signal;
Echo cancellation means for subtracting the pseudo echo signal from the input audio signal to cancel the acoustic echo of the output audio signal from the input audio signal;
For the input audio signal in which the acoustic echo is canceled, for each frequency band, based on the signal component in the range adjacent to the signal component in the frequency band in the time direction and the frequency direction, the acoustic signal is extracted from the signal component in the frequency band. And a residual echo canceling means for canceling a residual echo that is a residual component of the echo.

  According to the present invention, the echo suppression effect of the acoustic echo canceller can be further enhanced.

An embodiment of the present invention will be described by taking as an example the case where the present invention is applied to an echo canceller of a microphone array apparatus used in a conference system including a microphone and a speaker. Note that the microphone array device has a function of transmitting / receiving audio signals via an IP network and an acoustic echo canceller function.
[First embodiment]
FIG. 1 is a schematic configuration diagram of an echo canceller 1 according to the first embodiment of the present invention.

  As illustrated, the echo canceller 1 includes a far-end speech signal input unit 101, a near-end speech signal input unit 102, FFT units 103 and 104, a pseudo echo generation unit 105, a linear echo canceller unit 106, a spectrum, A subtraction unit 107; a residual echo canceling unit 108; an IFFT unit 109; and a near-end audio signal output unit 110.

  The far-end audio signal input unit 101 is an input terminal for a digital audio signal (hereinafter referred to as a far-end audio signal) received from a call partner. Note that the far-end audio signal is output from the speaker after DA conversion.

  The near-end audio signal input unit 102 is an input terminal for a digital audio signal (hereinafter referred to as a near-end audio signal) input to the microphone and subjected to AD conversion. The near-end voice signal is transmitted to the other party after the acoustic echo is erased by the echo canceller 1.

  The FFT unit 103 performs FFT (Fast Fourier Transform) on the far-end audio signal input to the far-end audio signal input unit 101 and outputs frequency domain information of the far-end audio signal.

  The FFT unit 104 performs FFT on the near-end audio signal input to the near-end audio signal input unit 102 and outputs frequency domain information of the near-end audio signal.

  The pseudo echo generation unit 105 estimates an acoustic echo component of the far end speech signal using an adaptive filter for each predetermined frequency band based on the frequency domain information of the far end speech signal output from the FFT unit 103. Thus, the frequency domain information of the pseudo echo signal is generated.

  When the reference signal (here, the far-end audio signal) is input to the filter, the adaptive filter is a change that occurs until the reference signal is output from the speaker and transmitted as a sound wave to reach the microphone. As long as it is a filter created to output the result (signal) of applying an equivalent change to its reference signal. Such an adaptive filter is created as a result of acoustic environment learning performed in accordance with the call situation. Various existing techniques including the technique described in Patent Document 1 can be used for the pseudo echo generation unit 105 and the adaptive filter.

  The linear echo canceller unit 106 subtracts the frequency domain information of the pseudo echo signal from the frequency domain information of the near end audio signal output from the FFT unit 104 to eliminate the acoustic echo included in the near end audio signal.

  The spectral subtraction unit 107, for the sound pressure spectrum information of the near-end speech signal from which the acoustic echo has been eliminated by the linear echo canceller unit 106, for each predetermined frequency band, the sound pressure, and the sound pressure spectrum information of the pseudo echo signal Are suppressed according to the sound pressure in the same frequency band. The spectrum subtraction unit 107 can use an existing spectrum subtraction technique.

  The residual echo canceling unit 108, for example, in units of one frame shift of the sound pressure spectrum information of the near-end speech signal output from the spectral subtraction unit 107, and the frequency band component and time direction of interest for each predetermined frequency band The residual component of the acoustic echo (residual echo) is eliminated from the component of the frequency band of interest based on the components in the adjacent range in the frequency direction.

  If there is no error in the acoustic environment learning necessary for creating the adaptive filter of the pseudo echo generation unit 105, sufficient acoustic echo suppression can be expected. However, the acoustic environment in actual use changes constantly. For this reason, it is difficult to completely learn the acoustic environment, and some errors or errors occur in the learning result. For this reason, in the sound pressure spectrum information of the near-end speech signal output from the spectrum subtraction unit 107, there are residual echoes that could not be eliminated. Therefore, in the present embodiment, in view of the fact that the characteristic of the residual echo component is different from the characteristic of the conversational voice component in the sound pressure spectrum information of the near-end speech signal output from the spectrum subtraction unit 107, the residual echo canceling unit By 108, the residual echo component is estimated from the sound pressure spectrum information, and the residual echo is suppressed. Details of the residual echo canceller 108 will be described later.

  IFFT section 109 performs IFFT (Inverse FFT) on the frequency domain information of the near-end audio signal output from residual echo cancellation section 108 and outputs a near-end audio signal.

  The near-end audio signal output unit 110 is an output terminal for the near-end audio signal output from the IFFT unit 109.

  Next, details of the residual echo canceling unit 108 will be described.

  As shown in FIG. 1, the residual echo canceller 108 includes a residual echo estimator 1081 and a residual echo suppressor 1082.

  The residual echo estimator 1081 performs a frequency band of interest, a time direction, and a frequency for each predetermined frequency band with respect to, for example, one frame shift unit of the sound pressure spectrum information of the near-end speech signal output from the spectrum subtraction unit 107. At least one component having a high sound pressure level is selected from components in a range adjacent in the direction. Then, based on the comparison result between the sound pressure level of the selected component and the sound pressure level of the component in the frequency band of interest, it is estimated whether or not the residual echo is included in the component of the frequency band of interest.

  The residual echo suppression unit 1082 generates a sound from a frequency band component estimated by the residual echo estimation unit 1081 to include a residual echo from components in a range adjacent to the frequency band in the time direction and the frequency direction. Select at least one component with a low pressure level. Then, according to the difference between the sound pressure level of the selected component and the sound pressure level of the component in this frequency band, the component in this frequency band is suppressed.

  FIG. 2 is a diagram for explaining a first example of residual echo estimation / suppression processing. Here, the far-end audio signal input to the far-end audio signal input unit 101 and the near-end audio signal input to the near-end audio signal input unit 102 are digital audio signals with a sampling frequency of 32 kHz, and the FFT units 103 and 104 However, it is assumed that the input digital audio signal is divided into 1024 frequency bands by FFT with a frame length of 2048 points and the frame shift unit is 1024 points (32 ms).

  In FIG. 2, reference numeral 20 denotes sound pressure spectrum information of the near-end speech signal output from the spectrum subtraction unit 107. Here, the vertical axis 21 is frequency and the horizontal axis 22 is time. As shown in the figure, the sound pressure spectrum information 20 of the near-end audio signal is divided into 1024 frequency bands, and the frame shift unit 23 is 32 ms.

  The residual echo estimation unit 1081 pays attention to each of 1024 frequency bands, for example, for each frame shift unit, and determines whether or not there is a residual echo component in the frequency band of interest in the following manner. .

  Let the frequency band of interest be the bandwidth of interest A. First, the residual echo estimation unit 1081 determines a processing target block including the target band A and the eight comparison candidate frequency bands B1 to B8. Specifically, from the frame shift unit immediately before the frame shift unit to which the target band A belongs, the same frequency band B4 as the target band A and frequency bands B1 and B6 adjacent to the upper and lower sides of the target band A are selected as comparison candidate frequency bands. To do. Further, frequency bands B2 and B7 adjacent to the top and bottom of the target band A are selected as comparison candidate frequency bands from the frame shift unit to which the target band A belongs. Further, from the frame shift unit immediately after the one frame shift unit to which the target band A belongs, the same frequency band B5 as the target band A and frequency bands B3 and B8 adjacent to the upper and lower sides of the target band A are selected as comparison candidate frequency bands.

  Next, the residual echo estimation unit 1081 compares the sound pressure level of the comparison candidate frequency bands B1 to B8 in the processing target block and the sound pressure level of the target band A, selected as described above, and determines the target band A. It is determined whether or not the sound pressure level deviates from the sound pressure levels of the comparison candidate frequency bands B1 to B8. Specifically, as shown in the condition 24 of FIG. 2, the sound pressure level D1 having the highest sound pressure level in the target band A among the sound pressure levels in the comparison candidate frequency bands B1 to B4 and B6 to B8, and It is determined whether or not it is higher than at least one of the highest sound pressure levels D2 among the sound pressure levels of the comparison candidate frequency bands B1 to B3 and B5 to B8. As a result, if the sound pressure level of the target band A deviates from the sound pressure levels of the comparison candidate frequency bands B1 to B8 (if it is higher than at least one of the sound pressure level D1 and the sound pressure level D2), this It is estimated that a residual echo component exists in the band A of interest.

  The residual echo suppression unit 1082 determines the sound pressure level of the target band A estimated by the residual echo estimation unit 1081 to have a residual echo from the sound pressure level of the comparison candidate frequency bands B1 to B8. Suppress it so that it does not diverge. Specifically, as indicated by condition 25 in FIG. 2, a value obtained by subtracting the lower sound pressure level of the sound pressure levels D1 and D2 from the sound pressure level of the target band A is used as the sound of the target band A. The pressure level suppression amount S is determined. Then, the sound pressure level of the band A of interest is suppressed by the suppression amount S. Thereby, the residual echo component which exists in the attention zone A is suppressed.

  For example, the sound pressure level of the target band A is 50, and the sound pressure levels of the comparison candidate frequency bands B1, B2, B3, B4, B5, B6, B7, and B8 are 10, 20, 10, 20, 60, 10, 10 and 10 are assumed. In this case, the highest sound pressure level D1 of comparison candidate frequency bands B1 to B4 and B6 to B8 is 20, and the highest sound pressure level D2 of comparison candidate frequency bands B1 to B3 and B5 to B8 is 60. Since the sound pressure level “50” of A is higher than one of the sound pressure levels D1, it is estimated that a residual echo component exists in the band A of interest. Then, a value “30” obtained by subtracting the lower sound pressure level “20” of the sound pressure levels D1 and D2 from the sound pressure level “50” of the target band A is the suppression amount S of the sound pressure level of the target band A. The sound pressure level of the band A of interest is suppressed by this suppression amount S. As a result, the sound pressure level of the band A of interest is “20”.

  FIG. 3 is a diagram for explaining a second example of the residual echo estimation / suppression process. Here, the far-end audio signal input to the far-end audio signal input unit 101 and the near-end audio signal input to the near-end audio signal input unit 102 are digital audio signals with a sampling frequency of 32 kHz, and the FFT units 103 and 104 However, it is assumed that the input digital audio signal is divided into 1024 frequency bands by FFT with a frame length of 2048 points and the frame shift unit is 512 points (16 ms).

  In FIG. 3, the same components as those in FIG. As shown in the figure, the sound pressure spectrum information 20 is divided into 1024 frequency bands, and the frame shift unit 23 is 16 ms.

  The residual echo estimation unit 1081 pays attention to each of the 1024 frequency bands for each frame shift unit, and determines whether or not there is a residual echo component in the frequency band of interest (target band) as follows. Judge with.

  Let the frequency band of interest be the bandwidth of interest A. First, the residual echo estimation unit 1081 determines a processing target block including the target band A and the 16 comparison candidate frequency bands C1 to C16. Specifically, from the frame shift unit three frames before the frame shift unit to which the target band A belongs, the same frequency band C7 as the target band A and the frequency bands C1 and C11 adjacent to the upper and lower sides of the target band A are compared with the comparison candidate frequency bands. Elected as. Further, from the frame shift unit two frames before the frame shift unit to which the target band A belongs, the same frequency band C8 as the target band A and the frequency bands C2 and C12 adjacent above and below the target band A are selected as the comparison candidate frequency bands. . Further, frequency bands C3 and C13 adjacent to the top and bottom of the target band A are selected as comparison candidate frequency bands from the frame shift unit immediately before the frame shift unit to which the target band A belongs. Further, frequency bands C4 and C14 adjacent to the top and bottom of the target band A are selected as comparison candidate frequency bands from the frame shift unit to which the target band A belongs. Further, from the frame shift unit immediately after the frame shift unit to which the target band A belongs, the same frequency band C9 as the target band A and frequency bands C5 and C15 adjacent to the upper and lower sides of the target band A are selected as the comparison candidate frequency bands. Further, the frequency band C10 that is the same as the target band A and the frequency bands C6 and C16 that are adjacent to the upper and lower sides of the target band A are selected as comparison candidate frequency bands from the second frame shift unit to which the target band A belongs. .

  Next, the residual echo estimation unit 1081 compares the sound pressure level of the comparison candidate frequency bands C1 to C16 in the processing target block selected as described above with the sound pressure level of the target band A, and the target band A It is determined whether or not the sound pressure level deviates from the sound pressure levels in the comparison candidate frequency bands C1 to C16. Specifically, as shown in the condition 26 of FIG. 3, the lower one of the sound pressure levels of the comparison candidate frequency bands C1 and C2 is the lower one of the sound pressure levels of the comparison candidate frequency bands C3 and C4. E2, the lower of the sound pressure levels of the comparison candidate frequency bands C5 and C6, E3, the lower of the sound pressure levels of the comparison candidate frequency bands C7 and C8, and the sound of the comparison candidate frequency bands C9 and C10. The lower of the pressure levels is E5, the lower of the sound pressure levels of the comparison candidate frequency bands C11 and C12 is E6, the lower of the sound pressure levels of the comparison candidate frequency bands C13 and C14 is E7, and the comparison candidate The lower one of the sound pressure levels in the frequency bands C15 and C16 is defined as E8. The sound pressure level in the bandwidth A of interest is the highest sound pressure level F1 among the sound pressure levels E1 to E4 and E6 to E8, and the highest sound pressure level F2 among the sound pressure levels E1 to E3 and E5 to E8. It is judged whether it is higher than at least one of these. As a result, if the sound pressure level of the target band A deviates from the sound pressure levels of the comparison candidate frequency bands C1 to C16 (if higher than at least one of the sound pressure level F1 and the sound pressure level F2), this target It is estimated that there is a residual echo component in band A.

  The residual echo suppression unit 1082 uses the sound pressure level of the target band A estimated by the residual echo estimation unit 1081 to have a residual echo component, and the sound pressure level of the target band A of the comparison candidate frequency bands C1 to C16. Suppress it so that it does not deviate from. Specifically, as shown in the condition 27 of FIG. 3, a value obtained by subtracting the lower sound pressure level of the sound pressure levels F1 and F2 from the sound pressure level of the target band A is used as the sound pressure of the target band A. The level suppression amount S is determined. Then, the sound pressure level of the band A of interest is suppressed by the suppression amount S. Thereby, the residual echo component which exists in the attention zone A is suppressed.

  2 and 3, when the highest frequency band or the lowest frequency band is the attention area A, there is no frequency band higher or lower than the attention band A. These exceptional handlings are not so important in canceling residual echoes in practical use. For example, the suppression amount S may be determined by regarding the sound pressure level in the adjacent frequency band that does not actually exist as 0. Alternatively, the suppression amount S may be determined to 0 unconditionally. The same applies to the case of the third example described below with reference to FIG.

  The first embodiment of the present invention has been described above.

  The residual echo estimation / suppression process shown in FIG. 2 and FIG. 3 includes the characteristics of the residual echo component included in the near-end speech signal from which the acoustic echo has been eliminated by the linear echo canceller unit 106 and the spectral subtraction unit 107, and the human conversation speech. Utilizing the fact that the characteristics of the components are different. In the sound pressure spectrum information of the near-end audio signal from which the acoustic echo has been eliminated, the characteristics of each component will be described with an arbitrary frequency band as the target band A.

  When the human conversation voice component is mainly in the target band A, the sound pressure level of the component of the target band A is close to the sound pressure level of the component of the frequency band adjacent to the target band A in the time direction and the frequency direction. The tendency to take is strong.

  On the other hand, when the residual echo component is dominant in the target band A, the sound pressure level of the component of the target band A tends to take a value that deviates from the sound pressure level of the frequency band component adjacent to the target band A in the frequency direction. Is strong. This is because acoustic environment learning and acoustic echo cancellation processing are performed for each frequency band, so there is a difference for each frequency band in the acoustic echo cancellation amount by the linear echo canceller unit 106 and the spectral subtraction unit 107. This is because the sound pressure level of the residual echo component that could not be canceled due to the above does not necessarily take a close value between neighboring bands.

  Also, residual echo components with high sound pressure tend to exist in particular discretely. This is because, in an echo canceller in which acoustic environment learning and acoustic echo cancellation processing operate in a stable manner, a residual echo component with high sound pressure is generated in the frequency band and time when the estimation error of acoustic echo is large. This is because the probability of occurrence thereof (as a proportion of all frequency band components in the shift unit) is low.

  Also, in the time direction (when viewed in a very short time), if the same frequency band is used, acoustic echo cancellation processing is performed using a common adaptive filter, and therefore, several frame shift times (for example, about 64 ms) A residual echo component with a close sound pressure level may continue. However, in a speech signal that has been subjected to acoustic echo cancellation processing of a general conversational speech, it is rare that a residual echo component exists continuously for a longer time.

  In the present embodiment, in view of the difference between the characteristics of the residual echo component in the near-end speech signal after the acoustic environment learning and echo cancellation processing for each frequency band as described above, and the characteristics of the human speech component, For the near-end audio signal output from the spectral subtraction unit 107, the echo canceling unit 108, for each frequency band, based on the frequency band of interest and the components in the range adjacent in the time direction and the frequency direction, The residual echo is eliminated from the component.

  Specifically, with respect to the near-end speech signal output from the spectral subtraction unit 107, the residual echo estimation unit 1081 determines, for each frequency band, from the frequency band of interest and the components in the adjacent range in the time direction and the frequency direction. Then, select at least one component with a high sound pressure level, and based on the comparison result between the sound pressure level of the selected component and the sound pressure level of the component of the frequency band of interest, residual echo is present in the component of the frequency band of interest. Estimate whether it is included. The residual echo suppression unit 1082 then transmits a signal having a low sound pressure level from components in a frequency band that is estimated to contain residual echo from components in a range adjacent to the frequency band in the time direction and the frequency direction. A component is selected, and the component in this frequency band is suppressed according to the difference between the sound pressure level of the selected component and the sound pressure level of the component in this frequency band.

Therefore, according to the present embodiment, it is possible to suppress residual echo that cannot be completely erased by the linear echo canceller unit 106 and the spectral subtraction unit 107, while suppressing distortion (suppression) of conversational speech, thereby reducing the echo canceller. 1 echo suppression effect can be enhanced.
[Second Embodiment]
FIG. 4 is a schematic configuration diagram of an echo canceller 1A according to the second embodiment of the present invention.

  As illustrated, the echo canceller 1A according to the present embodiment is different from the echo canceller 1 according to the first embodiment shown in FIG. 1 in that a residual echo canceling unit 108A is provided instead of the residual echo canceling unit 108. It is.

  Further, the residual echo canceling unit 108A is different from the residual echo canceling unit 108 in that there are a plurality of residual echo estimating units 1081 having different frequency band ranges that are comparison candidate frequency bands, and a plurality of residual echo estimating units 1081 Among them, a selection unit 1083 for selecting the residual echo estimation unit 1081 used for estimation of the residual echo is provided.

  The selection unit 1083 is output from the sound pressure spectrum information of the near-end speech signal output from the FFT unit 104, the sound pressure spectrum information of the pseudo echo signal generated by the pseudo echo generation unit 105, and the linear echo canceller unit 106. The size of the acoustic echo is estimated based on the sound pressure spectrum information of the near-end speech signal after the acoustic echo is erased. Then, the residual echo estimation unit 1081 is selected so that the range of the comparison candidate frequency band is widened according to the size of the acoustic echo.

  For example, the total value of the sound pressure levels in the entire frequency band of the pseudo echo signal is larger than the total value of the sound pressure levels in the entire frequency band of the near-end speech signal after acoustic echo cancellation by a certain percentage (predetermined reference value) or more. In this case, that is, when the acoustic echo is larger than the reference value with respect to the near-end speech signal, the residual echo estimation unit 1081 having a wide comparison candidate frequency band range, that is, a higher residual echo suppression effect is selected. . Then, the selected residual echo estimator 1081 is caused to perform residual echo estimation.

  FIG. 5 is a diagram for explaining a third example of residual echo estimation / suppression processing. Here, the far-end audio signal input to the far-end audio signal input unit 101 and the near-end audio signal input to the near-end audio signal input unit 102 are digital audio signals with a sampling frequency of 32 kHz, and the FFT units 103 and 104 However, it is assumed that the input digital audio signal is divided into 1024 frequency bands by FFT with a frame length of 2048 points and the frame shift unit is 512 points (16 ms).

  5, the same components as those in FIG. 3 are denoted by the same reference numerals. As in the case shown in FIG. 3, the sound pressure spectrum information 20 is divided into 1024 frequency bands, and the frame shift unit 23 is 16 ms.

  The residual echo estimation unit 1081 pays attention to each of the 1024 frequency bands for each frame shift unit, and determines whether or not there is a residual echo component in the frequency band of interest (target band) as follows. Judge with.

  Let the frequency band of interest be the bandwidth of interest A. First, the residual echo estimation unit 1081 determines a processing target block including the target band A and the 28 comparison candidate frequency bands C1 to C28. Specifically, from the frame shift unit three frames before the frame shift unit to which the target band A belongs, the same frequency band C13 as the target band A, frequency bands C7 and C17 adjacent to the upper and lower sides of the target band A, and the frequency band C7. , C17 and C17 adjacent to C17 are selected as comparison candidate frequency bands. Further, from the frame shift unit immediately before the frame shift unit to which the target band A belongs, the same frequency band C14 as the target band A, the frequency bands C8 and C18 adjacent to the upper and lower sides of the target band A, and the frequency bands C8 and C18. Adjacent frequency bands C2 and C24 are selected as comparison candidate frequency regions. Further, from the frame shift unit immediately before the frame shift unit to which the target band A belongs, the frequency bands C9 and C19 adjacent to the upper and lower sides of the target band A and the frequency bands C3 and C25 adjacent to the frequency bands C9 and C19 are compared with the comparison candidate frequencies. Elected as a band. Further, from the frame shift unit to which the target band A belongs, the frequency bands C10 and C20 adjacent to the upper and lower sides of the target band A and the frequency bands C4 and C26 adjacent to the frequency bands C10 and C20 are selected as comparison candidate frequency bands. Further, from the frame shift unit immediately after the frame shift unit to which the target band A belongs, the same frequency band C15 as the target band A, adjacent frequency bands C11 and C21 above and below the target band A, and adjacent to the frequency bands C11 and C21. The frequency bands C5 and C27 are selected as comparison candidates. Further, from the frame shift unit two frames after the frame shift unit to which the target band A belongs, the same frequency band C16 as the target band A, the frequency bands C12 and C22 adjacent to the upper and lower sides of the target band A, and the frequency bands C12 and C22. Adjacent frequency bands C6 and C28 are selected as comparison candidate frequency bands.

  Next, the residual echo estimation unit 1081 compares the sound pressure level of the comparison candidate frequency bands C1 to C28 in the processing target block and the sound pressure level of the target band A, selected as described above, and calculates the target band A. It is determined whether or not the sound pressure level deviates from the sound pressure levels in the comparison candidate frequency bands C1 to C16. Specifically, as shown in the condition 28 in FIG. 5, the lower one of the sound pressure levels in the comparison candidate frequency bands C1, C2, C7, and C8 is set to E1, and the comparison candidate frequency bands C3, C4, C9, and C10. The lower one of the sound pressure levels is E2, the lower one of the sound pressure levels in the comparison candidate frequency bands C5, C6, C11, and C12 is the lower one of the sound pressure levels in the comparison candidate frequency bands C13 and C14. E4, a lower one of the sound pressure levels of the comparison candidate frequency bands C15, C16 is E5, a lower one of the sound pressure levels of the comparison candidate frequency bands C17, C18, C23, C24 is E6, and a comparison candidate frequency band C19 , C20, C25, C26 among the low sound pressure levels E7, and low of the comparison candidate frequency bands C21, C22, C27, C28 among the sound pressure levels E To. And the sound pressure level of the zone A of interest is the highest sound pressure level F1 among the sound pressure levels E1 to E4 and E6 to E8, and the highest sound pressure level among the sound pressure levels E1 to E3 and E5 to E8. It is determined whether it is higher than at least one of F2. As a result, if the sound pressure level of the target band A deviates from the sound pressure levels of the comparison candidate frequency bands C1 to C28 (if it is higher than at least one of the sound pressure levels F1 and F2), the target band A It is estimated that there is a residual echo component.

  The residual echo suppression unit 1082 uses the sound pressure level of the target band A estimated by the residual echo estimation unit 1081 to have a residual echo component, and the sound pressure level of the target band A of the comparison candidate frequency bands C1 to C28. Suppress it so that it does not deviate from. Specifically, it is the same as the case of condition 27 in FIG.

  According to the third example shown in FIG. 5, the range of the comparison candidate frequency band is wide compared to the second example shown in FIG. The suppression effect is further increased. However, the distortion of conversational speech increases. Therefore, for example, in the normal case, that is, when the size of the acoustic echo is less than the reference value, the selection unit 1083 selects the residual echo estimation unit 1081 that operates in the second example illustrated in FIG. Only when the magnitude is equal to or larger than the reference value, the residual echo estimation unit 1081 operating in the third example shown in FIG. 5 is selected. Thereby, while suppressing the residual echo, the distortion of the conversation voice is prevented from occurring as much as possible.

  The second embodiment of the present invention has been described above.

  In the present embodiment, since residual echo estimation / suppression processing with different residual echo suppression effects is used depending on the size of the acoustic echo, the voice switch is used in the technique described in Patent Document 1. Even in an adverse condition where the voice must be completely blocked, the full-duplex call state can be maintained and a more comfortable call environment can be provided. Other effects are the same as those of the first embodiment.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the gist.

  For example, in each of the above-described embodiments, the configuration of the echo cancellers 1 and 1A may be implemented in hardware by an integrated logic IC such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array). Alternatively, it may be executed in software by a DSP (Digital Signal Processor). Alternatively, in a general-purpose computer such as an auxiliary storage device such as a CPU, memory, HDD, or DVD-ROM, a modem, and a PC (Personal Computer) equipped with a NIC (Network Interface Card), the CPU stores a predetermined program as an auxiliary storage device. It may be realized by loading the program from the memory into the memory and executing it.

FIG. 1 is a schematic configuration diagram of an echo canceller 1 according to the first embodiment of the present invention. FIG. 2 is a diagram for explaining a first example of residual echo estimation / suppression processing. FIG. 3 is a diagram for explaining a second example of the residual echo estimation / suppression process. FIG. 4 is a schematic configuration diagram of an echo canceller 1A according to the second embodiment of the present invention. FIG. 5 is a diagram for explaining a third example of residual echo estimation / suppression processing.

Explanation of symbols

  1, 1A: Echo canceller, 101: Far end audio signal input unit, 102: Near end audio signal input unit, 103, 104: FFT unit, 105: Pseudo echo generation unit, 106: Linear echo canceller unit, 107: Spectral subtraction 108, 108A: residual echo canceller, 109: IFFT unit, 110: near-end speech signal output unit, 1081: residual echo estimation unit, 1082: residual echo suppression unit, 1083: selection unit

Claims (4)

An echo canceler that eliminates an acoustic echo of an output audio signal that is an audio signal output from a speaker from an input audio signal that is an audio signal input to a microphone,
Pseudo echo generating means for estimating the acoustic echo of the output audio signal for each frequency band and generating a pseudo echo signal;
Echo cancellation means for subtracting the pseudo echo signal from the input audio signal to cancel the acoustic echo of the output audio signal from the input audio signal;
For the input audio signal in which the acoustic echo is canceled, the acoustic echo remains from the signal component in the frequency band based on the signal component in the range adjacent to the frequency band in the time direction and the frequency direction for each frequency band. An echo canceller comprising: a residual echo canceling unit that cancels a residual echo as a component. The echo canceller according to claim 1,
The residual echo canceling means includes:
For the input audio signal from which the acoustic echo has been canceled, at least a signal component having a high sound pressure level is selected from signal components in a range adjacent to the signal component in the frequency band in the time direction and the frequency direction for each frequency band. Based on the comparison result between the sound pressure level of the selected signal component and the sound pressure level of the signal component in the frequency band, whether or not the residual echo is included in the signal component in the frequency band Residual echo estimation means for estimating
For the signal component of the frequency band estimated to contain the residual echo, the signal component having a low sound pressure level is selected from the signal components in the range adjacent to the frequency band in the time direction and the frequency direction, An echo comprising: residual echo suppression means for suppressing the signal component of the frequency band in accordance with the difference between the sound pressure level of the selected signal component and the sound pressure level of the signal component of the frequency band. Canceller. The echo canceller according to claim 2,
The residual echo estimation means and the residual echo suppression means are:
An echo canceller, wherein a frequency range adjacent to the frequency band signal component is changed based on a sound pressure level of the pseudo echo signal. An echo cancellation method for canceling an acoustic echo of an output audio signal that is an audio signal output from a speaker from an input audio signal that is an audio signal input to a microphone,
Estimating the acoustic echo of the output audio signal for each frequency band to generate a pseudo echo signal,
Subtracting the pseudo echo signal from the input audio signal, canceling the acoustic echo of the output audio signal from the input audio signal,
For the input audio signal in which the acoustic echo is canceled, the acoustic echo remains from the signal component in the frequency band based on the signal component in the range adjacent to the frequency band in the time direction and the frequency direction for each frequency band. An echo cancellation method characterized by canceling residual echo as a component.

Images