JP4818014B2 - Signal processing device - Google Patents

Description

  The present invention relates to a signal processing apparatus and relates to a process for improving the quality of an audio signal.

Various processes are known for improving the quality of an audio signal, for example, a process for suppressing a signal other than a call signal, that is, an acoustic echo or the like when making a call in a call device or the like. In addition, it is known that high performance can be obtained by performing this process in two stages of a pre-stage process and a post-stage process.

For example, the first stage performs echo suppression processing, and the second stage performs short-term spectral amplitude estimation using the transmission output signal with echoes suppressed, which is the output of the second stage, and performs echo suppression based on the estimated value. The technique to perform is known (for example, refer patent document 1).

In the first stage, an adaptive filter performs processing to suppress echo included in the transmission input signal input from the microphone, and in the second stage, technology to suppress echo or noise based on the pseudo echo signal generated in the previous stage. It is known (for example, refer to Patent Document 2).
Japanese Patent No. 3420705 (page 2-7, FIG. 1) JP 2004-56453 A (page 2-4, FIG. 1)

However, in the method disclosed in Patent Document 1 described above, when the linear echo suppression process (adaptive filter) in the previous stage and the nonlinear echo suppression process in the subsequent stage are combined, the acoustics of the echo path can be detected at the time of double talk and echo path fluctuation. The filter coefficient of the adaptive filter is diverted assuming that the coupling amount (echo path loss) does not change significantly. For this reason, there is a problem that the echo cannot be sufficiently suppressed when the adaptive filter cannot follow, such as when the echo path loss varies during double talk.

In addition, since the method disclosed in Patent Document 2 always uses a pseudo echo signal, the echo path followability of the transmission output signal that is the final output depends on the echo path followability of the previous stage by the adaptive filter. There was a problem. In other words, it is assumed that the echo estimation by the adaptive filter can be performed with high accuracy in the first stage, and the echo tends to remain in the transmission output signal, which is the second stage output, due to a sudden echo path fluctuation or echo path loss fluctuation during double talk. There was a point.

That is, in the prior art as described above, the performance of echo suppression performed in the preceding stage is not sufficient regardless of whether the processing is linear processing or nonlinear processing, for example, when echo path loss fluctuation or abrupt echo path fluctuation during double talk is performed. In this case, in echo suppression processing or noise suppression processing performed in the subsequent stage, and further in a combination thereof, the accuracy of echo amount estimation and noise amount estimation deteriorates due to residual echo in the output signal of the previous stage. Due to this deterioration, there is a problem that the performance of the echo suppression processing performed at the later stage and the performance of the noise suppression processing are not sufficient. This problem is conspicuous also in echo suppression processing in a voice switch that is performed later.

The present invention has been made to solve the above problems, improve the echo suppression performance,
Another object of the present invention is to provide a signal processing apparatus that has improved noise suppression performance and improved robustness against echo path fluctuations including fluctuations in echo path loss.

According to the embodiment, the signal processing device inputs the first signal processing means for suppressing the echo included in the input signal and outputting the echo reduction signal, the input signal and the echo reduction signal, And second signal processing means for suppressing at least one of noise, and an echo suppression amount of the first signal processing means by calculating an amount indicating a difference between power spectra of the input signal and the echo reduction signal. And the second signal processing means determines that the echo suppression amount of the first signal processing means calculated by the echo suppression amount calculation means is not sufficient. If selects the input signal, otherwise have a selecting means for selecting the echo reduction signal, the input signal selected by said selecting means and said error It suppresses at least one of either said echo and said noise over reduction signal.

ADVANTAGE OF THE INVENTION According to this invention, the signal processing apparatus which improved the echo suppression performance or improved the noise suppression performance, and improved the robustness with respect to the echo path fluctuation | variation including the fluctuation | variation of an echo path loss can be provided.

  Embodiments of a signal processing apparatus according to the present invention will be described below with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram showing a configuration of a signal processing apparatus according to the first embodiment of the present invention.
The signal processing apparatus includes a communication unit (COM) 101, a delay processing unit (DELAY) 102,
D / A converter (D / A) 103, receiver amplifier 104, speaker 105, microphone 106, transmitter amplifier 107, A / D converter (A / D) 108, and high-pass filter section (HPF) ) 109, an echo canceller (EC) 110, and an echo reduction unit (
ER) 111.

FIG. 2 is a block diagram illustrating a configuration of the echo canceller unit (EC) 110. The echo canceller unit (EC) 110 is connected to an adaptive filter unit (ADF) 110a connected to the delay processing unit (DELAY) 102 and a high-pass filter unit (HPF) 109, and an echo reduction unit (ER) described later. ) 111 frequency domain transform processing unit (FT) 111
a signal subtraction processing unit 110b connected to c and a delay processing unit (DELAY) 102, and a frequency domain transform processing unit (FT) 111 of an echo reduction unit (ER) 111 described later.
c and a double talk detector (DTD) 110c connected to the controller (CTRL) 111k
It consists of.

FIG. 3 is a block diagram showing a configuration of the echo reduction unit (ER) 111. The echo reduction unit (ER) 111 includes a frequency domain conversion processing unit (FT) 111 a connected to the delay processing unit (DELAY) 102 and a frequency domain conversion processing unit (FT) connected to the high pass filter unit (HPF) 109. ) 111b, a frequency domain conversion processing unit (FT) 111c connected to the signal subtraction processing unit 110b of the echo canceller unit (EC) 110, a received power calculation unit (POW) 111d, and a transmission power calculation unit (POW) 111e, residual power calculation unit (POW) 111f, acoustic coupling amount estimation unit (ACLE) 111g, echo amount estimation unit (ELE) 111h, echo suppression amount estimation unit (ECLE) 111i, and frequency domain double talk Detection unit (FDTD) 111j and double-talk detection unit (DTD) 1 of echo canceller unit (EC) 110 Control unit connected with 0c (CTRL) and 111k, a spectrum selection portion 111L, gain storage section (GTBL) 111m and the echo suppression gain calculator (GC
AL) 111n, a signal suppression unit (SS) 111o, and a frequency domain inverse transform processing unit (IFT) 111p connected to the communication unit (COM) 101.

The operation of each part of the signal processing device according to the first embodiment of the present invention configured as described above will be described with reference to FIGS.

A communication unit (COM) 101 receives reception data received from a communication partner, and decodes it into a digital signal every predetermined unit of processing time, that is, every one frame (N samples). However, this sampling frequency is assumed to be f _S [Hz]. And communication department (COM)
101 receives the decoded digital signal for each frame as a received input signal x [n] (n
= 0, 1, ..., N-1).

The communication unit (COM) 101 receives the transmission signal s ′ [n] (n = 0, 1,..., N−1) output from the frequency domain inverse transform processing unit (IFT) 111p. , Encode
Output to the communication partner as transmission data.

The delay processing unit (DELAY) 102 receives the reception input signal x [n] output from the communication unit (COM) 101 for each frame, temporarily stores it, and stores it for a predetermined time (D Sample) Delayed by digital processing and output.

The D / A converter (D / A) 103 receives the received input signal x [n] output from the communication unit (COM) 101, converts it into an analog signal, and outputs it.

The receiving amplifier 104 receives the analog signal output from the D / A converter (D / A) 103, amplifies it, and outputs it.

The speaker 105 receives the amplified analog signal output from the receiving amplifier 104 and outputs it to the acoustic space as a signal x (t).

The microphone 106 collects a signal z (t) obtained by acoustically coupling the signal x (t) output from the speaker 105 to the acoustic space as described above and the transmitted voice signal s (t), and outputs an analog signal. Convert to and output.

The transmission amplifier 107 receives the analog signal output from the microphone 106, amplifies it, and outputs it.

An A / D converter (A / D) 108 receives the amplified analog signal output from the transmission amplifier 107, converts it into a digital signal for each frame, and outputs it.

The high-pass filter unit (HPF) 109 receives the digital signal output from the A / D converter (A / D) 108, removes the offset (DC component), and transmits the transmission input signal z [n].
(N = 0, 1,..., N−1) is output.

The echo canceller unit (EC) 110 transmits the transmission input signal z [n] output from the high-pass filter unit (HPF) 109 and the delayed reception input signal x [n− output from the delay processing unit (DELAY) 102. D] as an input, the echo component is suppressed from the transmission input signal z [n], and the signal after the echo suppression is used as a residual signal e [n] (n = 0, 1,..., N−1). Output as. Furthermore, double talk information ECstate [n] is output.

The adaptive filter unit (ADF) 110a has a length L filter coefficient h [i] (i = 0, 1,
..., L-1) is an adaptive filter composed of a variable transversal filter.

The adaptive filter unit (ADF) 110a receives the delayed received input signal x [n−D] output from the delay processing unit (DELAY) 102 and the remaining one sample before the echo suppression output from the signal subtraction processing unit 110b. Difference signal e [n−1] and double talk detector (DTD) 110c
The double talk information ECstate [n] output from the input is input, and when the double talk information ECstate [n] is not in the double talk state, the filter coefficient h [i] is adaptively learned for each sample n, and the double talk information is obtained. When ECstate [n] is in the double talk state, adaptive learning is not performed.

The adaptive filter unit (ADF) 110a uses the delayed received input signal x [n-D] output from the delay processing unit (DELAY) 102 and the filter coefficient h [i] to generate a pseudo echo signal y ′ [n. ] (N = 0, 1,..., N−1) are calculated and output.

The adaptive filter unit (ADF) 110a uses a fixed or variable step size μ _T [n] (n = 0, 1,..., N−1) for controlling the update width of the filter coefficient h [i]. Perform adaptive learning.

The adaptive filter unit (ADF) 110a includes, for example, an LMS (Least-Mean-Square) algorithm, an NLMS (Normalized-Least-Mean-Square) algorithm, a learning identification method, an affine projection (AP) algorithm, and a sequential. Least squares (RLS: Recursi
Non-linear adaptive algorithms such as adaptive filters based on linear adaptive algorithms such as the ve-Least-Squares algorithm, gradient-limited normalized-least-mean-square, and adaptive Volterra filters. Based on adaptive filter. In this embodiment, an example of a time domain type adaptive filter is shown, but an adaptive filter used in a subband type (band division type) or frequency domain type may be used.

The signal subtraction processing unit 110b receives the transmission input signal z [n] output from the high pass filter unit (HPF) 109 and the pseudo echo signal y ′ [n] output from the adaptive filter unit (ADF) 110a. The echo component is suppressed by subtracting the pseudo echo signal y ′ [n] from the transmission input signal z [n] for each sample n, and the residual signal e which is the signal after the echo suppression is suppressed.
[N] is output.

The double-talk detector (DTD) 110c transmits the transmission input signal z [n] output from the high-pass filter (HPF) 109 and the delayed reception input signal x [n] output from the delay processor (DELAY) 102. −D] and the residual signal e [n−1] one sample before output from the signal subtraction processing unit 110b are input, and it is determined whether or not the sample is in the double talk state for each sample n.

Specifically, the double talk detector (DTD) 110c has a power characteristic (power value or peak value; hereinafter referred to as “power characteristic”) P _Z [n] (n =
0, 1,..., N−1) and the delayed power input signal x [n−D] power characteristic P _X [n]
(N = 0, 1,..., N−1) and the power characteristic P _E [n] (n = 0,
1,..., N−1) for each sample n, and P _E [n]> λ [n] · P _X [n]
Alternatively, when P _Z [n]> δ · P _X [n], the double-talk state is determined. here,
λ [n] (n = 0, 1,..., N−1) is an estimate of the echo path loss, and the filter coefficient h [i] (i = 0, 1,. It is calculated for each sample n that has been adaptively learned, and is a variable that decreases as adaptive learning progresses and increases when adaptive learning is incorrect. Also,
δ is a fixed value that can be preset from the outside before the operation starts. And the double talk detector (
DTD) 110c is double talk information ECsta which is information indicating whether or not a double talk state exists.
te [n] is output.

In this case, the echo canceller unit (EC) 110 uses the filter coefficient h [i] (i = 0, 1).
,..., L-1), step size μ _T [n], echo path loss estimated value λ [n], double talk information ECstate [n], received input signal power characteristic P _X [n], transmission Input signal power characteristic P _Z [n], residual signal power characteristic P _E [n] (n = 0, 1,...,
N-1) is held in the memory as an internal state. Here, the internal state refers to a set of variables that change at least according to time, and a description thereof will be omitted.

Echo canceller (EC) 110 without double talk detector (DTD) 110c
It does not matter. In this case, the adaptive filter unit (ADF) 110a and the control unit (CTRL)
) 111k operates when the double talk information ECstate [n] indicates that it is not in the double talk state.

The echo reduction unit (ER) 111 includes a delayed received input signal x [n−D] output from the delay processing unit (DELAY) 102 and a transmitted input signal z [output from the high pass filter unit (HPF) 109. n] and the residual signal e [
n] and the double talk information ECs output from the double talk detector (DTD) 110c
ate [n] as an input, the echo component is suppressed from at least one of the transmission input signal z [n] or the residual signal e [n], and the signal after the echo suppression is transmitted as the transmission output signal s ′ [n. ] (
n = 0, 1,..., N−1) and output every frame.

The frequency domain transform processing unit (FT) 111a receives the delayed received input signal x [n−D] output from the delay processing unit (DELAY) 102 as an input, and performs FFT (Fast Fourier Transform).
rm) and the like, and the frequency spectrum X [f
, Ω] is calculated and output.

The frequency domain transform processing unit (FT) 111b converts the transmission input signal z [n] output from the high-pass filter unit (HPF) 109 into the frequency domain by FFT or the like, and the frequency spectrum Z [ f, ω] is calculated and output.

The frequency domain transform processing unit (FT) 111c transforms the residual signal e [n] output from the signal subtraction processing unit 110b into the frequency domain by FFT or the like, and the frequency spectrum E [f, ω] of the residual signal. Is calculated and output.

The frequency domain transformation processing unit (FT) 111a, the frequency domain transformation processing unit (FT) 111b, and the frequency domain transformation processing unit (FT) 111c appropriately use windowing with a Hamming window or the like, use past frames, or perform zero interpolation. Or overlap. For example, a signal corresponding to the number of FFT points is extracted from the past one frame and the corresponding frame, windowed by a Hamming window, and FF
Do T.

The received power calculation unit (POW) 111d receives the frequency spectrum X [f, ω] of the received input signal output from the frequency domain transform processing unit (FT) 111a, and receives the received power spectrum | X [ f, ω] | ² is calculated and output. Since the amount of acoustic coupling usually does not change abruptly in time, the amount of acoustic coupling can be estimated with higher accuracy by using a smoothed value than by using an instantaneous value. Therefore, the received power calculation unit (POW) 111d
For example, as shown in Equation 1, the received power spectrum | X _S [f, ω] | ² smoothed using the value | X _S [f−1, ω] | ² of the previous frame is calculated and output. To do. However,
α _X [ω] is preferably about 0.75 to 0.999.

The transmission power calculation unit (POW) 111e receives the frequency spectrum Z [f, ω] of the transmission input signal output from the frequency domain conversion processing unit (FT) 111b as an input, and the transmission power spectrum which is the power spectrum | Z [f, ω] | ² is calculated and output. Since the amount of acoustic coupling usually does not change abruptly in time, the amount of acoustic coupling can be estimated more accurately by using a smoothed value than by using an instantaneous value. Therefore, the transmission power calculation unit (POW) 111e.
Is, for example, the value of the previous frame as shown in Equation _{2 | Z S [f-1} , ω] | sending power spectra were smoothed using a _{^{2 | Z S [f, ω}} ] | 2 calculates the Output. However,
α _Z [ω] is preferably about 0.75 to 0.999.

The residual power calculation unit (POW) 111f receives the frequency spectrum E [f, ω] of the residual signal output from the frequency domain transform processing unit (FT) 111c as an input, and the residual power spectrum, which is the power spectrum | E [f, ω] | ² is calculated and output. Since the amount of acoustic coupling does not usually change suddenly in time, it is possible to estimate the amount of acoustic coupling more accurately by using a smoothed value than by using an instantaneous value. Therefore, the residual power calculation unit (POW) 111f is ,
For example, one frame previous value as shown in Equation _{3 | E S [f-1} , ω] | residual was smoothed using a ^second power spectrum _{| E S [f, ω]} | 2 calculated by the output To do. Where α
_E [ω] is preferably about 0.75 to 0.999.

The acoustic coupling amount estimation unit (ACLE) 111g includes a smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d, and a transmission power calculation unit (POW) 111e. Smoothed transmission power spectrum output from
Z _S [f, ω] | ² and frequency domain double talk information ERstate [f, ω] output from the frequency domain double talk detector (FDTD) 111j are input, and the performance of the echo canceller unit (EC) 110 is obtained. | E _S [f, ω] based on residual signal so as not to be affected by
Using | Z _S [f, ω] | ² based on the transmission input signal without using | ² , the acoustic coupling amount | H [f, ω] | ² for each frequency band ω, for example, Calculated by By doing in this way, compared with the signal processing apparatus which concerns on 5th Embodiment mentioned later, a calculation amount can be decreased and the memory amount required for storage of each parameter can also be decreased.

Then, the acoustic coupling amount estimation unit (ACLE) 111 g calculates and outputs the acoustic coupling amount | H _S [f, ω] | ² that is smoothed by using the value one frame before as shown in the following Expression 5. However, α _H [ω] is preferably about 0.03 to 0.99.

Here, at the initial stage of the call, for example, by increasing α _H [ω] for about 5 seconds from the start of the call, the update of the acoustic coupling amount | H _S [f, ω] | ² is accelerated. By doing so, since the acoustic coupling amount is initialized at the beginning of the call, it is possible to prevent the suppression amount from being reduced at the beginning of the call.

However, when the frequency domain double talk information ERstate [f, ω] output from the frequency domain double talk detector (FDTD) 111j is in a double talk state, or when the acoustic coupling amount changes abruptly, that is, | H _S [f, ω] | ² > β _H [ω] · | H _S [
If f−1, ω] | ² is satisfied, or if the received input signal is not sufficiently large, that is, if | X _S [f, ω] | ² <β _X [ω] is satisfied, the echo path The acoustic coupling amount estimation unit (ACLE) 111g does not update the acoustic coupling amount in order to avoid calculating the acoustic coupling amount in the frequency band that is a double talk while maintaining high-speed followability to fluctuation. 1
The past acoustic coupling amount | H _S [f−1, ω] | ² before the frame is used. Since an extreme change in the amount of acoustic coupling has the possibility of double talk, it is possible to prevent deterioration in transmitted sound quality by not updating the amount of acoustic coupling in this way. However, β _H [ω] is preferably about 0.9 to 30. β _X [ω] is desirably about 30 dB to 40 dB.

The echo amount estimation unit (ELE) 111h includes the smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d and the acoustic coupling amount estimation unit (ACLE) 111g. The output acoustic coupling amount | H _S [f, ω] | ² is used as an input, and the echo amount | Y [f, ω] | ² included in the frequency spectrum Z [f, ω] of the transmission input signal.
Is estimated and output for each frequency band ω as shown in Equation 6 below.

Then, the echo amount estimation unit (ELE) 111h can make the signal after echo suppression a more natural signal by using the smoothed value rather than using the instantaneous echo amount | Y [f, ω] | ^2. The amount of echo smoothed using the value one frame before | Y
_S [f, ω] | ² is calculated and output for each frequency band ω. However, αY [ω] is 0.7 to
About 0.99 is desirable.

The echo suppression amount estimation unit (ECLE) 111i includes a transmission power spectrum | Z [f, ω] | ² output from the transmission power calculation unit (POW) 111e, and a residual power calculation unit (POW).
The residual power spectrum | E [f, ω] | ² output from 111f is input, and an echo canceller (E
C) The echo suppression amount ECL [f, ω] suppressed at 110 is estimated and output for each frequency band ω. Specifically, it is calculated as shown in Equation 8 below. Of course, the difference between these two power spectra may be used.

The frequency domain double talk detector (FDTD) 111j is a received power calculator (POW).
Smoothed received power spectrum output from 111d | X _S [f, ω] | ²
And the smoothed residual power spectrum | E _S [f, ω] | ² output from the residual power calculation unit (POW) 111f and the frame one frame before output from the echo amount estimation unit (ELE) 111h. The echo amount | Y [f−1, ω] | ² is input, and it is determined whether or not the following Expression 9 is satisfied for each frequency band ω.

Here, the power spectrum after the echo suppression processing of the echo reduction unit (ER) 111 is substituted with the distance between the residual power spectrum and the echo amount, and it is determined whether or not it is sufficiently smaller than the received power spectrum. Use. If it is larger than the threshold β _Y [ω], that is, if Expression 9 is satisfied, it is determined that the state is a double talk state. If Expression 9 is not satisfied, it is determined that the state is not a double talk state. Such information is output as frequency domain double talk information ERstate [f, ω]. As a result, a frequency domain DTD with a small calculation amount can be realized.

Of course, an echo reduction unit (ER) 111 that does not include the frequency domain double talk detection unit (FDTD) 111j may be used. In this case, the acoustic coupling amount estimation unit (ACLE)
) 111g and control unit (CTRL) 111k are frequency domain double talk information ERstat.
The operation is performed when e [f, ω] indicates that it is not in the double talk state. However, β _Y [ω
] Is preferably about 1.0 to 20.

The control unit (CTRL) 111k includes the echo suppression amount ECL [f, ω] output from the echo suppression amount estimation unit (ECLE) 111i and the double talk information ECstate [n] output from the double talk detection unit (DTD) 110c. ] And a frequency domain double talk detector (FD
TD) frequency domain double talk information ERstate [f, ω] output from 111j, and a double talk state and echo canceller (EC) 110 for each frequency band ω.
The control information ERcontrol1 [f, ω, which is information indicating whether or not the frequency band in which the echo suppression amount is not sufficient is detected, and whether or not the frequency band in which the echo suppression amount of the echo canceller unit (EC) 110 is insufficient is detected. ] And ERcontrol2 [f, ω] are output.

First, the control unit (CTRL) 111k performs frequency domain double talk information ERstate [
f, ω] for each frequency band ω, a frequency domain double talk detector (FDTD) 111j
It is determined whether or not the double talk state is determined by the double talk detection unit (DTD) 110c of the echo canceller unit (EC) 110 using the double talk information ECstate [n]. It is determined whether or not a double talk state is set for each frequency band ω depending on whether or not it is performed.

Here, if one or both of the frequency domain double talk information ERstate [f, ω] and the double talk information ECstate [n] indicate that they are in a double talk state, the control unit (CTRL) 111k It is determined that the double talk state is set for each frequency band ω.

Thus, the double talk detector (DTD) 110 of the echo canceller (EC) 110
If the double talk information ECstate [n] output from c is used, a situation in which the echo suppression amount of the echo canceller (EC) 110 is not sufficient is likely to occur due to echo path fluctuations during double talk. The frequency band where the echo suppression amount of the unit (EC) 110 is not sufficient can be detected with high accuracy.

Next, the control unit (CTRL) 111k has ECL [f, ω]> β _Z1 [for each frequency band ω.
It is determined that the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient for the frequency band satisfying ω]. However, β _Z1 [ω] is preferably about +8 dB to −15 dB. In the present embodiment, hereinafter, β _Z1 [ω] will be described as 0 dB.

Since it is unlikely that the echo suppression amount of the echo canceller unit (EC) 110 is small only in one frequency band, that is, the performance of the echo canceller unit (EC) 110 is considered to be switched in the time domain, the control unit (CTRL) 111k is an echo canceller (E
C) In order to accurately detect a frequency band where the echo suppression amount of 110 is not sufficient,
It is determined that the amount of echo suppression is small in consideration of the number of bands count in which the inequality ECL [f, ω]> β _Z1 [ω] per frame is established.

That is, it is determined that the echo suppression amount is small when ECL [f, ω]> β _Z1 [ω] and count> β _C. However, β _C1 is desirably about 10% to 40% of the total number of frequency bands. By doing so, it is more robust to the volume difference between the received input signal and the transmitted input signal than it is determined only for each frequency band, and a frame with an insufficient echo suppression amount can be accurately detected. In addition to the determination of the double talk state for each frequency band ω, the control unit (CTRL) 111k detects the frequency band in which the double talk state and the echo suppression amount of the echo canceller unit (EC) 110 are not sufficient. Control information ERcontrol1 [f, ω
] Is output.

Similarly, the control unit (CTRL) 111k uses the equation ECL [f, ω]>
ECL [f, ω]> β _Z2 [ω] in consideration of the number of bands count in which β _Z2 [ω] is established.
When count> β _C2 , it is determined that the echo suppression amount is small, and the control information ERcontrol 2 [f, ω] is output for each frequency band ω together with the determination of the double talk state. However, β _Z2 [ω] is preferably about +8 dB to −15 dB, and β _C2 is preferably about 10% to 40% of the total number of frequency bands.

The spectrum selection unit 111L includes the frequency spectrum Z [f, ω] of the transmission input signal output from the frequency domain transform processing unit (FT) 111b and the frequency domain transform processing unit (FT) 111.
The frequency spectrum E [f, ω] of the residual signal output from c and the control unit (CTRL) 11
The control information ERcontrol1 [, which is information indicating whether or not the frequency range is the double talk state output from 1k and the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient.
f, ω] as an input, and one of the frequency spectrum Z [f, ω] of the transmission input signal and the frequency spectrum E [f, ω] of the residual signal is selected and output as a frequency spectrum.

Specifically, when the control information ERcontrol 1 [f, ω] is a frequency band that is detected as a double talk state and an echo suppression amount of the echo canceller unit (EC) 110 is not sufficient, the spectrum selection unit 111L The frequency spectrum Z [f, ω] of the transmission input signal is selected as the frequency spectrum. For the other frequency bands, the frequency spectrum E [f, ω] of the residual signal is selected as the frequency spectrum.

By doing so, when the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient for each frequency band, that is, when the estimation accuracy of the echo canceller unit (EC) 110 is not sufficient, the echo reduction unit (ER) 111 can be operated alone.

The gain storage unit (GTBL) 111m stores and outputs a parameter γ [ω] that controls a preset nonlinear echo suppression amount. However, γ [ω] is preferably about 1.0 to 2.0.

The echo suppression gain calculation unit (GCAL) 111n is a transmission power calculation unit (POW) 111.
The smoothed transmission power spectrum | Z _S [f, ω] | ² output from e and the smoothed echo amount output from the echo amount estimation unit (ELE) 111 h | Y _S [f,
ω] | ² and the parameter γ [ω] output from the gain storage unit (GTBL) 111m;
Echo suppression amount ECL [f, ω] output from the echo suppression amount estimation unit (ECLE) 111i
Control information ERcontrol2 [f, ω], which is information indicating whether the frequency range is a double talk state output from the control unit (CTRL) 111k and the echo suppression amount of the echo canceller unit (EC) 110 is insufficient. Is input and the echo suppression gain G [f, ω] is calculated and output.

Specifically, for a frequency band detected from the control information ERcontrol2 [f, ω] in a double talk state and the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient, an echo suppression gain calculation unit (GCAL) 111n Is the echo suppression gain G [f, ω
] Is calculated by Equation 10 using the Wiener Filter method. By calculating in this way, for the frequency band detected when the echo suppression amount of the double talk state and the echo canceller unit (EC) 110 is not sufficient, the echo reduction unit (ER
) The echo can be suppressed by 111 alone.

In other frequency bands, the echo suppression gain calculation unit (GCAL) 111n has a sufficient echo suppression amount of the echo canceller unit (EC) 110, and the echo canceller unit (EC).
110, the echo suppression amount ECL [f, f of the echo canceller unit (EC) 110 is increased so that the echo suppression amount of the echo reduction (ER) 111 is increased.
is corrected using [omega]], and the echo suppression gain G [f, [omega]] is calculated as shown in Equation 11 below.

By calculating in this way, compared with a signal processing apparatus according to a fifth embodiment to be described later,
The amount of calculation can be reduced, and the amount of memory required for storing each parameter can also be reduced.

Further, the echo suppression gain calculation unit (GCAL) 111n prevents the quality of the transmitted voice from being deteriorated due to excessive echo suppression, and prevents the background noise from being intermittently suppressed, so that the echo suppression gain G [f, Control is performed so that ω] does not fall below a predetermined lower limit.

FIG. 4 shows an example of the lower limit value. As indicated by the solid and broken lines in FIG. 4, the lower limit value of the echo suppression gain is bathtub-shaped for each band, that is, the lowest range (0 to f _S / 16 [Hz]) side and the highest range (7 · f _S / 16 To f _S / 2 [Hz]) side is increased from 0.04 to 0.12, and the vicinity of the middle range (f _S / 8 to 3 · f _S / 8 [Hz]) is 0.01 to 0.05. Set to be as small as possible. This is because it is harsh if an echo component that could not be suppressed remains on the lowest side or the highest side, so that the echo component can be easily suppressed and the transmitted voice can be improved in hearing. Of course, as shown by the dotted line in FIG. 4, the bathtub curve may be set as a constant value that changes stepwise.

Further, the echo suppression gain calculation unit (GCAL) 111n prevents the transmission voice quality from deteriorating due to excessive echo suppression, so that the transmission input signal z [n] or the residual signal e [
n], a background noise level may be calculated for each frequency band using a signal level of a section that is not a voiced section, and the echo suppression gain may be controlled so as not to suppress the background noise level.

Furthermore, the echo suppression gain calculation unit (GCAL) 111n sets the echo suppression gain G [f, ω] to the following expression 12-1 or expression 12- in order to prevent the transmission sound quality from deteriorating.
As indicated by 2, the output may be smoothed in the frequency direction. For example, ε _j is [0.
1, 0.2, 0.4, 0.2, 0.1], η _j is [0.1, 0.2, 0.4, 0.8,
0.4, 0.2, 0.1].

The signal suppression unit (SS) 111o receives the frequency spectrum output from the spectrum selection unit 111L and the echo suppression gain G [f, ω] output from the echo suppression gain calculation unit (GCAL) 111n as input. The echo of the frequency spectrum output from the unit 111L is suppressed and output as the spectrum S ′ [f, ω] of the transmission output signal as shown in the following expression 13-1 or expression 13-2.

At this time, even if any frequency spectrum of the frequency spectrum Z [f, ω] of the transmission input signal or the frequency spectrum E [f, ω] of the residual signal is selected by the spectrum selection unit 111L, the echo canceller unit ( Since EC) 110 is processing in the time domain, there is no difference in the phase spectrum, so that the output transmission output signal can be smoothly connected in terms of audibility.

Control information ER output from the control unit (CTRL) 111k to the spectrum selection unit 111L
The control 1 [f, ω] may be different from the control information ERcontrol 2 [f, ω] output from the control unit (CTRL) 111k to the echo suppression gain calculation unit (GCAL) 111n. For example, the values of the parameters β _Z1 [ω] and β _C1 and β _Z2 [ω] and β _C2 of the control unit (CTRL) 111k are set to different values, so that the control information ERcont
It is detected that the double talk state of control information ERcontrol2 [f, ω] and the amount of echo suppression are not sufficient so as to include the condition that the double talk state of roll1 [f, ω] is detected and the amount of echo suppression is not sufficient. Set a wide range of conditions. In this way, by using two pieces of control information, when the echo suppression gain is smoothed, it is possible to prevent deterioration of the quality of the transmitted voice.

The frequency domain inverse transform processing unit (IFT) 111p receives the frequency spectrum S ′ [f, ω] output from the signal suppression unit (SS) 111o as an input, and performs IFFT (Inverse Fast Fourier T).
ransform) or the like to calculate and output a transmission output signal s ′ [n] (n = 0, 1,..., N−1). At this time, considering the windowing of the frequency domain transform processing unit (FT) 111b and the frequency domain transform processing unit (FT) 111c as appropriate, a process of returning overlap using s ′ [n] of the past frame is performed. .

The processing flow of the signal processing apparatus according to the first embodiment configured as described above is shown in FIGS.
This will be described with reference to FIG. FIG. 5 is a flowchart showing the overall processing flow.
Is a flowchart showing the flow of processing in the echo canceller (EC) 110,
FIG. 7 is a flowchart showing the flow of processing in the echo reduction unit (ER) 111.

In FIG. 5, when there is an outgoing call or an incoming call, the communication unit (COM) 101 performs processing for establishing a communication link, and performs initial setting processing such as initialization of each parameter and each buffer (step S1001). When the communication link is established, it becomes possible to make a two-way call with the communication partner. When the two-way call is started, a decoder (not shown) in the communication unit (COM) 101 decodes every frame and receives the call. Read as input signal x [n]. Further, the transmission input signal z [n] is read through the microphone 106 (step S1002).

Then, the high pass filter unit (HPF) 109 performs an offset removal process on the transmission input signal z [n] (step S1003). The delay processing unit (DELAY) 102 performs processing for temporarily storing and delaying the received input signal x [n] (step S1004). The echo canceller unit (EC) 110 performs an echo canceller process using the delayed received input signal x [n−D] and the transmission input signal z [n] from which the offset is removed as input (step S1005).

Then, the delayed reception input signal x [n−D] and the transmission input signal z with the offset removed.
[N] and the residual signal e [n], which is the signal after error canceller processing output from the echo canceller (EC) 110, are input, and the echo reduction unit (ER) 111 is an echo that is nonlinear echo suppression processing. Reduction processing is performed (step S1006). Then, the processing from step S1002 to step S1006 is performed until the call is finished (step S1007).

The processing of the echo canceller (EC) 110 shown in FIG. 6 is performed as follows. First,
The double talk detection unit (DTD) 110c performs a double talk detection process (step S11).
01). Next, the adaptive filter unit (ADF) 110a double-talk information ECstate [
n], adaptive filter processing is performed (step S1102). Then, the signal subtraction processing unit 110b generates an adaptive filter unit (ADF) 110 from the transmission input signal z [n].
The pseudo echo signal y ′ [n] output from “a” is subtracted to calculate and output the residual signal e [n], and the echo canceller processing ends.

The processing of the echo reduction unit (ER) 111 shown in FIG. 7 is performed as follows. First, the frequency domain transform processing unit (FT) 111a, the frequency domain transform processing unit (FT) 111b, and the frequency domain transform processing unit (FT) 111c are respectively a reception frame and a transmission frame that are buffers for conversion to the frequency domain. , Update the residual frame (step S1201r,
S1201s, S1201e).

Next, the frequency domain transform processing unit (FT) 111a converts the delayed received input signal x [n-D] into the frequency domain and calculates the frequency spectrum X [f, ω] of the received input signal (step). S1202r), the received power calculation unit (POW) 111d receives the received power spectrum | X
[F, ω] | ² and the smoothed received power spectrum | X _S [f, ω] | ² are calculated (step S1203r).

Similarly, the frequency domain transform processing unit (FT) 111b converts the transmission input signal z [n] into the frequency domain and calculates the frequency spectrum Z [f, ω] of the transmission input signal (step S12).
02s), the transmission power calculation unit (POW) 111e transmits the transmission power spectrum | Z [f, ω].
| ² and the smoothed transmission power spectrum | Z _S [f, ω] | ² are calculated (step S1203s).

Similarly, the frequency domain transform processing unit (FT) 111c transforms the residual signal e [n] into the frequency domain and calculates the frequency spectrum E [f, ω] of the residual signal (step S120).
2e), the residual power calculation unit (POW) 111f receives the residual power spectrum | E [f, ω] |
² and the smoothed residual power spectrum | E _S [f, ω] | ² are calculated (step S1203e).

The frequency domain double talk detector (FDTD) 111j outputs frequency domain double talk information ERstate [f, ω], and the acoustic coupling amount estimator (ACLE) 111g
Smoothed reception power spectrum | X _S [f, ω] | ² and smoothed transmission power spectrum | Z _S [f, ω] | ² and frequency domain double talk information ERstat
The acoustic coupling amount | H _S [f, ω] | ² is calculated using e [f, ω] as an input (step S 1
204). Echo amount estimating unit (ELE) 111h is acoustic coupling amount | contained in ² and transmission input signal as an input ^{_{| H s [f, ω]}} | 2 and smoothing the received power spectrum _{| X} S [f, _ω] The echo amount | Y _S [f, ω] | ² is estimated (step S1205).

Next, the echo suppression amount estimation unit (ECLE) 111i transmits the transmission power spectrum | Z [f,
ω] | ² and residual power spectrum | E [f, ω] | ² are input, and the echo suppression amount ECL [f, ω] suppressed by the echo canceller unit (EC) 110 is estimated (step S1).
206). Then, the control unit (CTRL) 111k determines whether or not the state is a double talk state and whether or not the echo suppression amount of the echo canceller unit (EC) 110 is sufficient, and the control information ERcontrol1 [f, ω] and ERcontrol2 [f, ω] are output (step S1207).

The echo suppression gain calculation unit (GCAL) 111n is a transmission power calculation unit (POW) 111.
The smoothed transmission power spectrum | Z _S [f, ω] | ² output from e and the smoothed echo amount output from the echo amount estimation unit (ELE) 111 h | Y _S [f,
ω] | ² and the parameter γ [ω] output from the gain storage unit (GTBL) 111m;
Echo suppression amount ECL [f, ω] output from the echo suppression amount estimation unit (ECLE) 111i
And control information ERcontrol2 [f, ω output from the control unit (CTRL) 111k
] In the double talk state and the frequency band detected when the amount of echo suppression is not sufficient, and the other frequency band, the echo suppression gain G [
f, ω] is calculated. Further, the echo suppression gain calculation unit (GCAL) 111n controls the echo suppression gain G [f, ω] so as not to be equal to or lower than a predetermined lower limit value (step S1208).
).

On the other hand, the spectrum selection unit 111L receives the control information ERcontrol1 [f, ω] output from the control unit (CTRL) 111k as an input, and in the frequency band detected that the echo suppression amount is not sufficient in the double talk state, the frequency spectrum Is selected as the frequency spectrum Z [f, ω] of the transmission input signal, and the frequency spectrum E [f, ω] of the residual signal is selected as the frequency spectrum in other frequency bands (step S1209).

Then, the signal suppression unit (SS) 111o selects the frequency spectrum selected by the spectrum selection unit 111L and the echo suppression gain G [f, ω] calculated by the echo suppression gain calculation unit (GCAL) 111n as inputs. The echo of the frequency spectrum is suppressed (step S1210). Finally, the frequency domain inverse transform processing unit (IFT) 111p performs frequency inverse transform processing on the frequency spectrum S ′ [f, ω] output from the signal suppression unit (SS) 111o (step S1211), thereby performing echo. The reduction process ends.

The transmission output signal s ′ output from the echo reduction unit (ER) 111 in this way.
[N] is encoded in each frame by an encoder (not shown) in the communication unit (COM) 101, and data obtained by this encoding is transmitted to the communication partner as transmission data through the communication unit (COM) 101. Is done.

In the above description, the echo reduction unit (ER) 111 is described as operating as a method of processing for each frequency band in the frequency domain type by FFT. You may implement | achieve the frequency domain type | molds, such as the method which puts together the frequency band by FFT and processes for every frequency band group, and band division filters, such as a filter bank.

When echo path loss fluctuates during double talk, the echo canceller (EC) 11
Due to the processing of 0, the residual signal e [n] becomes larger than the transmission input signal z [n],
In some cases, the amount of echo suppression by the processing of the echo canceller unit (EC) 110 is not sufficient.

In this case, the echo canceller (EC) 11 is operated by the operation of the signal processing apparatus described above.
Echo suppression amount estimation unit (ECLE) in a frequency region where the echo suppression amount due to processing of 0 is not sufficient
111i and the control unit (CTRL) 111k determine, and the spectrum selection unit 111L and the echo suppression gain calculation unit (GCAL) 111n perform echo canceller units (E
C) Since it is possible to control to change the weight of the processing of 110 and the processing of the echo reduction unit (ER) 111, it is possible to make robust against the echo path loss fluctuation, and high quality transmission output It is possible to output a signal.

(Modification of the first embodiment)
The signal processing device according to the modification of the first embodiment of the present invention is different from the signal processing device according to the first embodiment in that an echo reduction unit (ER) 1112 is used instead of the echo reduction unit (ER) 111. It is a point which has. FIG. 8 is a block diagram showing the configuration of the echo reduction unit (ER) 1112 of the signal processing device according to the modification of the first embodiment of the present invention.

The echo reduction unit (ER) 1112 is different from the echo reduction unit (ER) 111 in that an acoustic coupling amount estimation unit (ACLE) is used instead of the acoustic coupling amount estimation unit (ACLE) 111g.
111g2 and an echo suppression gain calculation unit (GCAL) 111n2 instead of the echo suppression gain calculation unit (GCAL) 111n.

Further, in the echo reduction unit (ER) 1112, the transmission power calculation unit (POW)
The output of 111e is input only to the echo suppression amount estimation unit (ECLE) 111i, and the output of the residual power calculation unit (POW) 111f is input to the echo suppression amount estimation unit (ECLE) 111i and the frequency domain double talk detection unit (FDTD) 111j. In addition to the acoustic coupling amount estimation unit (ACLE) 1
11g2 and the echo suppression gain calculation unit (GCAL) 111n2 are input, and the other parts are the same. Therefore, in the echo reduction unit (ER) 1112, the same parts as those of the echo reduction unit (ER) 111 are denoted by the same reference numerals and description thereof is omitted.

The acoustic coupling amount estimation unit (ACLE) 111g2 includes the smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d, and the residual power calculation unit (POW) 111f. And the smoothed residual power spectrum | E _S [f, ω] | ² output from the frequency domain double talk information ERstate [f, ω] output from the frequency domain double talk detector (FDTD) 111j. As an input, for each frequency band ω, the acoustic coupling amount | H [f, ω] | ² is calculated by the following expression 14 using | E _S [f, ω] | ² .

Then, the acoustic coupling amount estimation unit (ACLE) 111g2 calculates and outputs a smoothed acoustic coupling amount | H _S [f, ω] | ² using a value one frame before as in the following Expression 15. However, α _H [ω] is preferably about 0.03 to 0.99.

The echo suppression gain calculation unit (GCAL) 111n2 is a residual power calculation unit (POW) 11
The smoothed residual power spectrum | E _S [f, ω] | ² output from If
Smoothed echo amount output from the echo amount estimation unit (ELE) 111h | Y _S [f
, Ω] | ² , the parameter γ [ω] output from the gain storage unit (GTBL) 111m, and the echo suppression amount ECL [f, ω output from the echo suppression amount estimation unit (ECLE) 111i.
] And control information ERcontrol2 [f, ω] which is information indicating whether the frequency range is a double talk state output from the control unit (CTRL) 111k and the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient. Are input and the echo suppression gain G [f, ω] is calculated and output.

Specifically, the echo suppression gain calculation unit (GCAL) 111n2 is a control unit (CTRL).
For the frequency band detected from the control information ERcontrol2 [f, ω] output from 111k in the double talk state and the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient, the echo of the echo canceller unit (EC) 110 is echoed. In order to use the suppression amount obtained by increasing the suppression amount as the echo suppression amount of the echo reduction (ER) 1112, the echo suppression gain G
[F, ω] is calculated by using the echo suppression amount ECL [f, ω] of the echo canceller unit (EC) 110 to correct the suppression amount, as shown in Equation 16 below.

In other frequency bands, the echo suppression gain calculation unit (GCAL) 111n2 uses the Wiener Filter method to express the echo suppression gain G [f, ω] using Equation 17
Calculated by -1. Alternatively, assuming that the echo canceller (EC) 110 is functioning normally, the echo suppression amount of the echo reduction (ER) 111 is increased.
Correction is performed using the echo suppression amount ECL [f, ω] of the echo canceller (EC) 110, and the echo suppression gain G [f, ω] is calculated as shown in the following Expression 17-2.

The echo canceller unit is used in applications in which the echo path fluctuation hardly occurs and the suppression performance of the echo canceller unit (EC) 110 is often stable due to the operation of the signal processing device according to the modification of the first embodiment described above. (EC) 110 and echo reduction (ER) 111
Since it becomes easy to select the serial connection of 2, it is possible to prevent excessive echo suppression and to prevent the transmission sound quality from deteriorating.

(Second Embodiment)
The signal processing apparatus according to the second embodiment is different from the signal processing apparatus according to the first embodiment in that it does not have the echo reduction unit (ER) 111 as shown in FIG. 9, but the noise reduction unit (NR). The other parts are the same. Therefore, the same parts are denoted by the same reference numerals and description thereof is omitted, and the noise reduction unit (NR) 211 according to the second embodiment will be described with reference to the drawings. This noise reduction part (NR)
In FIG. 21, the same parts as those of the echo reduction unit (ER) 111 according to the first embodiment are denoted by the same reference numerals and description thereof is omitted.

FIG. 10 shows a noise reduction unit (NR) 211 of the signal processing apparatus according to the second embodiment.
It is a block diagram which shows the structure of these. The noise reduction unit (NR) 211 includes a frequency domain conversion processing unit (FT) 111b, a frequency domain conversion processing unit (FT) 111c, a transmission power calculation unit (POW) 111e, and a residual power calculation unit (POW). ) 111f, echo suppression amount estimation unit (ECLE) 111i, control unit (CTRL) 211k, and spectrum selection unit 21
1L, a noise suppression gain calculation unit (GCAL) 211n, and a signal suppression unit (SS) 211o
A frequency domain inverse transform processing unit (IFT) 111p, and a noise level estimation unit (NLE) 211.
q.

The noise reduction unit (NR) 211 receives the transmission input signal z [n] output from the high-pass filter unit (HPF) 109 and the residual signal e [
n] as an input, a noise component is suppressed from at least one of the transmission input signal z [n] or the residual signal e [n], and the signal after the noise suppression is transmitted as a transmission output signal s ′ [n] ( n = 0, 1
,..., N-1) are output every frame.

The noise level estimator (NLE) 211q is a transmission spectrum Z [f, ω] output from the frequency domain transform processor (FT) 111b and a residual spectrum output from the frequency domain transform processor (FT) 111c. E [f, ω] as an input, a spectrum in a non-sound period is measured, and a noise level | N _Z [f included in the transmission spectrum Z [f, ω] for each frequency band ω.
, Ω] | ² and the noise level | N _E [f, ω] | ² included in the residual spectrum E [f, ω] are calculated and output.

The control unit (CTRL) 211k receives the echo suppression amount ECL [f, ω] output from the echo suppression amount estimation unit (ECLE) 111i as an input, and for each frequency band ω, an echo canceller unit (
EC) The frequency band where the echo suppression amount of 110 is not sufficient is detected, and the echo canceller (
EC) Outputs control information NRcontrol [f, ω], which is information indicating whether or not the echo suppression amount of 110 is an insufficient frequency band.

Specifically, the control unit (CTRL) 211k performs the inequality ECL [f, ω for each frequency band ω.
]> Β _Z [ω], and the inequality ECL [f,
When the number of bands count in which ω]> β _Z [ω] is established is larger than β _C, it is determined that the echo suppression amount is small. However, β _Z [ω] is preferably about +8 dB to −15 dB, and β _C is preferably about 10% to 40% of the total number of frequency bands. In the present embodiment, hereinafter, β _Z [ω] will be described as 0 dB.

The spectrum selection unit 211L includes the frequency spectrum Z [f, ω] of the transmission input signal output from the frequency domain transform processing unit (FT) 111b and the frequency domain transform processing unit (FT) 111.
The frequency spectrum E [f, ω] of the residual signal output from c and the control unit (CTRL) 21
The control information NRcontrol [f, ω], which is information indicating whether or not the echo suppression amount of the echo canceller (EC) 110 output from 1k is a frequency band that is not sufficient, is input,
Frequency spectrum Z [f, ω] of the transmission input signal or frequency spectrum E [
f, ω] is selected as a frequency spectrum and output.

Specifically, the control information NRcontrol [f, ω] is the echo canceller (EC) 1
If the frequency band detected that the echo suppression amount of 10 is not sufficient, the frequency spectrum Z [f, ω] of the transmission input signal is selected as the frequency spectrum. If the frequency band is other than that, the frequency spectrum E [f, ω] of the residual signal is selected as the frequency spectrum.

In this way, when the echo suppression amount of the echo canceller (EC) 110 is not sufficient for each frequency band, the noise reduction unit (NR) 211 is operated alone, and noise level estimation affects the influence of echo. It can be made not to receive.

The noise suppression gain calculation unit (GCAL) 211n is a noise level estimation unit (NLE) 211.
The noise level | N _Z [f, ω] | ² and | N _E [f, ω] | ² output from q and the transmission power spectrum | Z [output from the transmission power calculation unit (POW) 111e f, ω] | ²
And control information NRcontrol [f, ω] output from the control unit (CTRL) 211k.
And the spectrum S ′ [f−1, ω] of the transmission output signal one frame before output from the signal suppression unit (SS) 111o as an input, and calculate and output a noise suppression gain G [f, ω]. To do.

Specifically, the noise suppression gain calculation unit (GCAL) 211n first has a control unit (CTR)
L) For the frequency band detected by the control information NRcontrol [f, ω] output from 211k that the echo suppression amount is not sufficient, the echo canceller (EC) 110
It is determined that the accuracy of noise estimation has not been obtained due to the influence of the residual echo after the above processing, and | N _Z [f, ω] | ² is selected as the noise level. In other frequency bands, the noise level is | N
_E [f, ω] | ² is selected.

Then, based on the noise level selected for each frequency band ω, the noise suppression gain G [f, ω] is calculated by the following algorithm or a combination thereof. That is, the spectral subtraction method (SF) that is a general noise reduction.
Boll, “Suppression of acoustic noise in speech using spectral subtraction”, IE
EE Trans.Acoustics, Speech, and Signal Processing, vol.ASSP-29, pp.113-120 (197
9).), Wiener Filter method (JS Lim, AV Oppenheim, “En
hancement and bandwidth compression of noisy speech ”, Proc. IEEE Vol.67, No.12,
pp.1586-1604, Dec.1979.) and Maximum Likelihood method (RJ McAulay, M
L. Malpass, “Speech enhancement using a soft-decision noise suppression filte
r ”, IEEE Trans. on Acoustics, Speech, and Signal Processing, vol.ASSP-28, no.2,
pp.137-145, Apr.1980.).

Further, the pre-SNR SNR _PRIO [f, ω] and the post-SNR SNR _POST [f, ω] are estimated and calculated using the frequency spectrum S ′ [f−1, ω] one frame before the transmission output signal. To accurately estimate the noise suppression gain (eg, P. Scalart, JV Filho, “S
peech enhancement based on a priori signal to noise estimation ”, Proc. ICASSP96
, pp.629-632, May 1996), MMSE-STSA (Minimum Mean-Square Error Short-T
ime Spectral Amplitude estimator method (Y. Ephraim, D. Malah, “Speech enhancement
using a minimum mean-square error short-time spectral amplitude estimator ”, IE
EE Trans.on Acoustics, Speech, and Signal Processing, vol.ASSP-32, no.6, pp.110
9-1121, Dec. 1984.) and Joint MAP method (T. Lotter, P. Vary, “Noise reduction by max
imum a posteriori spectral amplitude estimation with super Gaussian speech model
ing ”, Proc. IWAENC, pp. 83-86, Sep. 2003.).

For example, when the noise level | N _Z [f, ω] | ² is selected, P [•] is set as a half-wave rectification, and the prior SN ratio SNR _PRIO [f, ω] and the subsequent SN ratio SNR _POST [f, ω] is obtained by the following equations 18-1 and 18-2. In this case, when the Wiener filter method is used, the noise suppression gain G [f, ω] is calculated by the following equation 18-3. The However, μ [ω] is a forgetting factor of about 0.9 to 0.999. On the other hand, noise level | N _E
When [f, ω] | ² is selected, | N _Z [f, ω] | ² in the following equations 18-1 and 18-2 is similarly changed to | N _E [f, ω] | ² Is replaced with the noise suppression gain G [f, ω
] Is calculated by the following equation 18-3.

In addition, the noise suppression gain calculation unit (GCAL) 211n prevents the transmission sound quality from being deteriorated due to excessive noise suppression, and prevents noise suppression gain G [f, ω] in order to prevent intermittent suppression of background noise. Is controlled so as not to fall below a predetermined lower limit. That is, the setting for each band of the lower limit value of the echo suppression gain described with reference to FIG. 4 is also applied to the setting for each band of the lower limit value of the noise suppression gain.

Further, in order to prevent the transmission sound quality from deteriorating due to excessive noise suppression, a frequency band is used by using a signal level of a section that is not a voiced section of the transmission input signal z [n] or the residual signal e [n]. The background noise level may be calculated for each, and the noise suppression gain may be controlled so as not to suppress the background noise level.

The signal suppression unit (SS) 211o receives the frequency spectrum output from the spectrum selection unit 211L and the noise suppression gain G [f, ω] output from the noise suppression gain calculation unit (GCAL) 211n as the spectrum selection. The noise of the frequency spectrum output from the unit 211L is suppressed and output as the spectrum S ′ [f, ω] of the transmission output signal as shown in the following Expression 19-1 or 19-2.

At this time, regardless of which frequency spectrum Z [f, ω] of the transmission input signal or frequency spectrum E [f, ω] of the residual signal is selected by the spectrum selection unit 211L, the echo canceller unit (EC ) 110 is a process in the time domain, so there is no difference in the phase spectrum, so that the output transmission output signal can be smoothly connected in terms of hearing.

Next, a processing flow of the signal processing device according to the second embodiment configured as described above will be described. FIG. 11 is a flowchart showing an overall processing flow of the signal processing apparatus according to the second embodiment. Note that the same operation steps as those of the signal processing apparatus according to the first embodiment described with reference to FIG. 5 are denoted by the same reference numerals, and description thereof is omitted.

After the echo canceller processing in step S1005, the signal processing apparatus replaces the echo reduction processing in step S1006 in the first embodiment with the echo obtained in step S1005 in the second embodiment. Noise reduction processing is performed using the signal after the canceller processing (step S2006). Then, the process proceeds to step S1007 to determine whether or not the call is an end story.

Since the echo cancellation unit (EC) 110 according to the second embodiment is the same as the echo cancellation unit (EC) 110 according to the first embodiment, the flow of processing in each embodiment is naturally the same. There is no explanation.

FIG. 12 is a flowchart showing the flow of processing in the noise reduction unit (NR) 211 according to the second embodiment. Note that the same operation steps as the operations in the echo reduction unit (ER) 111 according to the first embodiment described with reference to FIG. 7 are denoted by the same reference numerals, and description thereof is omitted.

The noise reduction unit (NR) 211 according to the second embodiment performs step S1201r.
Received frame update processing, received frequency conversion processing of step S1202r, step S12
There is no operation from the reception power spectrum calculation process of 03r, the acoustic coupling amount calculation process of step S1204, the echo level calculation process of step S1205, the determination process of step S1207 to the transmission signal suppression process of step S1210.

That is, the noise reduction unit (NR) 211 according to the second embodiment starts the noise reduction process, and after the transmission power spectrum calculation process in step S1203s and the residual power spectrum calculation process in step S1203e, Level estimation unit (NLE) 2
11q calculates the noise level included in the frequency spectrum of the transmission input signal and the noise level included in the frequency spectrum of the residual signal for each frequency band ω by measuring the spectrum of the section that is not a sound section. (Step S2205). Then, step S1
An echo suppression amount calculation process 206 is performed, and the control unit (CTRL) 211k performs a determination process as to whether or not the echo suppression amount of the echo canceller unit (EC) 110 is sufficient, and performs control information NR.
Control [f, ω] is output (step S2207).

The noise suppression gain calculation unit (GCAL) 211n then transmits the transmission power calculation unit (POW).
) 111e output power spectrum | Z [f, ω] | ² output from 111e and the control unit (CTR)
L) The control information NRcontrol [f, ω] output from 211k is used as an input, and the noise suppression gain G [ f, ω] is calculated. Further, the noise suppression gain calculation unit (GCAL) 211n controls the noise suppression gain G [f, ω] so as not to be equal to or lower than a predetermined lower limit value (step S2208).

Then, the spectrum selection unit 211L receives the control information NRcontrol [f, ω] output from the control unit (CTRL) 211k as an input, and transmits the transmission frequency as a frequency spectrum for the frequency band detected that the echo suppression amount is not sufficient. Signal frequency spectrum Z [
f, ω] is selected, and in the other frequency bands, the frequency spectrum E [f, ω] of the residual signal is selected as the frequency spectrum (step S2209).

The signal suppression unit (SS) 211o is selected with the frequency spectrum selected by the spectrum selection unit 211L and the noise suppression gain G [f, ω] calculated by the noise suppression gain calculation unit (GCAL) 211n as inputs. The frequency spectrum noise is suppressed (step S2210). Then, the process proceeds to the frequency inverse transform process in step S1211, and the noise reduction process ends.

As described above, the noise reduction unit (NR) 211 has been described as operating as a method of processing for each frequency band in the frequency domain type by FFT. You may implement | achieve the frequency domain type | molds, such as the method which puts together the frequency band by FFT and processes for every frequency band group, and band division filters, such as a filter bank.

At the time of echo path loss fluctuation during double talk, the residual signal e [n] becomes larger than the transmission input signal z [n] by the processing of the echo canceller unit (EC) 110, and the echo canceller unit (EC) 110. In some cases, the amount of echo suppression by this processing is not sufficient.

By the operation of the signal processing apparatus according to the second embodiment described above, an echo canceller (
EC) The frequency region where the echo suppression amount by the processing of 110 is not sufficient is selected as an echo suppression amount estimation unit (
ECLE) 111i and control unit (CTRL) 211k, and for each frequency band, noise level estimation unit (NLE) 211q, spectrum selection unit 211L, noise suppression gain calculation unit (GCAL) 211n of noise reduction unit (NR) 211 This makes it possible to prevent the noise level estimation from being affected by echoes, so that it is possible to make robust against echo path loss fluctuations and output high-quality signals.

(Third embodiment)
FIG. 13 is a block diagram illustrating a configuration of a signal processing device according to the third embodiment. This signal processing device is different from the signal processing device according to the first embodiment in that an echo reduction unit (
ER) 111 and an echo noise reduction unit (ENR) 311, and the other parts are the same. Therefore, the same parts are denoted by the same reference numerals and description thereof is omitted, and an echo noise reduction unit (ENR) according to the third embodiment with reference to the drawings.
311 will be described. In this echo noise reduction unit (ENR) 311, the same parts as those of the echo reduction unit (ER) 111 according to the first embodiment and the noise reduction unit (NR) 211 according to the second embodiment are denoted by the same reference numerals. The description is omitted.

FIG. 14 illustrates an echo noise reduction unit (ENR) of the signal processing device according to the third embodiment.
3) is a block diagram showing the configuration of 311. This echo noise reduction unit (ENR) 3
11 is a frequency domain transformation processing unit (FT) 111a and a frequency domain transformation processing unit (FT) 11
1b, a frequency domain conversion processing unit (FT) 111c, and a received power calculation unit (POW) 111
d, transmission power calculation unit (POW) 111e, residual power calculation unit (POW) 111f, acoustic coupling amount estimation unit (ACLE) 111g, echo amount estimation unit (ELE) 111h, echo suppression amount estimation unit (ECLE) 111i, control unit (CTRL) 211k, spectrum selection unit 211L, gain storage unit (GTBL) 111m, echo noise suppression gain calculation unit (GCAL) 311n, signal suppression unit (SS) 311o, Frequency domain inverse transform processing unit (I
FT) 111p and a noise level estimation unit (NLE) 311q.

An echo noise reduction unit (ENR) 311 is a high-pass filter unit (HPF) 10.
9 is input with the transmission input signal z [n] output from the signal 9 and the residual signal e [n] output from the signal subtraction processing unit 110b, and the transmission input signal z [n] or the residual signal e [ n] suppresses an echo component and a noise component from at least one of them, and a signal after the echo suppression and noise suppression is set as a transmission output signal s ′ [n] (n = 0, 1,..., N−1). Output every frame.

The noise level estimation unit (NLE) 311q receives the frequency spectrum (transmission spectrum Z [f, ω] or residual spectrum E [f, ω]) output from the spectrum selection unit 211L, and is not a voiced section And a noise level | N [f, ω] | ² included in the frequency spectrum is calculated and output for each frequency band ω.

The echo noise suppression gain calculation unit (GCAL) 311n is an echo amount estimation unit (ELE) 1
The smoothed echo amount | Y _S [f, ω] | ² output from 11h, the parameter γ [ω] output from the gain storage unit (GTBL) 111m, and the transmission power calculation unit (POW)
) Smoothed transmission power spectrum output from 111e | Z _S [f, ω] |
² and the noise level | N [f, ω] | output from the noise level estimator (NLE) 311q
² and control information NRcontrol [f, ω output from the control unit (CTRL) 211k
] Is input and the echo noise suppression gain G [f, ω] is calculated and output.

Specifically, the echo noise suppression gain calculation unit (GCAL) 311n first has a control unit (
(CTRL) 211k, the frequency band detected by the control information NRcontrol [f, ω] output from the control information NRcontrol is not smoothed in order to prevent the transmission sound quality from being deteriorated due to excessive echo suppression. Transmission power spectrum | Z _S [f
, Ω] | ² removes the noise level | N [f, ω] | ² from the echo suppression gain G _ER [f
, Ω] is calculated by Equation 20.

In other frequency bands, the echo cancellation amount of the echo canceller unit (EC) 110 is sufficient, and it is assumed that the echo canceller unit (EC) 110 is functioning normally, and an echo noise suppression gain calculation unit (GCAL) 311n is an echo noise reduction unit (ENR)
The echo suppression amount ECL [f, ω] of the echo canceller unit (EC) 110 is corrected so as to increase the echo suppression amount 311, and the echo suppression gain G _ER [f, ω] is expressed by the following equation 2
1 is calculated.

Also, the echo noise suppression gain calculation unit (GCAL) 311n, based on the noise level | N [f, ω] | ² selected for each frequency band ω, spectral subtraction (Spectral), which is the general noise reduction described above. The noise suppression gain G _NR [f, ω] is calculated by an algorithm such as a subtraction method, a Wiener Filter method, or a maximum likelihood estimation method.

Then, as shown in Equation 22, the product of the echo suppression gain G _ER [f, ω] and the noise suppression gain G _NR [f, ω] is calculated and set as the echo noise suppression gain G [f, ω].

In order to prevent the transmission sound quality from being deteriorated due to excessive echo suppression, the echo noise suppression gain calculation unit (GCAL) 311n prevents the echo noise suppression gain G [f, ω] from being below a predetermined lower limit value. To control. That is, the setting for each band of the lower limit value of the echo suppression gain described with reference to FIG. 4 is also applied to the setting for each band of the lower limit value of the echo noise suppression gain.

Further, in order to prevent the transmission sound quality from deteriorating due to excessive echo and noise suppression, the echo noise suppression gain calculation unit (GCAL) 311n performs the transmission input signal z [n] or the residual signal e [n]. A background noise level may be calculated for each frequency band using a signal level of a section that is not a sound section, and the echo noise suppression gain may be controlled so as not to suppress the background noise level.

Furthermore, the echo noise suppression gain calculation unit (GCAL) 311n sets the echo noise suppression gain G [f, ω] to the following Expression 23-1 or Expression 23-2 in order to prevent the transmission sound quality from deteriorating. As shown in FIG. 5, the output may be smoothed in the frequency direction. For example, ε
_j is [0.1, 0.2, 0.4, 0.2, 0.1], and η _j is [0.1, 0.2, 0.
4, 0.8, 0.4, 0.2, 0.1].

The signal suppression unit (SS) 311o receives the frequency spectrum output from the spectrum selection unit 211L and the echo noise suppression gain G [f, ω] output from the echo noise suppression gain calculation unit (GCAL) 311n as inputs. The echo and noise of the frequency spectrum output from the spectrum selection unit 211L are suppressed, and the spectrum S ′ [f, ω] of the transmission output signal is calculated and output as in the following Expression 24-1 or Expression 24-2. To do.

At this time, regardless of which frequency spectrum Z [f, ω] of the transmission input signal or frequency spectrum E [f, ω] of the residual signal is selected by the spectrum selection unit 211L, the echo canceller unit (EC ) 110 is a process in the time domain, so there is no difference in the phase spectrum, so that the output transmission output signal can be smoothly connected in terms of hearing.

Next, a processing flow of the signal processing device according to the third embodiment configured as described above will be described. FIG. 15 is a flowchart illustrating an overall processing flow of the signal processing apparatus according to the third embodiment. Note that the same operation steps as those of the signal processing apparatus according to the first embodiment described with reference to FIG. 5 are denoted by the same reference numerals, and description thereof is omitted.

After the echo canceller processing in step S1005, the signal processing apparatus replaces the echo reduction processing in step S1006 in the first embodiment with the echo obtained in step S1005 in the third embodiment. An echo noise reduction process is performed using the signal after the canceller process (step S3006). And step S10
The process proceeds to the judgment of whether or not the end talk of 07.

Since the echo cancellation unit (EC) 110 according to the third embodiment is the same as the echo cancellation unit (EC) 110 according to the first embodiment, the flow of processing in each embodiment is naturally the same. There is no explanation.

FIG. 16 is a flowchart showing the flow of processing in the echo noise reduction unit (ENR) 311 according to the third embodiment. The operation of the echo reduction unit (ER) 111 according to the first embodiment described with reference to FIG. 7 and the operation of the noise reduction unit (NR) 211 according to the second embodiment described with reference to FIG. About the same operation step, the same code | symbol is attached | subjected and description of the part is abbreviate | omitted.

The echo noise reduction unit (ENR) 311 according to the third embodiment starts an echo noise reduction process, and steps S1201r to S1203r, S1201s to S
Frame update processing, frequency conversion processing, power spectrum calculation processing, acoustic coupling amount calculation processing in step S1204, echo level calculation processing in step S1205, and step S1206 in steps 1203s and S1201e to S1203e. Echo suppression amount calculation processing is performed.

Then, after the determination processing in step S2207 and the spectrum selection processing in step S2209, the noise level | N [f, ω] output from the noise level estimation unit (NLE) 311q.
| ² is calculated (step S3205).

Then, the echo noise suppression gain calculation unit (GCAL) 311n has a noise level estimation unit (
NLE) The noise level | N [f, ω] | ² output from 311q and the control unit (CTRL)
The control information NRcontrol [f, ω] output from the 211k is used as an input, and the echo noise suppression gain G [] is calculated by different calculations for the frequency band detected when the echo suppression amount is not sufficient and the other frequency bands. f, ω] is calculated. Further, the echo noise suppression gain calculation unit (GCAL) 311n controls the echo noise suppression gain G [f, ω] so as not to be equal to or lower than a predetermined lower limit value (step S3208).

Then, the signal suppression unit (SS) 311o receives the frequency spectrum selected by the spectrum selection unit 211L and the echo noise suppression gain G [f, ω] calculated by the echo noise suppression gain calculation unit (GCAL) 311n as inputs. The echo and noise of the selected frequency spectrum are suppressed (step S3210). Then, the process proceeds to the frequency inverse transform process in step S1211, and the echo noise reduction process ends.

As described above, the echo noise reduction unit (ENR) 311 has been described as operating in a frequency domain type by FFT and processing for each frequency band. You may implement | achieve the frequency domain type | molds, such as the method which puts together the frequency band by FFT and processes for every frequency band group, and band division filters, such as a filter bank.

At the time of echo path loss fluctuation during double talk, the residual signal e [n] becomes larger than the transmission input signal z [n] by the processing of the echo canceller unit (EC) 110, and the echo canceller unit (EC) 110. In some cases, the amount of echo suppression by this processing is not sufficient.

By the operation of the signal processing apparatus according to the third embodiment described above, the echo canceller unit (
EC) The frequency region where the echo suppression amount by the processing of 110 is not sufficient is selected as an echo suppression amount estimation unit (
ECLE) 111i and control unit (CTRL) 211k make determination, and spectrum selection unit 211
Since L and the echo noise suppression gain calculation unit (GCAL) 311n can be controlled to change the weights of the processing of the echo canceller unit (EC) 110 and the processing of the echo noise reduction unit (ENR) 311 for each frequency band. It is possible to make robust against echo path loss fluctuations and to output a high-quality signal.

In addition, echo reduction in the frequency domain, which is nonlinear suppression processing, and noise reduction in the frequency domain are performed by an echo noise suppression gain calculation unit (GCAL) 311n and a signal suppression unit (SS).
) By using 311o as the common echo noise reduction unit (ENR) 311, the amount of calculation can be reduced, and deterioration of the transmission sound quality can be prevented as compared with the case where echo reduction and noise reduction are connected in series. .

(Fourth embodiment)
FIG. 17 is a block diagram illustrating a configuration of a signal processing device according to the fourth embodiment. This signal processing device is different from the signal processing device according to the first embodiment in that it does not have the echo reduction unit (ER) 111 and has the echo suppressor unit (ES) 411 as shown in FIG. A point having a delay processing unit (DELAY) 402 instead of the delay processing unit (DELAY) 102 and a point having a D / A converter (D / A) 403 instead of the D / A converter (D / A) 103 And other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

Further, the delay processing unit (DELAY) 402 and the D / A converter (D / A) 403 may receive the reception output signal x ′ [n] output from the echo suppressor unit (ES) 411. Although different, the delay processing unit (DELAY) 102 and the D / A converter (D / A)
Since the same operation as that of 103 is performed, the description thereof is omitted, and an echo suppressor unit (ES) 411 according to the fourth embodiment is described with reference to the drawings.

FIG. 18 is a block diagram illustrating a configuration of an echo suppressor unit (ES) 411 of the signal processing device according to the fourth embodiment. The echo suppressor (ES) 411 includes a reception power calculation unit (POW) 411d, a transmission power calculation unit (POW) 411e, a residual power calculation unit (POW) 411f, and an echo suppression amount estimation unit (ECLE). 411i and the control unit (CTRL
) 411k, a signal selection unit 411L, a gain storage unit (GTBL) 411m, an echo suppression gain calculation unit (GCAL) 411n, a transmission signal suppression unit (SS) 411o, and a reception signal suppression unit (SS) 411r, Consists of.

The echo suppressor unit (ES) 411 includes a reception input signal x [n] output from the communication unit (COM) 101 and a transmission input signal z output from the high-pass filter unit (HPF) 109.
[N] and the residual signal e [n] output from the signal subtraction processing unit 110b are input, and the received output signal x ′ [n] and the transmitted output signal s ′ [n] (n = 0, 1). ,..., N-1) are output for each frame. Here, at least one of the reception input signal x [n] or the transmission signal (transmission input signal z [n] or residual signal e [n]) is subjected to suppression processing, and the signal after the echo suppression is output. .

The reception power calculation unit (POW) 411d receives the reception input signal x [n] output from the communication unit (COM) 101, and receives the reception power P _X [f] as shown in the following expression 25-1. Calculate and output in units. As the reception power P _X [f], a value P _SX [f] smoothed by using a value P _SX [f−1] one frame before may be used as shown in the following Expression 25-2. However, α _X is preferably about 0.375 to 0.999.

The transmission power calculation unit (POW) 411e receives the transmission input signal z [n] output from the high-pass filter unit (HPF) 109, and the transmission power P _Z [f] is expressed by the following equation 26-1. As shown, it is calculated and output in units of frames. The transmission power P _Z [f] is expressed by the following equation 26-2.
The value _P SZ [f was smoothed using one previous frame values _P SZ [f-1] as shown in the
] May be used. However, α _Z is, is preferably about 0.375 to 0.999.

The residual power calculation unit (POW) 411f receives the residual signal e [n] output from the signal subtraction processing unit 110b, and the residual power P _E [f] is expressed by the following equation 27-1. Calculate and output in units of frames. Residual power _P E [f] may be used a value _P SE [f] were smoothed using the following one frame as shown by equation 27-2 previous value _P SE [f-1]. However, α _E is preferably about 0.375 to 0.999.

The echo suppression amount estimation unit (ECLE) 411i transmits the transmission power P _Z [f] output from the transmission power calculation unit (POW) 411e and the residual power output from the residual power calculation unit (POW) 411f. P _E [f] is input, and the echo suppression amount ECL [f] suppressed by the echo canceller unit (EC) 110 is estimated and output. Specifically, it is calculated by the following equation 28. Of course, the difference between these two powers may be used.

The control unit (CTRL) 411k receives ECL [f] output from the echo suppression amount estimation unit (ECLE) 411i as an input, and determines whether the echo suppression amount of the echo canceller unit (EC) 110 is insufficient. Control information EScontr which is the determination result information
ol [f] is output. Specifically, the frame filled with ECL [f]> β _Z determines that the echo suppression amount of the echo canceller (EC) 110 is small. However, β _Z + 8dB
About -15 dB is desirable. In the present embodiment will be described below beta _Z as 0 dB.

Further, the control unit (CTRL) 411k receives the reception power P _X [f] output from the reception power calculation unit (POW) 411d and the transmission power P _Z [output from the transmission power calculation unit (POW) 411e. f] as an input, the received signal suppression unit (SS) 411r suppresses the received input signal x [n], or the transmitted signal suppression unit (SS) 411o transmits the transmitted signal (transmitted input signal z [n]. Alternatively, it is determined whether or not the residual signal e [n]) is to be suppressed, and state information ESstate [f] that is the determination result information is output.

For example, using the reception power P _X [f] and a variable threshold value representing the noise level on the reception side, it is detected whether or not the reception side is voiced, and the transmission power P _Z [f] and the noise level on the transmission side are detected. The state information ESstate [f] is set so as to suppress the transmission signal if the receiving side is sounded, and the transmitting side sets the state information ESstate [f]. If there is sound, the state information ESstate [f] is set so as to suppress the reception signal, and the previous state information ESstate until the difference between the reception power P _X [f] and the transmission power P _Z [f] becomes large. [F] is set to continue.

Here, as the transmission power P _S [f], P _Z [f] is used for a frame for which the echo suppression amount is determined to be insufficient by the control information EScontrol [f], and it is determined that the echo suppression amount is sufficient. P _E [f] may be used in the frame that has been processed.

The signal selection unit 411L includes a transmission input signal z [n] output from the high-pass filter unit (HPF) 109, a residual signal e [n] output from the signal subtraction processing unit 110b, and a control unit (
CTRL) 411k and control information EScontrol [f] are input, and either one of transmission input signal z [n] or residual signal e [n] is output in frame units. Specifically, the control information EScontrol [f] is stored in the echo canceller (EC) 110.
If the frame is detected to have an insufficient echo suppression amount, the transmission input signal Z [n] is selected as a signal. In other frames, the residual signal e [n] is selected as a signal.

In this way, when the echo suppression amount is not sufficient for each frame, that is, when the estimation accuracy of the echo canceller (EC) 110 is not sufficient, the echo suppressor (ES
) 411 can be operated alone.

The gain storage unit (GTBL) 411m stores a gain γ set in advance. However, γ is an analog front end (D / A converter (D / A) 403) in advance before a call is started.
, Receiver amplifier 104, speaker 105, acoustic space, microphone 106, transmitter amplifier 10
7. It is desirable to set based on the echo return loss of the A / D converter (A / D) 108).

The echo suppression gain calculation unit (GCAL) 411n is a gain storage unit (GTBL) 411m.
Gain γ output from E, and E output from echo suppression amount estimation unit (ECLE) 411i.
CL [f] and control information EScontrol output from the control unit (CTRL) 411k
[F] and state information ESstate [f] are input, and an echo suppression gain G [f] is calculated and output.

Specifically, the echo suppression gain calculation unit (GCAL) 411n receives the state information ESstat.
When e [f] indicates that it is determined to suppress the transmission signal (transmission input signal z [n] or residual signal e [n]), the following is performed. First, when the control information EScontrol [f] is a frame detected when the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient, the echo suppression gain G [f] is set to G [f] = γ. .

In other frames, the echo suppression gain G [f] is set to the echo canceller (EC) 1
10 is sufficient, it is determined that the amount of suppression may be increased, and G [f] = γ · ECL [
f]. Alternatively, an echo canceller (EC) 11 is used to prevent excessive suppression.
G [f] = γ / ECL [f] is calculated so that the echo suppression amount of 0 is reduced. On the other hand, when the state information ESstate [f] indicates that it is determined to suppress the incoming input signal, the echo suppression gain G [f] is set to G [f] = γ.

Further, in order to prevent the transmission sound quality from being deteriorated due to excessive echo suppression, background noise is used by using the signal level of the transmission input signal z [n] or the residual signal e [n] that is not a voiced section. The level may be calculated, and the echo suppression gain may be controlled so that the level is not suppressed below the background noise level.

The transmission signal suppression unit (SS) 411o includes state information ESstate [f] output from the control unit (CTRL) 411k, echo suppression gain G [f] output from the echo suppression gain calculation unit (GCAL) 411n, and , The transmission signal (transmission input signal z [n] or residual signal e [n]) selected by the signal selection unit 411L is input, and the transmission signal (transmission input signal) is transmitted by the state information ESstate [f]. z [n] or the residual signal e [n]) is determined to be suppressed, the signal selected by the signal selection unit 411L and the echo suppression gain calculation unit (G
CAL) Using the suppression gain G [f] calculated in 411n, the following Expression 29-1 or Expression 2
The signal is suppressed as indicated by 9-2, and the transmission output signal s ′ [n] is calculated and output.

If it is determined that the transmission signal (transmission input signal z [n] or residual signal e [n]) is not suppressed by the state information ESstate [f], the following Expression 29-3 or Expression 29-
As shown in FIG. 4, the transmission signal is output as it is as the transmission output signal s ′ [n].

At these times, regardless of which signal is selected by the signal selection unit 411L, the echo suppression gain G [f smoothed in the time direction is smoothly connected so that the transmitted transmission output signal s ′ [n] is smoothly connected for hearing. ] May be multiplied.

The received signal suppression unit (SS) 411r includes state information ESstate [f] output from the control unit (CTRL) 411k, echo suppression gain G [f] output from the echo suppression gain calculation unit (GCAL) 411n, When the reception input signal x [n] output from the communication unit (COM) 101 is input, and it is determined that the reception input signal x [n] is suppressed by the state information ESstate [f], the reception signal x [n] Echo suppression gain calculator (GCA)
L) Using the echo suppression gain G [f] calculated in 411n, the signal is suppressed as shown in the following Expression 30-1, and the received output signal x ′ [n] is calculated and output.

If it is determined that the received input signal x [n] is not suppressed by the state information ESstate [f], the received input signal x [n] is directly used as the received output signal x ′ as shown in the following Expression 30-2. Output as [n].

Next, a processing flow of the signal processing device according to the fourth embodiment configured as described above will be described. FIG. 19 is a flowchart showing an overall processing flow of the signal processing apparatus according to the fourth embodiment. Note that the same operation steps as those of the signal processing apparatus according to the first embodiment described with reference to FIG. 5 are denoted by the same reference numerals, and description thereof is omitted.

After the offset removal processing in step S1003, the delay processing unit (DELAY) 402 performs processing for temporarily storing and delaying the reception output signal x ′ [n] output from the reception signal suppression unit (SS) 411r (step S1003). S4004). Then, the process proceeds to the echo canceller process in step S1005. After the echo canceller process in step S1005, the signal processing apparatus
Instead of performing echo reduction processing in step S1006 in the first embodiment, echo suppression processing is performed using the signal after echo canceller processing obtained in step S1005 in the fourth embodiment ( Step S4006). Then, the process proceeds to step S1007 to determine whether or not the call is an end story.

Since the echo cancellation unit (EC) 110 according to the fourth embodiment is the same as the echo cancellation unit (EC) 110 according to the first embodiment, the flow of processing in each embodiment is naturally the same. There is no explanation.

FIG. 20 is a flowchart showing the flow of processing in the echo suppressor unit (ES) 411 according to the fourth embodiment. First, the received power calculator (POW) 411d calculates the received power P _X [f] from the received input signal x [n] (step S4201r), and the transmitted power calculator (POW) 411e receives the transmitted input signal. The transmission power P _Z [f] is calculated from z [n] (step S4201s), and the residual power calculation unit (POW) 411f receives the residual signal e [n.
] To calculate the residual power P _E [f] (step S4201e).

The echo suppression amount estimation unit (ECLE) 411i transmits the transmission power P _Z [f] and the residual power P _E.
Echo suppression amount EC suppressed by echo canceller unit (EC) 110 using [f] as input
L [f] is estimated (step S4202).

The control unit (CTRL) 411k receives the echo suppression amount ECL [f] as an input, determines whether or not the echo suppression amount is sufficient for each frame, and outputs control information EScontrol [f]. In addition, the control unit (CTRL) 411k receives the reception power P _X [f] and the transmission power P _Z [f].
], The received input signal x [n] is suppressed by the received signal suppression unit (SS) 411r, or the transmitted signal (transmitted input signal z [n] or the remaining signal is suppressed by the transmitted signal suppressing unit (SS) 411o. It is determined whether the difference signal e [n]) is suppressed, and the state information ESstate [f] is output. (
Step S4203).

Then, the signal selection unit 411L receives the control information EScontrol [f] as an input, selects and outputs z [n] as a signal for the frame determined to have insufficient echo suppression amount, and the echo suppression amount is sufficient. In a frame determined to be present, e [n] is selected and output as a signal (step S4204).

The echo suppression gain calculation unit (GCAL) 411n controls the control information EScontrol [f].
And the echo suppression amount ECL [f] as an input, the echo suppression gain G [f] is calculated by a different calculation for each of the frames detected as having insufficient echo suppression amount and the other frames.
] Is calculated and output (step S4205).

The transmission signal suppression unit (SS) 411o has an echo suppression gain G [f] and state information ESsta.
When it is determined to suppress the transmission signal (transmission input signal z [n] or residual signal e [n]) using te [f] as an input, the signal selected by the signal selection unit 411L and echo suppression The echo suppression gain G [f] calculated by the gain calculation unit (GCAL) 411n is used to suppress the signal and output it as a transmission output signal s ′ [n]. Alternatively, if it is determined not to suppress the transmission signal, the signal selected by the signal selection unit 411L is output as the transmission output signal s ′ [n] (step S4206s). On the other hand, the received signal suppression unit (SS) 411r receives the echo suppression gain G [f] and the state information ESstate [f] as input, and receives the received input signal x [n].
Is determined to be suppressed, an echo suppression gain calculation unit (GC) is applied to the received signal x [n].
AL) Using the echo suppression gain G [f] calculated in 411n, the signal is suppressed and output as a reception output signal x ′ [n]. Alternatively, when it is determined not to suppress the reception input signal x [n], the reception signal x [n] is output as the reception output signal x ′ [n] (step S).
4206r). As a result, the echo suppressor process ends.

As described above, the echo suppressor unit (ES) 411 has been described as operating as a time domain type processing method for each frame. You may implement | achieve by the system processed to a frequency domain type | mold using band division filters, such as FFT and a filter bank.

At the time of echo path loss fluctuation during double talk, the residual signal e [n] becomes larger than the transmission input signal z [n] by the processing of the echo canceller unit (EC) 110, and the echo canceller unit (EC) 110. In some cases, the amount of echo suppression by this processing is not sufficient.

By the operation of the signal processing apparatus described above, the echo suppression amount estimation unit (ECLE) 411i and the control unit (CTRL) 411k detect when the echo suppression amount due to the processing of the echo canceller unit (EC) 110 is not sufficient, and the echo Since the suppression gain calculation unit (GCAL) 411n and the transmission signal suppression unit (SS) 411o take into account the echo suppression amount of the echo canceller unit (EC) 110, it is possible to avoid suppression of excessive transmission speech. It is possible to output a high-quality signal that is robust against path loss fluctuations.

(Fifth embodiment)
The signal processing device according to the fifth embodiment is different from the signal processing device according to the first embodiment in that it does not have the echo reduction unit (ER) 111, but the echo reduction unit (ER) 511.
It is in the point which has. On the other hand, the other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

FIG. 21 shows an echo reduction unit (ER) of a signal processing device according to the fifth embodiment of the invention.
) Shows a block diagram showing the configuration of 511. The difference between the echo reduction unit (ER) 511 of the signal processing device according to the fifth embodiment of the present invention and the echo reduction unit (ER) 111 of the signal processing device according to the first embodiment will be described below. Note that the echo reduction unit (ER) 111 according to the first embodiment and the noise reduction unit (N) according to the second embodiment.
The same parts as those in R) 211 are given the same reference numerals and their description is omitted.

The echo reduction unit (NR) 511 includes a frequency domain transform processing unit (FT) 111a,
A frequency domain transform processing unit (FT) 111b, a frequency domain transform processing unit (FT) 111c,
Received power calculation unit (POW) 111d, transmission power calculation unit (POW) 111e, residual power calculation unit (POW) 111f, acoustic coupling amount estimation unit (ACLE) 511g, and echo amount estimation unit (ELE) 511h, an echo suppression amount estimation unit (ECLE) 111i, a control unit (
CTRL) 511k, spectrum selection unit 511L, and gain storage unit (GTBL) 111
m, an echo suppression gain calculator (GCAL) 511n, and a frequency domain inverse transform processor (IFT)
) 111p, a transmission signal suppression unit (SS) 511s, and a residual signal suppression unit (ES) 511t.

The echo reduction unit (ER) 511 receives the delayed received input signal x [n−D] output from the delay processing unit (DELAY) 102 and the transmitted input signal z [output from the high pass filter unit (HPF) 109. n] and the residual signal e [
n] as an input, the echo component is suppressed from at least one of the transmission input signal z [n] or the residual signal e [n], and the signal after the echo suppression is transmitted as the transmission output signal s ′ [n] ( n = 0, 1
,..., N-1) are output every frame.

The acoustic coupling amount estimation unit (ACLE) 511g includes the smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d and the transmission power calculation unit (POW) 111e. Smoothed transmission power spectrum output from
Z _s [f, ω] | ² and the smoothed residual power spectrum | E _s [f, ω] | ² output from the residual power calculation unit (POW) 111f are input, and the acoustic coupling amount | H _Z [f
, Ω] | ² and | H _E [f, ω] | ² are calculated by Expression 31-1 and Expression 31-2, respectively.

And the acoustic coupling amount estimation part (ACLE) 511g is the following formula 32-1 and formula 32-2.
Indicates that the amount of acoustic coupling smoothed using the value one frame before | H _ZS [f, ω]
| ² and | H _ES [f, ω] | ² are calculated and output. However, α _HZ [ω] and α _HE
[Ω] is preferably about 0.03 to 0.99.

Here, at the beginning of the call, for example, about 5 seconds after the start of the call, α _HZ [ω] and α _HE [
The acoustic coupling amount | H _ZS [f, ω] | ² and | H _ES [f, ω] | ² by increasing ω].
Speed up the update. By doing so, since the acoustic coupling amount is initialized at the beginning of the call, it is possible to prevent the suppression amount from being reduced at the beginning of the call.

However, when the amount of acoustic coupling changes abruptly, that is, the inequality | H _Z [f, ω] | ² > β _Z
H [ω] · | H _ZS [f−1, ω] | ² and | H _E [f, ω] | ² > β _EH [ω] · | H
_{If ES} [f−1, ω] | ² holds and if the received input signal is not large enough, that is, if the inequality | X _S [f, ω] | ² <β _X [ω] holds, In order not to calculate the amount of acoustic coupling in the frequency band where double talk occurs while maintaining high-speed follow-up to echo path fluctuations, the past acoustic coupling amount one frame before without updating the acoustic coupling amount | H _ZS
[F-1, ω] | ² and | H _ES [f-1, ω] | ² are used.

Since an extreme change in the amount of acoustic coupling has the possibility of double talk, it is possible to prevent deterioration in transmitted sound quality by not updating the amount of acoustic coupling in this way. However, β _ZH [ω] and β _EH [ω] are preferably about 0.9 to 30. β _X [ω] is desirably about 30 dB to 40 dB.

The echo amount estimation unit (ELE) 511h includes the smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d and the acoustic coupling amount estimation unit (ACLE) 511g. Output acoustic coupling amount | H _ZS [f, ω] | ² and | H _E
S [f, ω] | ² as an input, the amount of transmitted echoes | Y _Z [f, ω] | ² included in the frequency spectrum Z [f, ω] of the transmission input signal and the frequency spectrum of the residual signal The residual echo amount | Y _E [f, ω] | ² included in _E [f, ω] is expressed by the following equations 33-1 and 33-2, respectively.
As shown in FIG.

Then, the echo amount estimation unit (ELE) 511h provides an instantaneous echo amount | Y _Z [f, ω] | ²
And | Y _E [f, ω] | ² , the smoothed value can be used to make the signal after echo suppression a more natural signal. Therefore, the following equations 34-1 and 34-2 are shown. 1
Echo amount smoothed using values before frame | Y _ZS [f, ω] | ² and | Y _ES
[F, ω] | ² is calculated and output for each frequency band ω. Where α _ZY [ω] and α _EY
[Ω] is preferably about 0.7 to 0.99.

The control unit (CTRL) 511k receives the echo suppression amount ECL [f, ω] output from the echo suppression amount estimation unit (ECLE) 111i, and the frequency band where the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient. Control information ERcontrol which is information on whether or not
l [f, ω] is output. Specifically, the control unit (CTRL) 511k is a frequency band satisfying ECL [f, ω]> β _Z [ω] for each frequency band ω, and the expression ECL [f, ω] per frame. When the number of bands count in which> β _Z [ω] is established is larger than β _C, it is determined that the echo suppression amount is small. However, β _Z [ω] is preferably about +8 dB to −15 dB, and β _C is preferably about 10% to 40% of the total number of frequency bands. In the present embodiment, β _Z [ω] is set to 0 below.
It will be described as dB.

The echo suppression gain calculation unit (GCAL) 511n is a transmission power calculation unit (POW) 111.
The smoothed transmission power spectrum | Z _S [f, ω] | ² output from e and the smoothed residual power spectrum | E _S [f, output from the residual power calculation unit (POW) 111 f ω] | ² and the smoothed transmission echo amount | Y _ZS [f, ω] | ² and the smoothed residual echo amount | Y output from the echo amount estimation unit (ELE) 511h
_ES [f, ω] | ² and the parameter γ output from the gain storage unit (GTBL) 111m
[Ω] as an input, the transmission echo suppression gain G _Z [f, ω] and the residual echo suppression gain G
_E [f, ω] is calculated and output.

Specifically, the echo suppression gain calculation unit (GCAL) 511n converts the transmission echo suppression gain G _Z [f, ω] and the residual echo suppression gain G _E [f, ω] to a Wiener filter (Wiener
Using the (Filter) method, calculation is performed according to the following Expression 35-1 and Expression 35-2.

Further, the echo suppression gain calculation unit (GCAL) 511n prevents the transmission sound quality from being deteriorated due to excessive echo suppression, and prevents the background noise from being intermittently suppressed, so that the transmission echo suppression gain G _Z [f , Ω] and residual echo suppression gain G _E [f, ω] are controlled so as not to fall below a predetermined lower limit value.

Further, the echo suppression gain calculation unit (GCAL) 511n prevents the transmission sound quality from deteriorating due to excessive echo suppression, so that the voiced interval of the transmission input signal z [n] or the residual signal e [n] The background noise level is calculated for each frequency band using the signal level of the section that is not, and the transmission echo suppression gain G _Z [f, ω] and the residual echo suppression gain G are prevented from being suppressed below the background noise level. _E [f, ω] may be controlled.

Furthermore, the echo suppression gain calculation unit (GCAL) 511n sets the echo suppression gain G _Z [f, ω] to the following Expression 36-1 or Expression 36 in order to prevent the transmission sound quality from deteriorating.
-2 and G _E [f, ω] may be smoothed and output in the frequency direction as shown in the following Expression 36-3 or Expression 36-4. For example, ε _j is [0.1, 0
. 2, 0.4, 0.2, 0.1], η _j is [0.1, 0.2, 0.4, 0.8, 0.4
, 0.2, 0.1].

The transmission signal suppression unit (SS) 511s includes a frequency spectrum Z [f, ω] of the transmission input signal output from the frequency domain conversion processing unit (FT) 111b, and an echo suppression gain calculation unit (GC).
AL) The transmission echo suppression gain G _Z [f, ω] output from 511n is used as an input to suppress the echo of the frequency spectrum of the transmission input signal, and the spectrum S _Z ′ [ f, ω] is output.

The residual signal suppression unit (ES) 511t includes a frequency spectrum E [f, ω] of the residual signal output from the frequency domain transform processing unit (FT) 111c, and an echo suppression gain calculation unit (GCAL).
) The residual echo suppression gain G _E [f, ω] output from 511n is input, and the echo of the frequency spectrum of the residual signal is suppressed, and the spectrum S _E ′ [f, ω] is output as follows: To do.

The spectrum selection unit 511L includes the frequency spectrum S _Z ′ [f, ω] of the echo-suppressed transmission input signal output from the transmission signal suppression unit (SS) 511s and the residual signal suppression unit (ES
) Frequency spectrum S _E ′ [f, ω of echo-suppressed residual signal output from 511t
] And control information ERcon that is information indicating whether or not the echo suppression amount of the echo canceller unit (EC) 110 output from the control unit (CTRL) 511k is an insufficient frequency band
trol [f, ω] as an input and the frequency spectrum S of the echo-suppressed transmission input signal
Either _Z ′ [f, ω] or the frequency spectrum S _E ′ [f, ω] of the echo-suppressed residual signal is selected and output as a frequency spectrum.

Specifically, the control information ERcontrol [f, ω] is the echo canceller (EC) 1
When the frequency band detected that the echo suppression amount of 10 is not sufficient, the spectrum selection unit 511L selects S ′ _Z [f, ω] as the frequency spectrum. For other frequency bands, S ′ _E [f, ω] is selected as the frequency spectrum.

By doing so, when the echo suppression amount of the echo canceller unit (EC) 110 is not sufficient for each frequency band, that is, when the estimation accuracy of the echo canceller unit (EC) 110 is not sufficient, the echo reduction unit (ER) 511 can be operated alone.

Next, a processing flow of the signal processing device according to the fifth embodiment configured as described above will be described. FIG. 22 is a flowchart showing an overall processing flow of the signal processing apparatus according to the fifth embodiment. Note that the same operation steps as those of the signal processing apparatus according to the first embodiment described with reference to FIG. 5 are denoted by the same reference numerals, and description thereof is omitted.

After the echo canceller process in step S1005, the signal processing apparatus replaces the echo reduction process in step S1006 in the first embodiment with the echo obtained in step S1005 in the fifth embodiment. An echo reduction process is performed using the signal after the canceller process (step S5006). Then, the process proceeds to step S1007 to determine whether or not the call is an end story.

Since the echo cancellation unit (EC) 110 according to the fifth embodiment is the same as the echo cancellation unit (EC) 110 according to the first embodiment, the flow of processing in each embodiment is naturally the same. There is no explanation.

FIG. 23 is a flowchart showing the flow of processing in the echo reduction unit (ER) 511 according to the fifth embodiment. The operation of the echo reduction unit (ER) 111 according to the first embodiment described with reference to FIG. 7 and the operation of the noise reduction unit (NR) 211 according to the second embodiment described with reference to FIG. About the same operation step, the same code | symbol is attached | subjected and description of the part is abbreviate | omitted.

The echo reduction unit (ER) 511 according to the fifth embodiment starts the echo reduction process, and the received power spectrum calculation process in step S1203r, step S12.
Sending power spectrum calculation process of 03s, and after the residual power spectrum calculation processing in step S1203e, the acoustic coupling amount estimating unit (ACLE) 511 g is smoothed received power spectrum _{| X S [f, ω]} | 2 and Smoothed transmission power spectrum | Z
_The acoustic coupling amount | H _ZS [f, ω] | ² is calculated by using _S [f, ω] | ² as an input (step S5204s). Similarly, the acoustic coupling amount estimating unit (ACLE) 511 g is smoothed received power spectrum _{| X S [f, ω]} | 2 and the smoothed residual power spectrum _{| E S [f, ω]} | Input ² As a result, the acoustic coupling amount | H _ES [f, ω] | ² is calculated (step S5204e).

Then, the echo estimation unit (ELE) 511h is acoustic coupling amount _{| H ZS [f, ω]} | 2 and smoothing the received power spectrum _{| X S [f, ω]} | 2 and transmission echo amount as input | Y _ZS [f, ω] | ² is estimated (step S5205s). Similarly, the echo estimation unit (ELE) 511h is acoustic coupling amount _{| H ES [f, ω]} | 2 and smoothing the received power spectrum _{| X S [f, ω]} | 2 and residual echo quantity as an input | Y _ES [f, ω] | ²
Is estimated (step S5205e).

Then, the echo suppression gain calculation unit (GCAL) 511n is an echo amount estimation unit (ELE).
511h and the transmission echo amount | Y _ZS [f, ω] | ² and the transmission power calculation unit (P
OW) 111e and the smoothed transmission power spectrum | Z _S [f, ω
] | ² is input, and the transmission echo suppression gain G _Z [f, ω] is calculated, and control is performed so as not to be below a predetermined lower limit value (step S1208s).

Similarly, an echo suppression gain calculation unit (GCAL) 511n is an echo amount estimation unit (ELE).
511h output residual echo amount | Y _ES [f, ω] | ² and residual power calculation unit (P
OW) 111f smoothed residual power spectrum | E _S [f, ω
] ² is input, and the residual echo suppression gain G _E [f, ω] is calculated and controlled so as not to be equal to or lower than a predetermined lower limit (step S1208e).

After that, the transmission signal suppression unit (SS) 511s is a frequency domain transform processing unit (FT) 111b.
The frequency spectrum Z [f, ω] of the transmission input signal output from the signal and the transmission echo suppression gain G _Z [f, ω] calculated by the echo suppression gain calculation unit (GCAL) 511n are input. The echo of the frequency spectrum of the input signal is suppressed (step S5210s).

Similarly, the residual signal suppression unit (ES) 511t is a frequency domain transform processing unit (FT) 111c.
The frequency spectrum E [f, ω] of the residual signal output from, and the echo suppression gain calculation unit (
GCAL) The residual echo suppression gain G _E [f, ω] calculated in 511n is used as an input to suppress the echo of the frequency spectrum of the residual signal (step S5210e).

Next, after the echo suppression amount calculation processing in step S1206 and the determination processing in step S2207, the spectrum selection unit 511L performs the frequency spectrum of the echo-suppressed transmission input signal output from the transmission signal suppression unit (SS) 511s. S _Z '[f, ω] and the residual signal suppression unit (
ES) Frequency spectrum S _E ′ [f of the echo-suppressed residual signal output from 511t
, Ω] and control information ERcontrol [f output from the control unit (CTRL) 511k.
, Ω] as an input, and in the frequency band where the echo suppression amount is detected to be insufficient, the frequency spectrum S _Z ′ [f, ω] of the echo-suppressed transmission input signal is selected and output as the frequency spectrum. In other frequency bands, the frequency spectrum S _E ′ [f, ω] of the residual signal subjected to echo suppression is selected and output as a frequency spectrum (step S5209).

Then, the process proceeds to the frequency inverse transform process in step S1211, and the echo reduction process ends.

As described above, the echo reduction unit (ER) 511 has been described as operating as a method of processing for each frequency band in the frequency domain type by FFT. You may implement | achieve the frequency domain type | molds, such as the method which puts together the frequency band by FFT and processes for every frequency band group, and band division filters, such as a filter bank.

At the time of echo path loss fluctuation during double talk, the residual signal e [n] becomes larger than the transmission input signal z [n] by the processing of the echo canceller unit (EC) 110, and the echo canceller unit (EC) 110. In some cases, the amount of echo suppression by this processing is not sufficient.

By the operation of the signal processing apparatus described above, the echo suppression amount estimation unit (ECLE) 111i and the control unit (CTRL) 511k determine the frequency region where the echo suppression amount by the processing of the echo canceller unit (EC) 110 is not sufficient, An echo canceller (EC) 110 for each frequency band in the spectrum selector 511L and the echo suppression gain calculator (GCAL) 511n.
And the processing weight of the echo reduction unit (ER) 511 can be controlled to be changed, so that it is possible to make robust against echo path loss fluctuations and to output a high-quality signal. Is possible.

Also, the acoustic coupling amount estimation unit (ACLE) 511g of the echo reduction unit (ER) 511
And the echo amount estimation unit (ELE) 511h separately estimate the echo amount contained in the transmission input signal z [n] and the echo amount contained in the residual signal e [n], and the echo suppression gain calculation unit (G
(CAL) 511n separately calculates the echo suppression gain, and the transmission signal suppression unit (SS) 5
11s and the residual signal suppression unit (ES) 511t are separately suppressed, so that it is possible to prevent excessive echo suppression and prevent deterioration of the transmission sound quality.

(Sixth embodiment)
FIG. 24 is a block diagram illustrating a configuration of a signal processing device according to the sixth embodiment. This signal processing device is different from the signal processing device according to the first embodiment in that it does not have an echo canceller (EC) 110 and an echo reduction unit (ER) 111 as shown in FIG. EC) 610 and an echo reduction unit (ER) 611, and an echo suppression amount estimation unit (ECLE) 612 outside the echo reduction unit (ER) 611. The other parts are the same.

Therefore, the same parts as those of the signal processing apparatus according to the first embodiment are denoted by the same reference numerals and description thereof is omitted. In the echo reduction unit (ER) 611, the same parts as those of the echo reduction unit (ER) 111 according to the first embodiment are denoted by the same reference numerals, and description thereof is omitted.

FIG. 25 is a block diagram showing the configuration of the echo canceller unit (EC) 610. The echo canceller (EC) 610 includes a frequency domain transform processor (FT) 610d, a frequency domain adaptive filter unit (FDADF) 610e, and a frequency domain inverse transform processor (IFT) 610f.
, A signal subtraction processing unit 610g, a frequency domain conversion processing unit (FT) 610h, and a frequency domain double talk detection unit (FDDTD) 610i.

FIG. 26 is a block diagram showing a configuration of the echo reduction unit (ER) 611. The echo reduction unit (ER) 611 includes a frequency domain transformation processing unit (FT) 111a, a frequency domain transformation processing unit (FT) 111b, a frequency domain transformation processing unit (FT) 111c, and a received power calculation unit (POW). 111d, transmission power calculation unit (POW) 111e, acoustic coupling amount estimation unit (ACLE) 611g, echo amount estimation unit (ELE) 111h, frequency domain double talk detection unit (FDTD) 611j, and control unit ( CTRL) 611k, spectrum selection unit 111L, gain storage unit (GTBL) 111m, echo suppression gain calculation unit (
GCAL) 611n, signal suppressor (SS) 111o, and frequency domain inverse transform processor (IF
T) 111p and a transmission output power calculation unit (POW) 611u.

FIG. 27 is a block diagram illustrating a configuration of the acoustic coupling amount estimation unit (ACLE) 611g. The acoustic coupling amount estimation unit (ACLE) 611g is an acoustic coupling amount estimation unit (CACL) 611g1.
And an acoustic coupling amount correction unit (ADJ) 611g2 and an acoustic coupling amount smoothing unit (SMACL) 611
g3.

The operation of each part of the signal processing device according to the sixth embodiment of the present invention configured as described above will be described with reference to FIGS.

The echo canceller unit (EC) 610 includes a transmission input signal z [n] output from the high-pass filter unit (HPF) 109 and a delayed reception input signal x [n− output from the delay processing unit (DELAY) 102. D] as an input, Overlap-Save Method
), Or based on the overlap-add method, the transmission input signal z [
n] suppresses the echo component, and the signal after the echo suppression is the residual signal e [n] (n = 0, 1
,..., N−1), and the filter coefficient H _FDAF [f, ω] and the estimated echo path loss λ _FDAF [f, ω] are output.

The frequency domain transform processing unit (FT) 610d receives the delayed received input signal x [n−D] output from the delay processing unit (DELAY) 102 as input, and performs FFT (Fast Fourier Transform).
) _{Or the like, and} the frequency spectrum _XFDAF [f
, Ω] is calculated and output. At this time, the overlap-save method (Overlap-Save Metho
d) Or, based on the overlap-add method, perform windowing using a Hamming window, etc., use past frames, perform zero interpolation, or overlap.

The frequency domain adaptive filter unit (FDADF) 610e includes a filter coefficient H _FDAF [f,
ω] is a frequency domain adaptive filter composed of a variable transversal filter. Also, the frequency domain adaptive filter unit (FDADF) 610e is a frequency spectrum X of the received input signal output from the frequency domain transform processing unit (FT) 610d.
_FDAF [f, ω], the frequency spectrum E _FDAF [f−1, ω] of the residual signal one frame before output from the frequency domain transform processing unit (FT) 610h, and the frequency domain double talk detecting unit (FDDTD) ) Double talk information ECstate [f, output from 610i
ω] and the double talk information ECstate [f, ω] is not in the double talk state, the filter coefficient H _FDAF [f, ω] is adaptively learned for each frame f and frequency band ω, and the double talk information is obtained. If ECstate [f, ω] is in the double talk state, adaptive learning is not performed. In this way, the filter coefficient H _FDAF [f, ω] is calculated and output.

In addition, the frequency domain adaptive filter unit (FDADF) 610e includes a frequency domain conversion processing unit (
FT) The frequency spectrum Y ′ of the pseudo echo signal using the frequency spectrum X _FDAF [f, ω] of the received input signal output from _610d and the filter coefficient H _FDAF [f, ω].
_FDAF [f, ω] is changed to Y ′ _FDAF [f, ω] = H _FDAF [f, ω] · X _FDAF [f
, Ω].

The frequency domain adaptive filter unit (FDADF) 610e includes a filter coefficient H _FDAF [f,
Adaptive learning is performed using a fixed or variable step size μ _F [f, ω] for controlling the update width of ω].

Further, the frequency domain adaptive filter unit (FDADF) 610e is, for example, an LMS (Least-Me).
an-Square) algorithm, NLMS (Normalized-Least-Mean-Square) algorithm,
Learning identification method, affine projection (AP) algorithm, sequential least squares (
Adaptive filter based on linear adaptive algorithm such as RLS (Recursive-Least-Squares) algorithm and gradient-limited learning identification method (Gradient-limited Normalized-Least-Mean-Square)
e), composed of adaptive filters based on non-linear adaptive algorithms such as Adaptive Volterra Filter. In this embodiment, there is no gradient constraint (gradient
An example of an unconstrained frequency domain adaptive filter, but with gradient constraints (grad
(ient constrained) frequency domain type adaptive filter.

The frequency domain inverse transform processing unit (IFT) 610f includes a frequency domain adaptive filter unit (FDAD).
F) Frequency spectrum Y ′ _FDAF [f, ω] of the pseudo echo signal output from _610e
, And the pseudo echo signal y ′ _FDAF [n] (n = 0, 1,..., N−1) is calculated and output by IFFT (Inverse Fast Fourier Transform) or the like. At this time, the overlap-save method (overlap-save method) or the overlap-add method (overl
Based on ap-Add Method), the past frame is used, zero interpolation is performed, and overlap is returned.

Signal subtraction unit 610g includes a high-pass filter unit and (HPF) output from 109 sending speech input signal z [n], the pseudo echo signal y _'FDAF [n output from the frequency domain inverse transform unit (IFT) 610f ] And the pseudo-echo signal y from the transmission input signal z [n]
' _FDAF [n] is subtracted for each sample n, the echo component is suppressed, and a residual signal e [n] which is a signal after the echo suppression is calculated and output.

The frequency domain transform processing unit (FT) 610h receives the time domain residual signal e [n] output from the signal subtraction processing unit 610g, converts it into the frequency domain by FFT (Fast Fourier Transform), etc. The frequency spectrum E _FDAF [f, ω] of the difference signal is calculated and output. At this time, based on the overlap-save method (Overlap-Save Method) or overlap-add method (Overlap-Add Method),
Use past frames, perform zero interpolation, or overlap.

The frequency domain double talk detector (FDDTD) 610i is a frequency domain conversion processor (F
T) Frequency spectrum X _FDAF [f, ω] of the received signal output from 610d and frequency spectrum E _FDAF [f−1] of the residual signal one frame before output from the frequency domain transform processing unit (FT) 610h. , Ω] as input, it is determined for each frame f and frequency band ω whether or not it is in a double talk state, double talk information ECstate [f, ω] that is information indicating whether or not it is in a double talk state, and an echo Λ _FDAF [f, ω] which is the estimated path loss
Is calculated and output.

Specifically, first, the frequency domain double-talk detector (FDDDTD) 610i determines the power spectrum of the received signal | X _FDAF [f, ω] | ² and the power spectrum of the residual signal one frame before | E _FDAF [f− 1, ω] | ² for each frame f and frequency band ω. And the inequality | E _FDAF [f−1, ω] | ² > λ _FDAF [f, ω] · | X _{FDA
If F} [f, ω] | ² holds, it is determined that the state is a double talk state. Where λ _FDAF [f
, Ω] is an estimated value of the echo path loss, and is a variable amount that decreases as adaptive learning progresses and increases when adaptive learning is incorrect. Also, λ _FDAF [f, ω] is the filter coefficient H
_FDAF [f, ω] is updated and calculated for each adaptively learned frame f and frequency band ω.

In this case, the echo canceller unit (EC) 610 has a filter coefficient H _FDAF [f, ω].
, Step size μ _F [f, ω], estimated echo path loss λ _FDAF [f, ω], double talk information ECstate [f, ω], received signal power spectrum | X _FDAF [f
, Ω] | ² and the power spectrum of the residual signal | E _FDAF [f, ω] | ² are held in the memory as internal states.

An echo suppression amount estimation unit (ECLE) 612 is a frequency domain adaptive filter unit (FDADF).
The filter coefficient H _FDAF [f, ω] output from the 610e and the estimated echo path loss λ _FDAF [f, ω] output from the frequency domain double talk detector (FDDTD) 610i.
ω] as an input and the echo suppression amount ECL suppressed by the echo canceller unit (EC) 610
[F, ω] is estimated and output for each frequency band ω. Specifically, when not in the double talk state, | E _FDAF [f, ω] | ^{2 is changed} to | E _FDAF [f, ω] | ² = λ _FDAF [f, ω
] · | X _FDAF [f, ω] | ² and approximate the power spectrum of the transmission input signal | Y
'The _{ratio of FDAF} [f, ω] + E _FDAF [f, ω] | ² is the echo suppression amount ECL [f, ω
] Is calculated as shown in Equation 39 below.

The echo reduction unit (ER) 611 receives the received signal x [n] and the transmitted input signal z [n].
And the echo component is suppressed based on the residual signal e [n], and the signal after the echo suppression is transmitted as the transmission output signal s ′ [n] (n = 0, 1,..., N− Output as 1).

The acoustic coupling amount estimation unit (ACLE) 611g includes the smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d, and the transmission power calculation unit (POW) 111e. Smoothed transmission power spectrum output from
Z _S [f, ω] | ² , the echo suppression amount ECL [f, ω] output from the echo suppression amount estimation unit (ECLE) 612, and the frequency output from the frequency domain double talk detection unit (FDTD) 611j Area double talk information ERstate [f, ω] and an echo amount estimation unit (EL
E) 111h echo of one frame before output from _{| Y S [f-1,} ω] | ² and the input smoothing correction acoustic coupling amount _{| H S [f, ω]} | 2 calculates the Output.

The acoustic coupling amount estimation unit (CACL) 611g1 includes the smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d and the transmission power calculation unit (POW) 111e. sending power spectrum is smoothed output from _{| Z S [f, ω]} | ² and the input, so as not to be affected by the performance of the echo canceller (EC) 610 based on the residual signal | E _S [f, ω] ^{| 2} based on the transmission input signal without using the _{| Z S [f, ω]} | 2 with an acoustic coupling amount for each frequency band ω | H [f, ω] | 2 below Is calculated and output as shown in Equation 40.

The acoustic coupling amount correction unit (ADJ) 611g2 is an acoustic coupling amount estimation unit (CACL) 611g1.
Acoustic output amount | H [f, ω] | ² output from the signal and an echo suppression amount estimation unit (ECLE) 61
2 and echo transmission amount ECL [f, ω] output from the transmission power calculation unit (POW) 111.
The smoothed transmission power spectrum | Z _S [f, ω] | ² output from e and the echo amount one frame before output from the echo amount estimation unit (ELE) 111 h | Y _S [f−1
, Ω] | ² as an input, and the acoustic coupling amount | H [f based on the echo suppression amount ECL [f, ω]
, Ω] | ² , the corrected acoustic coupling amount | H ′ [f, ω] | ² is calculated and output.

Specifically, the acoustic coupling amount correction unit (ADJ) 611g2 first determines whether the echo suppression amount suppressed by the echo canceller unit (EC) 610 is sufficient as follows. For each frequency band ω, ECL [f, ω]> β _ACL [ω] and count> β _CACL are also taken into account, with the number of bands per frame in which the inequality ECL [f, ω]> β _ACL [ω] is established. It is determined that the echo suppression amount of the echo canceller unit (EC) 610 is not sufficient. However, β _ACL [ω] is preferably about +8 dB to −35 dB, and β _CACL is preferably about 10% to 40% of the total number of frequency bands. In the present embodiment, hereinafter, β _ACL [ω] is described as 0 dB.

For the frequency band for which the echo suppression amount is determined to be sufficient, the acoustic coupling amount correction unit (AD
J) 611g2 is an expression 41 shown below only when the transmission power spectrum | Z _S [f, ω] | ² is larger than the echo amount | Y _S [f-1, ω] | ^{2 of} one frame before. The acoustic coupling amount | H [f, ω] | ² is corrected as shown below, and the corrected acoustic coupling amount | H ′ [f, ω] | ² is calculated.

When the above condition is not satisfied, and in a frequency band in which the echo suppression amount of the echo canceller unit (EC) 610 is determined to be insufficient, the acoustic coupling amount correction unit (ADJ) 611g2 performs | H ′ [f, ω] ^{| 2 = | H [f,} ω] | 2 as not corrected.

The acoustic coupling amount smoothing unit (SMCL) 611g3 is an acoustic coupling amount correction unit (ADJ) 611g.
2 and the corrected acoustic coupling amount | H ′ [f, ω] | ² output from the frequency domain double talk information ERstate [f] output from the frequency domain double talk detector (FDTD) 611j.
As input and omega], smoothed by using the value of one frame before as shown in Equation 42 below _{| H S [f, ω]} | 2 and outputs them. However, α _H [ω] is 0.03 to 0.
. About 99 is desirable.

Here, the acoustic coupling amount smoothing unit (SMCL) 611g3 increases the acoustic coupling amount | H _S [f, ω] | ² by increasing α _H [ω] at the beginning of the call, for example, for about 5 seconds from the start of the call. Speed up the update. By doing so, since the acoustic coupling amount is initialized at the beginning of the call, it is possible to prevent the suppression amount from being reduced at the beginning of the call.

However, if the frequency domain double talk information ERstate [f, ω] output from the frequency domain double talk detector (FDTD) 611j is determined to be in the double talk state, the acoustic coupling amount changes abruptly, that is, , Inequality | H _S [f, ω] | ² > β _H [
ω] · | H _S [f−1, ω] | ² holds, and when the received signal is not sufficiently large, that is, the inequality | X _S [f, ω] | ² <β _X [ω] holds. In this case, the acoustic coupling amount smoothing unit (SMCL) 611g3 does not calculate the acoustic coupling amount in the frequency band where double talk is maintained while maintaining high-speed followability to the echo path fluctuation. The past acoustic coupling amount | H _S [f−1, ω] | ² one frame before is used without updating the amount.

Since an extreme change in the amount of acoustic coupling has the possibility of double talk, it is possible to prevent deterioration in transmitted sound quality by not updating the amount of acoustic coupling in this way. However, β _H [ω] is 0.
. About 9-30 is desirable. β _X [ω] is desirably about 30 dB to 40 dB.

The frequency domain double talk detector (FDTD) 611j is a received power calculator (POW).
Smoothed received power spectrum output from 111d | X _S [f, ω] | ²
And the transmission output power spectrum | S ′ _S [f−1, ω] | ² that has been smoothed one frame before output from the transmission output power calculation unit (POW) 611u, and is in a double talk state. The frequency domain double talk information ERstate [f, ω] is calculated and output as information indicating whether or not. Specifically, the frequency domain double talk detector (FDTD) 611j
Inequality _{| S 'S [f-1} , ω] | 2> λ ER [f, ω] · | X S [f, ω] | determines that the double-talk state in the case ^{where 2} is true.

Here, λ _ER [f, ω] is an estimated value of the echo path loss, and the acoustic coupling amount | H _S [f, ω
] ² is updated for each updated frame f and frequency band ω, and the acoustic coupling amount | H _S [f, ω
] | ^{2 is} a variable amount that decreases as the update proceeds, and increases as the acoustic coupling amount | H _S [f, ω] | ² does not progress.

The control unit (CTRL) 611k includes an echo suppression amount ECL [f, ω] output from the echo suppression amount estimation unit (ECLE) 612 and a frequency domain double talk detection unit (FDTD) 611.
frequency domain double talk information ERstate [f, ω] output from j as input,
For each frequency band ω, the frequency band ω detects a frequency band in which the double talk state and the echo suppression amount of the echo canceller (EC) 610 are not sufficient, thereby detecting the double talk state and the echo canceller (EC
) The control information ERcontrol [f, ω], which is information indicating whether the echo suppression amount of 610 is a frequency band that is not sufficient, is output.

First, whether the control unit (CTRL) 611k is in the double talk state for each frequency band ω depending on whether or not the frequency domain double talk detection unit (FDTD) 611j determines that it is in the double talk state for each frequency band ω. Determine whether or not.

Next, the control unit (CTRL) 611k performs the inequality ECL [f, ω]> β for each frequency band ω.
_{When Z} [ω] is satisfied and the inequality count> β _C is satisfied, it is determined that the echo suppression amount of the echo canceller (EC) 610 is small in the frequency band. However, count is the number of bands in which the expression ECL [f, ω]> β _Z [ω] per frame is established, and β _Z [ω]
Is preferably about +8 dB to −15 dB, and β _C is preferably about 10% to 40% of the total number of frequency bands.
In the present embodiment, the following description will be made assuming that β _Z [ω] = 2.5 (+8 dB).

The echo suppression gain calculation unit (GCAL) 611n is a transmission power calculation unit (POW) 111.
The smoothed transmission power spectrum | Z _S [f, ω] | ² output from e and the smoothed echo amount output from the echo amount estimation unit (ELE) 111 h | Y _S [f,
ω] | ² and the echo suppression amount ECL output from the echo suppression amount estimation unit (ECLE) 612
[F, ω] and control information ERcontrol output from the control unit (CTRL) 611k
[F, ω] and the parameter γ [ω] output from the gain storage unit (GTBL) 111m are input, and the smoothed echo suppression gain G _S [f, ω] is calculated and output.

Specifically, the echo suppression gain calculation unit (GCAL) 611n controls the control information ERcont.
For the frequency band detected from rol [f, ω] in a double talk state and the echo suppression amount of the echo canceller (EC) 610 is not sufficient, the echo suppression gain G [f, ω]
Is calculated by Equation 43 using the Wiener Filter method.

In other frequency bands, the echo suppression amount of the echo canceller unit (EC) 610 is sufficient and the echo canceller unit (EC) 610 is regarded as functioning normally, and the echo suppression gain calculation unit (GCAL) 611n. Is the echo suppression amount ECL [f of the echo canceller unit (EC) 610 so that the echo suppression amount of the echo reduction unit (ER) 611 is increased.
, Ω] as an argument, the function F (ECL [f [f] satisfying the inequality 0 ≦ F (ECL [f, ω]) ≦ 1.
, Ω]), and the echo suppression gain G [f, ω] is calculated as shown in Equation 44 below.

Here, the function F (ECL [f, ω]) is as follows. However, [delta] _C it is desirable to positively value considering the number of frequency bands.
When 0 ≦ ECL [f, ω] ≦ 1, Equation 45-1 gives
When 1 ≦ ECL [f, ω] ≦ β _Z [ω]
_{β Z [ω] ≦ ECL [} f, ω] according to equation 45-3 or Formula 45-4 when.

Thus, the function F (ECL [f, ω]) as the correction gain of the suppression gain is expressed by the inequality 0 ≦
By setting so as to satisfy F (ECL [f, ω]) ≦ 1, it is possible to prevent the corrected suppression gain from becoming an unnatural value, and it is possible to prevent deterioration of the transmission sound quality. Further, by making the function F (ECL [f, ω]) a non-linear function in this way, the echo canceller (E
C) When the echo suppression amount of 610 is not sufficient (β _Z [ω] ≦ ECL [f, ω]), the gain can be corrected so as to increase the suppression amount.

The echo suppression gain calculation unit (GCAL) 611n prevents the transmission sound quality from being deteriorated due to excessive echo suppression, and prevents the background noise from being intermittently suppressed.
] Is controlled so as not to fall below a predetermined lower limit value.

Further, in order to prevent the transmission sound quality from deteriorating due to excessive echo suppression, the echo suppression gain calculation unit (GCAL) 611n is provided with a sound interval of the transmission input signal z [n] or the residual signal e [n]. A background noise level may be calculated for each frequency band using a signal level of a section that is not, and the echo suppression gain may be controlled so as not to suppress the background noise level.

Furthermore, the echo suppression gain calculation unit (GCAL) 611n sets the echo suppression gain G [f, ω] to the following Expression 46-1 or Expression 46- in order to prevent the transmission sound quality from deteriorating.
As indicated by 2, the output may be smoothed in the frequency direction. For example, ε _j is [0.
1, 0.2, 0.4, 0.2, 0.1], η _j is [0.1, 0.2, 0.4, 0.8,
0.4, 0.2, 0.1].

The transmission output power calculation unit (POW) 611u receives the frequency spectrum S ′ [f, ω] of the transmission output signal output from the signal suppression unit (SS) 111o as an input, and the transmission output power that is the power spectrum. spectrum | S '[f, ω] | 2 is calculated, and the previous frame values as in equation 47 below _{| S' S [f-1} , ω] | 2 sending output power was smoothed using The spectrum | S ′ _S [f, ω] | ² is calculated and output. However, α _S [ω] is 0
. About 75 to 0.999 is desirable.

Next, a processing flow of the signal processing device according to the sixth embodiment configured as described above will be described. FIG. 28 is a flowchart showing an overall processing flow of the signal processing apparatus according to the sixth embodiment. Note that the same operation steps as those of the signal processing apparatus according to the first embodiment described with reference to FIG. 5 are denoted by the same reference numerals, and description thereof is omitted.

After the delay process in step S1004, the echo canceller unit (EC) 610 performs the echo canceller process using the delayed received input signal x [n-D] and the transmission input signal z [n] from which the offset is removed as input (step S1004). Step S6005).

Next, the echo suppression amount estimation unit (ECLE) 612 includes a frequency domain adaptive filter unit (FDA).
DF) 610e filter coefficient H _FDAF [f, ω] and frequency domain double talk detector (FDDTD) 610i estimated echo path loss λ _FDAF
[F, ω] is used as an input, and the echo suppression amount ECL [f, ω] suppressed by the echo canceller (EC) 610 is estimated (step S6008).

Then, the delayed reception input signal x [n−D] and the transmission input signal z with the offset removed.
[N] and the residual signal e [n], which is a signal after error canceller processing output from the echo canceller unit (EC) 610, are input, and the echo reduction unit (ER) 611 is an echo that is nonlinear echo suppression processing. Reduction processing is performed (step S6006). Then, the process proceeds to step S1007 to determine whether or not the call is an end story.

FIG. 29 is a flowchart showing the flow of processing in the echo canceller unit (EC) 610 according to the sixth embodiment.

The processing of the echo canceller (EC) 610 is performed as follows. First, the received input signal x [n−D] is converted into the frequency domain, and the frequency spectrum X _FDAF [
f, ω] is calculated, the residual signal e [n] is converted into the frequency domain, and the frequency spectrum E _FDAF [f, ω] of the residual signal is calculated. Next, the frequency domain double talk detector (FDDT)
D) 610i uses the frequency spectrum _XFDAF [f, ω] of the received signal and the frequency spectrum E _FDAF [f-1, ω] of the residual signal one frame before to perform frequency domain double talk detection processing. (Step S6101). The frequency domain adaptive filter unit (FDA
DF) 610e is double-talk information ECstate [f, ω] while receiving control of the frequency spectrum _{X FDAF} of the received signal [f, omega] a, the frequency spectrum of the previous frame of the residual signal _E FDAF [f-1, ω] is used to perform frequency domain adaptive filter processing (step S6102). Next, the frequency spectrum Y ′ _FDAF [f, ω] of the pseudo echo signal is inversely transformed in the frequency domain to calculate the pseudo echo signal y ′ _FDAF [n]. Then, the signal subtraction processing unit 610g generates a frequency domain adaptive filter unit (FDADF) from the transmission input signal z [n].
) The pseudo echo signal y ′ _FDAF [n] output from 610 e is subtracted and the residual signal e [n]
] Is calculated and the output and echo canceller processing ends.

FIG. 30 is a flowchart showing the flow of processing in the echo reduction unit (ER) 611 according to the sixth embodiment. Note that the same operation steps as the operations in the echo reduction unit (ER) 111 according to the first embodiment described with reference to FIG. 7 are denoted by the same reference numerals, and description thereof is omitted.

The echo reduction unit (ER) 611 according to the sixth embodiment performs step S1203e.
There is no operation of the residual power spectrum calculation process.

That is, the echo reduction unit (ER) 611 according to the sixth embodiment starts the echo reduction process, and performs the received power spectrum calculation process in step S1203r, the transmitted power spectrum calculation process in step S1203s, and the remaining steps in step S1202e. After the difference frequency conversion process, the frequency domain double talk detection unit (FDTD) 611j calculates and outputs frequency domain double talk information ERstate [f, ω], which is information indicating whether or not it is in a double talk state. The estimation unit (ACLE) 611g includes the smoothed reception power spectrum | X _S [f, ω] | ² and the smoothed transmission power spectrum | Z _S [f,
ω] | ² and the echo suppression amount ECL output from the echo suppression amount estimation unit (ECLE) 612
Based on [f, ω] and the frequency domain double talk information ERstate [f, ω], a corrected acoustic coupling amount | H _S [f, ω] | ² is calculated (step S6204).

After performing echo level calculation processing in step S1205, the control unit (CTRL) 6
11k is an echo suppression amount ECL [f, ω] and frequency domain double talk information ERstat.
e [f, ω] as input, a frequency band in which the double talk state and the echo suppression amount of the echo canceller (EC) 110 are not sufficient is detected for each frequency band ω, and the control information ERcon is detected.
trol [f, ω] is output (step S6207).

Next, the echo suppression gain calculation unit (GCAL) 611n includes the smoothed transmission power spectrum | Z _S [f, ω] | ² , the echo amount | Y _S [f, ω] | ^2, and the echo suppression amount. ECL [f, ω], control information ERcontrol [f, ω], and parameter γ [ω] are input, and echo suppression gain G [f, ω] is calculated and output (step S6208).
.

Then, after performing spectrum selection processing in step S1209 and transmission signal suppression processing in step S1210, the transmission output power calculation unit (POW) 611u calculates a transmission output power spectrum (step S6203ss). Then, the process proceeds to the frequency inverse transform process in step S1211, and the echo reduction process ends.

In the above, the echo canceller unit (EC) 610 shows an example of a frequency domain adaptive filter based on the overlap-save method or the overlap-add method in this embodiment. However, the circular convolution method (Circular-Con
A frequency domain type adaptive filter based on the “volution method” may be used.

In the above description, the echo reduction unit (ER) 611 is described as operating in a frequency domain type using FFT and processing for each frequency band. You may implement | achieve the frequency domain type | molds, such as the method which puts together the frequency band by FFT and processes for every frequency band group, and band division filters, such as a filter bank.

By the operation of the signal processing apparatus described above, the echo suppression amount estimation unit (ECLE) 612 and the control unit (CTRL) 611k determine the frequency region where the echo suppression amount due to the processing of the echo canceller unit (EC) 610 is not sufficient, The spectrum selection unit 111L and the echo suppression gain calculation unit (GCAL) 611n can be controlled to change the weights of the processing of the echo canceller unit (EC) 610 and the processing of the echo reduction unit (ER) 611 for each frequency band. Therefore, it is possible to make robust against echo path loss fluctuations and to output a high quality signal.

(Modification of the sixth embodiment)
FIG. 31 is a block diagram showing a configuration of an echo reduction unit (ER) 6112 of a signal processing device according to a modification of the sixth embodiment of the present invention. The echo reduction unit (ER) 6112 according to the sixth embodiment is the echo reduction unit (ER) 6 according to the sixth embodiment.
11 is that the acoustic coupling amount estimation unit (ACLE) 611g is not included and the acoustic coupling amount estimation unit (ACLE) 611g-2 is included, and the other portions are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

FIG. 32 shows an acoustic coupling amount estimation unit (ACLE) 611g-2 according to a modification of the sixth embodiment.
It is a block diagram which shows the structure of these. The acoustic coupling amount estimation unit (ACLE) 611g-2 includes an acoustic coupling amount estimation unit (CACL) 611g1-2 and an acoustic coupling amount smoothing unit (SMACL) 611.
g3-2 and an acoustic coupling amount correction unit (ADJ) 611g2-2.

The acoustic coupling amount estimation unit (CACL) 611g1-2 receives the received power calculation unit (POW) 111.
The smoothed reception power spectrum | X _S [f, ω] | ² output from d and the smoothed transmission power spectrum | Z _S [f, ω output from the transmission power calculation unit (POW) 111e. ] ² and the frequency domain double talk information ERstate [f, ω] output from the frequency domain double talk detection unit (FDTD) 611j, and the performance of the echo canceller unit (EC) 610 is not affected. Based on the residual signal | E _S [f
, Ω] | ^2, and using | Z _S [f, ω] | ² based on the transmission input signal, the acoustic coupling amount | H [f, ω] | ² is shown below for each frequency band ω. It calculates and outputs like Formula 48.

However, when the frequency domain double talk information ERstate [f, ω] output from the frequency domain double talk detection unit (FDTD) 611j indicates that the double talk state is determined, or the amount of acoustic coupling is If it changes rapidly, that is, the inequality | H [f
, Ω] | ² > β _H [ω] · | H [f−1, ω] | ² and when the received signal is not sufficiently large, that is, the inequality | X _S [f, ω] | ² When <β _X [ω] is satisfied, an acoustic coupling amount estimation unit is provided so as not to calculate the acoustic coupling amount in the frequency band that causes double talk while maintaining high-speed follow-up to echo path fluctuations. (CACL) 611g1-2
Does not update the acoustic coupling amount, and the past acoustic coupling amount one frame before | H [f−1, ω] | ²
Is used.

Since an extreme change in the amount of acoustic coupling has the possibility of double talk, it is possible to prevent deterioration in transmitted sound quality by not updating the amount of acoustic coupling in this way. However, β _H [ω] is 0.
. About 9-30 is desirable. β _X [ω] is desirably about 30 dB to 40 dB.

The acoustic coupling amount smoothing unit (SMACL) 611 g 3-2 is an acoustic coupling amount estimation unit (CACL) 6.
11g1-2, the acoustic coupling amount | H [f, ω] | ² and the echo suppression amount estimation unit (E
CLE) echo suppression amount output from 612 ECL [f, ω] as input and the acoustic coupling amount was smoothed by using the value of the previous frame as in Equation 49 below | H _S '[f, ω
] | ² is calculated and output.

However, α _H [ω] is preferably about 0.03 to 0.99. At this time, α _H [ω] is made variable for each frequency band ω based on the echo suppression amount ECL [f, ω] suppressed by the echo canceller unit (EC) 610. That, as the echo suppression amount ECL [f, ω] is enough, close the alpha _H [omega] 0, as the echo suppression amount ECL [f, omega] is not sufficient alpha _H [omega]
Is brought close to 1.

The acoustic coupling amount correction unit (ADJ) 611g2-2 is connected to an acoustic coupling amount smoothing unit (SMACL) 61.
The smoothed acoustic coupling amount | H _S ′ [f, ω] | ² output from 1g3-2 and the echo suppression amount ECL [f, ω] output from the echo suppression amount estimation unit (ECLE) 612 are input. The corrected acoustic coupling amount | H _S [f, ω] | ² is calculated and output.

Specifically, the acoustic coupling amount correction unit (ADJ) 611g2-2 first determines whether or not the echo suppression amount ECL [f, ω] suppressed by the echo canceller unit (EC) 610 for each frequency band is sufficient. Determine. Then, in a frequency band in which the echo suppression amount ECL [f, ω] suppressed by the echo canceller unit (EC) 610 is determined to be sufficient, 0 <β _HACL [ω] ≦
The smoothed acoustic coupling amount | H _S ′ [f, ω] | ² using the coefficient β _HACL [ω] of 1 is _corrected as shown in Equation 50 below, and the corrected acoustic coupling amount | H _S [f, ω] | ² is calculated and output.

On the other hand, the echo suppression amount ECL [f, ω] of the echo canceller unit (EC) 610 is the determined frequency bands not sufficient, the acoustic coupling amount correction unit (ADJ) 611g2-2 is, | _{H S}
^{[F, ω] | 2 =} | H S '[f, ω] | smoothed acoustic coupling amount as ² | H _s'
[F, ω] | ² is not corrected. By doing so, it is possible to prevent the transmission sound quality from deteriorating when the echo path fluctuation does not occur so much.

(Seventh embodiment)
FIG. 33 is a block diagram illustrating a configuration of a signal processing device according to the seventh embodiment. This signal processing device is different from the signal processing device according to the sixth embodiment in that an echo canceller (EC) 610, an echo reduction (ER) 611, and an echo suppression amount estimator (ECLE) as shown in FIG. 612, but has an echo canceller unit (EC) 710, an echo reduction unit (ER) 711, and an echo suppression amount estimation unit (ECLE) 712, and other parts are the same. Therefore, the same parts as those of the signal processing apparatus according to the sixth embodiment are denoted by the same reference numerals, and the description thereof is omitted.

Further, the echo reduction unit (ER) 711 according to the seventh embodiment is the same as the echo reduction unit (ER) 111 according to the first embodiment and the echo reduction unit (ER) 611 according to the sixth embodiment. Are denoted by the same reference numerals, and the description thereof is omitted.

FIG. 34 is a block diagram showing the configuration of the echo canceller unit (EC) 710. The echo canceller unit (EC) 710 includes a time domain echo canceller unit (TDAF) and a frequency domain echo canceller unit (FDAF), which are connected in series. That is, processing of the frequency domain echo canceller (FDAF) is performed using the signal after processing in the time domain echo canceller (TDAF).

The time domain echo canceller (TDAF) is the same as the echo canceller (EC) 110 according to the first embodiment, and the frequency domain echo canceller (FDAF)
This is the same as the echo canceller unit (EC) 610 according to the embodiment.

The delayed received input signal x [n output from the delay processing unit (DELAY) 102
-D] is input to the echo canceller unit (EC) 110 and the echo canceller unit (EC) 610. Further, the residual signal e [
n] is input to the echo canceller (EC) 610 as a transmission input signal z [n],
Connect in series. In this case, the echo canceller (EC) 710 is connected to the echo canceller (
The internal state of the EC) 110 and the internal state of the echo canceller unit (EC) 610 are stored in the memory as the internal state.

Therefore, the echo canceller unit (EC) 110 according to the first embodiment and the echo canceller unit (EC) 610 according to the sixth embodiment are denoted by the same reference numerals, and the echo canceller unit (E
C) Description of 710 is omitted.

FIG. 35 is a block diagram showing the configuration of the echo reduction unit (ER) 711. As shown in FIG. The echo reduction unit (ER) 711 includes a frequency domain transformation processing unit (FT) 111a, a frequency domain transformation processing unit (FT) 111b, a frequency domain transformation processing unit (FT) 111c, and a received power calculation unit (POW). 111d, transmission power calculation unit (POW) 111e, residual power calculation unit (POW) 111f, echo amount estimation unit (ELE) 711h, frequency domain double talk detection unit (FDTD) 711j, control unit ( CTRL) 611k, spectrum selection unit 111L, gain storage unit (GTBL) 111m, echo suppression gain calculation unit (G
CAL) 611n, signal suppressor (SS) 111o, frequency domain inverse transform processor (IFT)
) 111p.

FIG. 36 is a block diagram showing a detailed configuration of the echo amount estimation unit (ELE) 711h.
The echo amount estimation unit (ELE) 711h includes a signal subtraction processing unit 711h1, an echo amount estimation control unit (ELECTRL) 711h2, variable gains 711h3 and 711h4, and a signal addition processing unit 711h5.

The operation of each part of the signal processing device according to the seventh embodiment of the present invention configured as described above will be described with reference to FIGS.

First, an echo suppression amount estimation unit (ECLE) 712 includes a high-pass filter unit (HPF) 10.
9 and the time domain residual signal e [n] output from the signal subtraction processing unit 610g of the echo canceller unit (EC) 710 as inputs, and the echo canceller unit ( EC) The echo suppression amount ECL [f, ω] suppressed in 610 is estimated for each frequency band ω and output.

Specifically, the echo suppression amount estimation unit (ECLE) 712 first windows and overlaps the transmission input signal z [n], converts it to the frequency domain by FFT, and transmits the transmission power spectrum | Z ′ [ f, ω] | ² is calculated. Next, the residual signal e [n] is windowed and overlapped and transformed into the frequency domain by FFT, and the residual power spectrum | E ′ [f, ω]
| ² is calculated. Then, the transmission power spectrum | Z ′ [f,
ω] | ² and the residual power spectrum | E ′ [f, ω] | ²
) Calculate and output as the echo suppression amount ECL [f, ω] suppressed in 710.

Next, referring to FIG. 35 and FIG. 36, the echo reduction unit (ER
) The operation of 711 will be described.

The echo reduction unit (ER) 711 receives the reception signal x [n] and the transmission input signal z [n].
And the echo component is suppressed based on the residual signal e [n], and the signal after the echo suppression is transmitted as the transmission output signal s ′ [n] (n = 0, 1,..., N− Output as 1).

The echo amount estimation unit (ELE) 711h receives the smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d and the transmission power calculation unit (POW) 111e. Output smoothed transmission power spectrum | Z _S
[F, ω] | ² , the smoothed residual power spectrum output from the residual power calculation unit (POW) 111 f | E _S [f, ω] | ^2, and the echo suppression amount estimation unit (ECLE) 7
12 and the echo suppression amount ECL [f, ω] output from the frequency domain double talk detector (
FDTD) frequency domain double talk information ERstate [f, ω output from 711j
] Is input and a smoothed echo amount | Y _S [f, ω] | ² is calculated and output.

The signal subtraction processing unit 711h1 transmits the transmission power spectrum | Z _S [f, ω] | ² output from the transmission power calculation unit (POW) 111e and the residual output from the residual power calculation unit (POW) 111f. differential power spectrum _{| E S [f, ω]} | ² and the input, sending power spectrum _{| Z S [f, ω]} | 2 residual power spectrum from _{| E S [f, ω]} | ² frames f
And the subtracted signal | Z _S [f, ω] | ² − | E _S [f, ω]
| ² is output.

The echo amount estimation control unit (ELECTRL) 711h2 is a received power calculation unit (POW) 1
The smoothed reception power spectrum | X _S [f, ω] | ² output from 11d and the smoothed transmission power spectrum | Z _S [f, ω output from the transmission power calculation unit (POW) 111e. ] | ² , the smoothed residual power spectrum output from the residual power calculation unit (POW) 111 f | E _S [f, ω] | ^2, and output from the echo suppression amount estimation unit (ECLE) 712 Echo suppression amount ECL [f, ω] and frequency domain double talk information ERst output from frequency domain double talk detector (FDTD) 711j
ate [f, ω] is input, and two variable gains G ₁ [f, ω] and G ₂ [f, ω] are calculated and output.

Specifically, first, the echo amount estimation control unit (ELECTRL) 711h2 in the single talk state except when the frequency domain double talk information ERstate [f, ω] indicates that it is determined that the state is the double talk state. In order to estimate the amount of echo | Y _S [f, ω] | ² with high accuracy, G _HZ , G as shown in Equations 52-1, 52-2, and 52-3 shown below.
Three gains _HE and _GHZE are calculated.

Next, the echo amount estimation control unit (ELECTRL) 711h2 selects the echo suppression amount ECL [f
, Ω] as an argument, and three functions F _HZ (ECL [f, ω]), F having a range of 0 to 1
_HE (ECL [f, ω]) and F _HZE (ECL [f, ω]) are controlled as follows. If the echo suppression amount ECL [f, ω] is not sufficient, the residual power spectrum | E _S [f
, Ω] | ² to reduce the influence of F _HE (ECL [f, ω]) and F _HZE (EC
L [f, ω]) is set to a value close to 0 or small, and F _HZ (ECL [f, ω]) is set.
) Is brought close to 1.

On the other hand, if the echo suppression amount ECL [f, ω] is sufficient, the residual power spectrum | _{E S}
F _HZ (ECL [f, ω]) is set to a value close to 0 or small so that [f, ω] | ² is used reliably, and F _HE (ECL [f, ω]) and F _HZE (ECL [f, ω
]) Close to 1. At this time, a determination condition similar to that of the control unit (CTRL) 611k may be used as a determination condition as to whether or not the echo suppression amount ECL [f, ω] is sufficient. And
Two variable gains G ₁ [f, ω] and G ₂ [f, ω] are expressed by the following equations 53-1:
, As shown in Equation 53-2.

The variable gain 711h3 includes a smoothed reception power spectrum | X _S [f, ω] | ² output from the reception power calculation unit (POW) 111d, and an echo amount estimation control unit (EL
ECTh) 711h2 and variable gain G ₁ [f, ω] as input, | X _S
[F, ω] | ² is multiplied by G ₁ [f, ω], and the received power spectrum | X _S [f, ω] | ²
Is amplified or attenuated and output.

The variable gain 711h4 is the signal | Z _S [f, output from the signal subtraction processing unit 711h1.
ω] | ² − | E _S [f, ω] | ² and an echo amount estimation control unit (ELECTRL) 711h2.
And variable gain _G 2 [f, _ω] and an input that is output _{from, | Z S [f, ω} ] | 2 - | E S
Multiply [f, ω] | ² by G ₂ [f, ω], and output after amplification or attenuation.

Signal addition processing unit 711h5 is, _G 1 output from the variable gain 711h3 [f, ω] multiplied by _{| X S [f, ω]} | a ^2, _G 2 outputted from the variable gain 711h4 [f, ω] Is inputted as | Z _S [f, ω] | ² − | E _S [f, ω] | ² as an input, and a smoothed echo amount | Y _S [f, ω] | ² is calculated and output. . Specifically, the following formula 54-1
| X _S [f, ω] | multiplied by G ₁ [f, ω] for each frame f and frequency band ω
² multiplied by G ₂ [f, ω] | Z _S [f, ω] | ² − | E _S [f, ω] | ² is added to calculate the echo amount | Y [f, ω] | ² To do.

And as shown in Equation 54-2 below, the echo amount ^{| Y [f, ω] |} 2 smoothed echo amount _{| Y s [f, ω]} | 2 and outputs them. However, αY [ω] is 0.7-0.
. About 99 is desirable. Of course, the echo amount | Y [f, ω] | ² that is not smoothed may be output.

The frequency domain double talk detector (FDTD) 711j is a received power calculator (POW).
Smoothed received power spectrum output from 111d | X _S [f, ω] | ²
, The smoothed echo amount | Y _S [f−1, ω] | ² output from the echo amount estimation unit (ELE) 711h and the smoothing output from the residual power calculation unit (POW) 111f residual power spectrum _{| E S [f, ω]} | ² and the input is information indicating whether or not a double-talk state frequency domain double talk information ERstate [f,
ω] is calculated and output.

Specifically, the frequency domain double-talk detector (FDTD) 711j determines that the state is a double-talk state if the frequency band ω is larger than the threshold as follows, that is, if the inequality 55 shown below holds. However, β _Y [ω] is preferably about 1.0 to 20.

Next, a processing flow of the signal processing device according to the seventh embodiment configured as described above will be described. FIG. 37 is a flowchart showing an overall processing flow of the signal processing apparatus according to the seventh embodiment. Note that the same operation steps as those of the signal processing apparatus according to the first embodiment described with reference to FIG. 5 are denoted by the same reference numerals, and description thereof is omitted.

After the delay processing in step S1004, the echo canceller unit (EC) 710 performs echo canceller processing using the delayed reception input signal x [n-D] and the transmission input signal z [n] from which the offset is removed as input (FIG. Step S7005).

Next, an echo suppression amount estimation unit (ECLE) 712 includes a high-pass filter unit (HPF) 10.
Echo suppression suppressed by an echo canceller (EC) 710 using the transmission input signal z [n] output from 9 and the time domain residual signal e [n] output from the signal subtraction processor 610g as inputs. The quantity ECL [f, ω] is estimated (step S7008).

Then, the delayed reception input signal x [n−D] and the transmission input signal z with the offset removed.
[N] and the residual signal e [n], which is the signal after error canceller processing output from the echo canceller unit (EC) 710, are input, and the echo reduction unit (ER) 711 is an echo that is nonlinear echo suppression processing. Reduction processing is performed (step S7006). Then, the process proceeds to step S1007 to determine whether or not the call is an end story.

FIG. 38 is a flowchart showing the flow of processing in the echo canceller (EC) 710 according to the seventh embodiment. Note that the same operation steps as the operation of the echo canceller unit (EC) 110 according to the first embodiment described with reference to FIG. 6 are denoted by the same reference numerals, and description thereof is omitted.

The echo canceller unit (EC) 710 first performs double talk detection processing in step S1101, and then performs adaptive filter processing in step S1102. The signal subtraction processing unit 110b uses the transmission input signal z [n]. The pseudo echo signal y ′ [n] output from the adaptive filter unit (ADF) 110a is subtracted to calculate a residual signal e _TDAF [n]. Next, frequency domain double talk detection processing is performed (step S6101), frequency domain adaptive filter processing is performed, and the signal subtraction processing unit _{610g performs} frequency domain adaptive filter unit (FDADF) 610e from the residual signal e _TDAF [n]. _Is subtracted from the pseudo echo signal y ′ _FDAF [n] output from, and the residual signal e [n] is calculated and output (step S6102), and the echo canceller processing is terminated.

FIG. 39 is a flowchart showing the flow of processing in the echo reduction unit (ER) 711 according to the seventh embodiment. The operation of the echo reduction unit (ER) 111 according to the first embodiment described with reference to FIG. 7 and the operation of the echo reduction unit (ER) 611 according to the sixth embodiment described with reference to FIG. About the same operation step, the same code | symbol is attached | subjected and description of the part is abbreviate | omitted.

The echo reduction unit (ER) 711 according to the seventh embodiment starts the echo reduction process, and the received power spectrum calculation process in step S1203r, step S12.
After the transmission power spectrum calculation process of 03s and the residual power spectrum calculation process of step S1203e, the frequency domain double talk detection unit (FDTD) 711j is frequency domain double talk information that is information indicating whether or not a double talk state is present. ERstate [f, ω] is output, and the echo amount estimation unit (ELE) 711h receives the received power spectrum | X _S [f, ω] |
² , transmission power spectrum | Z _S [f, ω] | ² and residual power spectrum | E _S [f
, Ω] | ² , echo suppression amount ECL [f, ω], and frequency domain double talk information ERst
Using ate [f, ω] as an input, the smoothed echo amount | Y _S [f, ω] | ² is calculated and output (step S7205).

Then, after performing the determination process in step S6207, the echo suppression gain calculation process in step S6208, the spectrum selection process in step S1209, and the transmission signal suppression process in step S1210, the process proceeds to the frequency inverse transform process in step S1211, and echo reduction is performed. The process ends.

In the above description, the echo reduction unit (ER) 711 has been described as operating as a method of processing for each frequency band in the frequency domain type by FFT. You may implement | achieve the frequency domain type | molds, such as the method which puts together the frequency band by FFT and processes for every frequency band group, and band division filters, such as a filter bank.

By the operation of the signal processing apparatus described above, the echo suppression amount estimation unit (ECLE) 712 and the control unit (CTRL) 611k determine the frequency region where the echo suppression amount by the processing of the echo canceller unit (EC) 710 is not sufficient, The spectrum selection unit 111L and the echo suppression gain calculation unit (GCAL) 611n can be controlled to change the weights of the processing of the echo canceller unit (EC) 710 and the processing of the echo reduction unit (ER) 711 for each frequency band. Therefore, it is possible to make robust against echo path loss fluctuations and to output a high quality signal.

(Modification of the seventh embodiment)
FIG. 40 is a block diagram showing a configuration of a signal processing device according to a modification of the seventh embodiment of the present invention. The signal processing device according to the modification of the seventh embodiment is different from the signal processing device according to the seventh embodiment in that an echo canceller (EC) 710, an echo reduction unit (E
R) 711 and echo suppression amount estimation unit (ECLE) 712 instead of echo canceller unit (E
C) 710r, echo reduction unit (ER) 711r, and echo suppression amount estimation unit (EC
LE) 712r. Then, the transmission input signal storage unit (BUFF) 713, the reception input signal storage unit (BUFF) 714, and the delay processing unit (DELAY) 702r are included.

Then, in the signal processing apparatus according to the seventh embodiment shown in FIG. 33, the echo generated when the reception output signal wraps around the transmission input signal and is acoustically coupled is suppressed from the transmission input signal with reference to the reception input signal. In contrast, in the signal processing apparatus according to the modification of the seventh embodiment, the transmission output signal circulates at the far end (not shown) and the acoustically coupled echo is received with reference to the transmission output signal. This is a configuration that suppresses the input signal, and the other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

In the transmission input signal storage unit (BUFF) 713 and the reception input signal storage unit (BUFF) 714, for example, the technique described in Japanese Patent Application Laid-Open No. 2005-142886 or the cross-correlation between the transmission input signal and the reception input signal is used. With the delay time estimation technique, an echo generated at a far end (not shown) is synchronized using a process such as a delay.

The delay processing unit (DELAY) 702r, the echo canceller unit (EC) 710r, the echo reduction unit (ER) 711r, and the echo suppression amount estimation unit (ECLE) 712r are different in input signals, but are related to the seventh embodiment. The same processing as that of the delay processing unit (DELAY) 102, the echo canceller unit (EC) 710, the echo reduction unit (ER) 711, and the echo suppression amount estimation unit (ECLE) 712 is performed.

By adopting such a configuration, it is possible to suppress echo even for an acoustic echo generated at the far end.

Note that the present invention is applied to a call device, and the call device generally refers to a device having a call function. For example, a hands-free call device having a hands-free call function, a telephone, an interphone, a mobile phone, PHS, VoIP software, VoIP system, TV
Includes a conference system.

As an example of signal processing for reducing at least the echo component, echo canceller, echo reduction, echo suppressor, and echo noise reduction have been described, and as an example of signal processing for reducing at least the noise component, noise reduction and echo noise reduction have been described. This is because they are general names, and the present invention is not limited to these names. In addition, the present invention can be implemented in a combination of these signal processings without departing from the gist of the present invention.

The present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the invention at the stage of implementation. Further, the above embodiments include inventions at various stages, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements.

In addition, for example, even if some configuration requirements are deleted from all the configuration requirements shown in each embodiment,
When the problem described in the column of the problem to be solved by the invention can be solved and the effect described in the effect of the invention can be obtained, a configuration from which this constituent requirement is deleted can be extracted as the invention. The present invention is established not only as a device but also as a method and program for realizing the function of the device.

1 is a block diagram showing a configuration of a signal processing device according to a first embodiment of the present invention. The block diagram which shows the structure of the echo canceller part (EC) which concerns on the 1st Embodiment of this invention. The block diagram which shows the structure of the echo reduction part (ER) which concerns on the 1st Embodiment of this invention. The figure which shows an example of the lower limit value of the suppression gain which concerns on embodiment of this invention. 3 is a flowchart showing the operation of the signal processing apparatus according to the first embodiment of the present invention. The flowchart which shows operation | movement of the echo canceller part (EC) which concerns on the 1st Embodiment of this invention. The flowchart which shows operation | movement of the echo reduction part (ER) which concerns on the 1st Embodiment of this invention. The block diagram which shows the structure of the echo reduction part (ER) which concerns on the modification of the 1st Embodiment of this invention. The block diagram which shows the structure of the signal processing apparatus which concerns on the 2nd Embodiment of this invention. The block diagram which shows the structure of the noise reduction part (NR) which concerns on the 2nd Embodiment of this invention. The flowchart which shows operation | movement of the signal processing apparatus which concerns on the 2nd Embodiment of this invention. The flowchart which shows operation | movement of the noise reduction part (NR) which concerns on the 2nd Embodiment of this invention. The block diagram which shows the structure of the signal processing apparatus which concerns on the 3rd Embodiment of this invention. The block diagram which shows the structure of the echo noise reduction part (ENR) which concerns on the 3rd Embodiment of this invention. The flowchart which shows operation | movement of the signal processing apparatus which concerns on the 3rd Embodiment of this invention. The flowchart which shows operation | movement of the echo noise reduction part (ENR) which concerns on the 3rd Embodiment of this invention. The block diagram which shows the structure of the signal processing apparatus which concerns on the 4th Embodiment of this invention. The block diagram which shows the structure of the echo suppressor part (ES) which concerns on the 4th Embodiment of this invention. The flowchart which shows operation | movement of the signal processing apparatus which concerns on the 4th Embodiment of this invention. The flowchart which shows operation | movement of the echo suppressor part (ES) which concerns on the 4th Embodiment of this invention. The block diagram which shows the structure of the echo reduction part (ER) which concerns on the 5th Embodiment of this invention. The flowchart which shows operation | movement of the signal processing apparatus which concerns on the 5th Embodiment of this invention. The flowchart which shows operation | movement of the echo reduction part (ER) which concerns on the 5th Embodiment of this invention. The block diagram which shows the structure of the signal processing apparatus which concerns on the 6th Embodiment of this invention. The block diagram which shows the structure of the echo canceller part (EC) which concerns on the 6th Embodiment of this invention. The block diagram which shows the structure of the echo reduction part (ER) which concerns on the 6th Embodiment of this invention. The block diagram which shows the structure of the acoustic coupling amount estimation part (ACLE) which concerns on the 6th Embodiment of this invention. The flowchart which shows operation | movement of the signal processing apparatus which concerns on the 6th Embodiment of this invention. The flowchart which shows operation | movement of the echo canceller part (EC) which concerns on the 6th Embodiment of this invention. The flowchart which shows operation | movement of the echo reduction part (ER) which concerns on the 6th Embodiment of this invention. The block diagram which shows the structure of the echo reduction part (ER) which concerns on the modification of the 6th Embodiment of this invention. The block diagram which shows the structure of the acoustic coupling amount estimation part (ACLE) which concerns on the modification of the 6th Embodiment of this invention. The block diagram which shows the structure of the signal processing apparatus which concerns on the 7th Embodiment of this invention. The block diagram which shows the structure of the echo canceller part (EC) which concerns on the 7th Embodiment of this invention. The block diagram which shows the structure of the echo reduction part (ER) which concerns on the 7th Embodiment of this invention. The block diagram which shows the structure of the echo quantity estimation part (ELE) which concerns on the 7th Embodiment of this invention. The flowchart which shows operation | movement of the signal processing apparatus which concerns on the 7th Embodiment of this invention. The flowchart which shows operation | movement of the echo canceller part (EC) which concerns on the 7th Embodiment of this invention. The flowchart which shows operation | movement of the echo reduction part (ER) which concerns on the 7th Embodiment of this invention. The block diagram which shows the structure of the signal processing apparatus which concerns on the modification of the 7th Embodiment of this invention.

Explanation of symbols

110, 610, 710, 710r Echo canceller (EC)
110a Adaptive filter unit (ADF)
110b, 610g, 711h1 Signal subtraction processor 110c Double talk detector (DTD)
111, 511, 611, 711, 711r, 1112 Echo reduction part (ER)
111a, 111b, 111c, 610d, 610h Frequency domain transform processing unit (FT)
111d, 411d Received power calculation unit (POW)
111e, 411e Transmission power calculator (POW)
111f, 411f Residual power calculator (POW)
111g, 111g2, 511g, 611g, 611g-2 acoustic coupling amount estimation unit (ACL)
E)
111h, 511h, 711h Echo amount estimation unit (ELE)
111i, 411i, 612, 712, 712r Echo suppression amount estimation unit (ECLE)
111j, 611j, 711j Frequency domain double talk detector (FDTD)
111k, 211k, 411k, 511k, 611k control unit (CTRL)
111L, 211L, 511L Spectrum selection unit 111m, 411m Gain storage unit (GTBL)
111n, 111n2, 411n, 511n, 611n Echo suppression gain calculator (GC
AL)
111o, 211o, 311o Signal suppression unit (SS)
111p, 610f Frequency domain inverse transform processing unit (IFT)
211 Noise Reduction Unit (NR)
211n Noise suppression gain calculator (GCAL)
211q, 311q Noise level estimator (NLE)
311 Echo Noise Reduction Unit (ENR)
311n Echo noise suppression gain calculator (GCAL)
411 Echo Suppressor (ES)
411L Signal selection unit 411o, 511s Transmission signal suppression unit (SS)
411r Received signal suppressor (SS)
511t Residual signal suppression unit (ES)
610e Frequency domain adaptive filter unit (FDADF)
610i Frequency domain double talk detector (FDDTD)
611g1, 611g1-2 Acoustic coupling amount estimation unit (CACL)
611g2, 611g2-2 Acoustic coupling amount correction unit (ADJ)
611g3, 611g3-2 Acoustic coupling amount smoothing section (SMACL)
611u Transmission output power calculation unit (POW)
711h2 Echo amount estimation control unit (ELECTRL)
711h3, 711h4 Variable gain 711h5 Signal addition processing unit

Claims (16)

First signal processing means for suppressing an echo contained in the input signal and outputting an echo reduction signal;
Second signal processing means for inputting the input signal and the echo reduction signal and suppressing at least one of echo and noise;
Echo suppression amount calculating means for calculating an echo suppression amount of the first signal processing means by calculating an amount indicating a difference in power spectrum between the input signal and the echo reduction signal;
With
The second signal processing means includes
If an echo suppression amount of the first signal processing means which is calculated by the echo suppression quantity calculating means is determined not to be sufficient to select the input signal, otherwise selecting the echo reduction signal Having a selection means,
You suppress at least one of either said echo and said noise selected the input signal and the echo-reduced signal by said selecting means
Signal processing apparatus. The second signal processing means includes
At least one of a suppression amount calculated by estimating at least one of the echo and the noise included in at least one of the input signal and the echo reduction signal, and an echo suppression amount calculated by the echo suppression amount calculation means And a suppression amount setting means for setting the suppression amount of the signal selected by the selection means,
You suppressed by the suppression amount set by said suppression amount setting means
The signal processing apparatus according to 請 Motomeko 1. The second signal processing means includes
Based on the echo suppression amount calculated by the echo suppression amount calculation means , included in the first suppression amount and the echo reduction signal signal calculated by estimating at least one of the echo and the noise included in the input signal Suppression amount setting means for selecting one of the second suppression amounts calculated by estimating at least one of the echo and the noise ,
It suppressed by one suppression amount either previous SL suppression amount the first suppression amount is selected by the setting means and said second suppression amount
The signal processing apparatus according to 請 Motomeko 1. The second signal processing means includes
When the input signal is selected by the selection unit, the first suppression amount calculated by estimating the echo or the noise included in the input signal is selected, and the echo reduction signal is selected by the selection unit. If there is a suppression amount setting means for selecting a second suppression amount calculated by estimating the echo or the noise included in the echo reduction signal ,
It suppressed by suppressing the amount selected by the suppression amount setting means
The signal processing apparatus according to 請 Motomeko 1. First signal processing means for suppressing an echo at least with respect to an input signal and outputting an echo reduction signal;
Second signal processing means for inputting the input signal and the echo reduction signal, and suppressing and outputting at least one of echo and noise;
Echo suppression amount calculating means for calculating an echo suppression amount of the first signal processing means by calculating an amount indicating a difference in power spectrum between the input signal and the echo reduction signal;
With
The second signal processing means includes
First suppression amount setting means for calculating a first suppression amount by estimating at least one of the echo and the noise included in the input signal;
Second suppression amount setting means for calculating a second suppression amount by estimating at least one of the echo and the noise included in the echo reduction signal;
When it is determined that the echo suppression amount of the first signal processing means calculated by the echo suppression amount calculation means is not sufficient, at least one of the echo and the noise included in the input signal is used as the first suppression. A signal suppressed by the first suppression amount estimated by the amount setting means is selected; otherwise , at least one of the echo and the noise included in the echo reduction signal is selected by the second suppression amount setting means. and selection means for selecting a signal which has been suppressed in the second suppression amount estimated,
You output the signal selected by said selection means
Signal processing apparatus. The suppression amount setting means includes:
Echo amount estimation means for estimating an echo amount based on the echo suppression amount calculated by the echo suppression amount calculation means;
To set the reduction quantity of the second signal processing means using the echo quantity estimated by said echo estimation means
The signal processing device according to claim 2 . The suppression amount setting unit estimates an echo amount for each frequency region by the echo amount estimation unit, and uses the echo amount estimated for each frequency region to determine the suppression amount of the second signal processing unit for each frequency region. the signal processing apparatus according to 請 Motomeko 6 to set. It said suppression amount setting means, the signal processing apparatus according to any one of the second claim to set the amount of suppression of the signal processing means for each frequency region 2 to claim 5. The echo amount estimating means includes
An acoustic coupling amount estimating means for estimating an acoustic coupling amount;
An acoustic coupling amount correction unit that corrects the acoustic coupling amount estimated by the acoustic coupling amount estimation unit based on the echo suppression amount calculated by the echo suppression amount calculation unit;
Have
The signal processing apparatus according to 請 Motomeko 6 you estimate the echo amount using the acoustic coupling amount corrected by the acoustic coupling amount correcting means. The echo amount estimation means estimates the echo amount for each frequency region,
The acoustic coupling amount estimation means estimates the acoustic coupling amount for each frequency region,
The acoustic coupling amount correcting means, correct the acoustic coupling amount for each of the frequency domain
The signal processing apparatus according to 請 Motomeko 9. The suppression amount setting means includes:
Echo amount estimation means for estimating the echo amount;
A suppression gain calculating means for calculating a suppression gain using the echo amount estimated by the echo amount estimating means;
Suppression gain correction means for correcting the suppression gain calculated by the suppression gain calculation means based on the echo suppression amount calculated by the echo suppression amount calculation means;
Have
To set the suppression gain corrected by the suppression gain correcting means as a suppression amount
The signal processing device according to claim 2 . The echo amount estimation means estimates the echo amount for each frequency region,
The suppression gain calculation means calculates the suppression gain for each frequency domain,
The suppression gain correcting means, you correcting the suppression gain for each frequency domain
The signal processing apparatus according to 請 Motomeko 11. It said selecting means, the signal processing apparatus according to any one of 請 Motomeko 1 to claim 12 that select either of the two signals for each frequency region. The echo suppression amount calculation means calculates the echo suppression amount of the first signal processing means for each frequency domain ,
The selection means selects one of the two signals for each frequency domain.
The signal processing device according to any one of claims 1 to 13 . Wherein the input signal is a signal processing apparatus according to any one of the transmission signal der Ru請 Motomeko 1 to claim 14 picked up by the microphone. Wherein the input signal is a signal processing apparatus according to any one of the received signals Der Ru請 Motomeko 1 to claim 15.

Images