JP2001094480A - Method and device for suppressing echo - Google Patents

Abstract

PROBLEM TO BE SOLVED: To suppress an echo signal by working out a desired echo suppression amount that is in matching with various loudspeaking environments and universal to users. SOLUTION: A masking threshold with respect to echo signals is obtained from all masker signals masking the echo signals, a desired echo suppression amount is worked out from the masking threshold and the echo signals, and the echo signals are suppressed based thereon.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for optimally suppressing echo signals that cause howling and impair hearing in the speech quality, for example, in a two-wire / four-wire conversion system and a loudspeaker system. The present invention relates to a method for deriving an echo suppression amount.

[0002]

2. Description of the Related Art FIG. 1 is a schematic diagram of a loudspeaker system. In FIG. 1, reference numerals 1 and 3 denote transmission microphones, reference numerals 2 and 4 denote reception speakers, reference numerals 5 and 7 denote transmission signal amplifiers, reference numerals 6 and 8 denote reception signal amplifiers, reference numeral 9 denotes a transmission line, and reference numerals 10 and 11 denote speakers. , 12 represent the listener, respectively. The transmitted voice uttered by the transmitter 10 is transmitted to the transmitter 11 via the transmission microphone 1, the transmission signal amplifier 5, the transmission path 9, the reception signal amplifier 8, and the reception speaker 4. Unlike the conventional telephone communication system, this loudspeaker communication system does not require the handset to be held in the hand, so it has the advantages of being able to make a call while working and of realizing a natural face-to-face call. It is widely used for teleconferences, videophones, loudspeakers, and the like.

On the other hand, as a drawback of this communication system, the existence of an echo is a problem. That is, in FIG. 1, the sound transmitted from the receiving speaker 4 to the receiving side is received by the transmitting microphone 3, and passes through the transmitting signal amplifier 7, the transmission line 9, the receiving signal amplifier 6, and the receiving speaker 2. Played back to the sender. For the sender 10 and the receiver 12, this phenomenon is an echo phenomenon in which the voice uttered by oneself is reproduced from the receiving speaker 2, and is called an acoustic echo or the like. This echo phenomenon causes adverse effects such as trouble in communication and discomfort in a voice communication system. Further, the sound reproduced from the receiving speaker 2 is received by the transmitting microphone 1 to form a closed loop of the signal. And
If the loop gain is larger than 1, a howling phenomenon occurs, and a call cannot be made.

An echo canceller is used to overcome such a problem of the voice communication system. FIG.
The echo canceller 21 is roughly divided into an adaptive filter processing unit 22 and a loss insertion processing unit 23.
Consists of The adaptive filter processing unit 22 estimates the impulse response of the acoustic echo path, generates a pseudo echo signal, and cancels the echo by subtracting it from the microphone output signal. Here, the adaptive filter estimates an impulse response using an adaptive algorithm in order to follow the temporal variation of the acoustic echo path. The adaptive algorithm is an algorithm that determines an estimated value that minimizes the power of an error signal obtained by subtracting a pseudo echo signal from a microphone output signal.

[0005] A loss insertion processing unit 23 such as a voice switch or an echo suppressor suppresses residual echo that cannot be eliminated by the adaptive filter by inserting a loss into the line. When designing an echo canceller composed of such an adaptive filter processing unit and a loss insertion circuit, it is necessary to determine "to what level should the echo be reduced?" Is important.

The reason is that if the optimum desired echo suppression amount can be determined, the number of taps in the adaptive filter processing unit can be optimized. This is because an increase in scale can be prevented. Further, in the loss insertion processing unit, it is possible to insert a loss only for the residual echo that has not been completely eliminated by the adaptive filter processing unit, thereby preventing deterioration of transmitted voice due to excessive loss insertion and unnatural intermittent feeling of background noise. This is because you can do it.

Until now, the amount of desired echo suppression for a band has been studied in ITU-T recommendation G.167 and the like.
Furthermore, since the auditory characteristics have a frequency-dependent property,
The desired echo suppression amount for each frequency band is also disclosed in JP-A-11-122144.
No. 9-278444.

[0008]

Since all of the methods for determining the desired echo suppression amount described in the prior art are based on subjective evaluation, there is a problem that the obtained experimental values are poor in objective life. The first problem is that the experimental value obtained by the subjective evaluation depends on the loudspeaker environment in which the experiment was performed, for example, the acoustic characteristics and the transmission characteristics. In this case, there is no problem in a loudspeaker environment using an assumed echo canceller, but it cannot be applied to a loudspeaker environment having at least one characteristic. That is, in order to obtain a value suitable for various loudspeaker environments, it is necessary to perform a subjective evaluation in all loudspeaker environments, which is practically impossible. However, the use environment of loudspeakers is diversifying from conventional conference rooms to homes, busy streets, cars, etc.
It is necessary to determine the desired amount of echo suppression in various environments.

The second problem is that, in the subjective evaluation, the desired echo suppression amount is determined using the evaluation result of the evaluator.
However, due to the diversification of users, a problem arises in terms of generality if only a limited number of raters are used. In addition, in order to obtain a more accurate value, a large number of raters that are statistically sufficient are required. As described above, in the subjective evaluation, it is difficult to obtain an optimum desired echo suppression amount for various loudspeaker environments and users. When the desired echo suppression amount is determined using the experimental values obtained in the subjective evaluation,
Since the operation and loss insertion of the adaptive filter as described above cannot be performed optimally, an increase in hardware scale and deterioration in speech quality are caused.

[0010] It is an object of the present invention to provide a system which is suitable for each individual environment in various voice communication environments.
It is an object of the present invention to provide an objective method for a user to derive a universal desired echo suppression amount.

[0011]

In order to objectively derive the desired echo suppression amount, it is necessary to quantitatively treat all of the acoustic characteristics, transmission characteristics, and the auditory characteristics of a user who makes a loudspeaker call environment. There is. That is, since the acoustic characteristics and the transmission characteristics are different in various use environments, substitution is performed as a parameter at the time of derivation. As the auditory characteristics, known auditory characteristic data is used so as to cover various users.

The most characteristic feature of the present invention is to objectively derive a desired echo suppression amount by using parameters of various characteristics of a loudspeaker environment and data of known auditory characteristics. In the prior art, there was a problem that the subjective evaluation could not be performed for all of the various loudspeaker environments and the result depended on the evaluator performing the subjective evaluation. However, the present invention is different in that the objective evaluation can be performed.

[0013]

In the present invention, since the acoustic characteristics and the transmission characteristics of the loudspeaker environment are used as parameters, it is possible to individually derive desired echo suppression amounts suitable for various use environments. On the other hand, since the desired echo suppression amount is derived using a known auditory characteristic, it is possible to determine an acoustically optimal value regardless of the user.
Therefore, it is possible to derive the optimum desired echo suppression amount, and it is possible to derive the desired echo suppression amount individually in various communication environments, which is the object of the present invention, and also for the user. . Further, by controlling the loss insertion processing unit using the desired echo suppression amount obtained in this way, it is possible to promote a reduction in the size of the echo canceller and an improvement in communication quality in the design of the echo canceller.

[0014]

FIG. 3 shows the configuration of an echo canceller 21 according to the present invention. The echo canceller 21 includes an adaptive filter processing unit 22, a loss insertion processing unit 23, and a microphone 3 and a speaker 4 connected to the adaptive filter processing unit 22. The loss insertion processing unit 23 includes a masker and masking discrimination unit 30, a frequency analysis unit A31, a frequency analysis unit B32, and a desired echo suppression amount (= insertion loss amount) determination unit 3.
3. It comprises loss insertion means 34 and frequency synthesis means 35. (Here, the masker means not only the received voice, but also all the sounds that the listener hears, including the transmitted voice, the received voice, and the ambient noise, and the masky indicates (residual) echo.) The filter processor 22 estimates the impulse response of the echo path, generates a pseudo echo signal, and cancels the echo by subtracting it from the microphone output signal.

Each of a transmission input signal (transmission voice s), a residual echo e, and a reception input signal (ambient noise n, a reception signal x) is input to the masker / maski discriminating means 30 to discriminate the masker and masky. Do. Frequency analysis means A31
Performs frequency analysis (conversion to the frequency domain) of the received input signal,
Further, the frequency analysis means B32 analyzes the frequency of the (residual) echo e (converts it into a frequency domain). Then, based on the output signals of the frequency analyzing means A31, the frequency analyzing means B32, the masker and the masking discriminating means 30, the calculation of the masker, the calculation of the echo, the calculation of the masking threshold and the calculation of the desired echo suppression amount are performed for each frequency. To determine the desired echo suppression amount and output it to the loss insertion means. Loss insertion means 34
The filter inserts a desired loss into the line and suppresses (eliminates) a residual echo that cannot be eliminated by the adaptive filter processing unit 22 to a level that is inaudible by acting on the residual echo for each frequency. Then, the signal is frequency-synthesized by the frequency synthesizing means 35 to obtain a transmission output signal.

Referring to FIG. 4, a method of deriving a desired echo suppression amount will be described. The method of deriving the desired echo suppression amount is roughly divided into masker calculation, echo calculation, masking threshold calculation, and desired echo suppression amount calculation. First, the calculation of the masker and the calculation of the echo will be described. (Calculation of masker) a-1: Signals that can be the masker of the echo include the voice (x) of the own side, the voice (s) of the partner side, the ambient noise (n), and the like, and these signals are input. a-2: To these maskers, acoustic characteristics and transmission characteristics which are parameters of the loudspeaker environment are added. Specifically, since the echo is detected at the ear of the user, the acoustic characteristics (impulse response) from the source of all the maskers to the ear of the user who detects it are convolved with all the maskers. a-3: In addition, a transmission delay, a transmission loss, and a transmission line frequency characteristic, which are transmission characteristics, are added to a masker detected by a user through a communication network. In this way, the characteristics for various use environments are introduced as parameters, and the masking is calculated (echo calculation). B-1 to b-3: (residual) echo (masking) calculation is performed in the same manner. (Calculation of Masking Threshold) Next, calculation of a masking threshold is performed. In the present invention, a calculation method using 32 kHz sampling will be described. c-1: cut out all of the masker m _i to 1024 samples of the frame at the beginning. i represents discrete time. Apply a 1024-point hamming window to the sample cut in that frame.

Mw _i = m _i × [0.5−0.5 cos (2π (i−0.5) / 1024)] A sample subjected to a Hamming window is transformed into a frequency domain using an FFT. Calculate the polar representation of the values obtained by the FFT, the amplitude and phase components mr _w, and mf _w. c-2: Calculate energy (masker energy for each band) in each region where threshold calculation is performed. here,
The region values wlow _b (the lowest frequency component in the region), w
The combination of high _b (the highest frequency component in the area) is MPEG
According to the values shown in Table D.3a of the psychoacoustic model II of (ISO 11172-3). In addition, the combination table of them is shown in FIG. (In FIG. 5, when audio in a 16 kHz band is sampled at a sampling frequency of 32 kHz, if the audio is cut into a frame of 1024 samples, 513 samples are obtained for audio in a 16 kHz band.
FIG. 5 shows wlow, whigh and bark value bval of each area obtained by dividing the area into 1 to 49 areas (index b of area calculation). ) _Mb
= Σmr _w ² (w = low _b to w high _b ) c-3: Influence on other bands (a masker in one band (region) is replaced by another band (region)
Effect on the musky).

Using the spread function, the divided energy is convolved. mcb _b = Σm _bb × sprdng f (bval _bb , bval _b ) (from bb = 1
bmax) bval _b is the bark value at the center of the area, and the value is shown in FIG. The spread function is D.2. Of the psychoacoustic model II mentioned earlier.
Use the following function defined in section 3. tmpx = 1.05 (j−i) x = 8 × min ((tmpx−0.5) ² −2 (tmpx−0.5), 0) tmpy = 15.811389 + 7.5 (tmpx + 0.474) −17.5 (1.0+ (tmpx + 0.474) ² ) ^0.5 if (tmpy <-100) then (sprdngf (i, j) = 0) else (sprdngf (i, j) = 10 ^{(x + tmpy) / 10} ) c-4: Simultaneous masking threshold (with Calculation of instantaneous masking threshold) Calculate the denormalized energy.

Mn _b = mcb _b × rnorm _{b The} normalization constant is calculated by the following equation. rnorm _b = 1 / {sprdngf (bval _bb , bval _b )} (from bb = 0
bmax) The index b in the area calculation is converted into an index w on the frequency. _{mb w = mn b / (whigh} b -wlow b +1) c-5: temporal masking to calculate the (a certain moment of sound gives masking at the moment of the next) threshold. c-6: Comparison with audible level / determination of masking threshold Simultaneous masking threshold mb _w and audible energy threshold abs
Compared to thr _w , a larger value is set as a masking threshold thr _w .

Thr _w = max (mb _w , absthr _w ) where the value of the audible energy threshold absthr _w is MPEG
According to the values shown in Table D.4a of the psychoacoustic model II of (ISO 11172-3). These tables are shown in FIG. (FIG. 6 shows an absolute threshold (audible energy threshold absthr _w ) for the index.) (Calculation of echo energy) d-1: Calculate the energy of the echo to which the communication environment parameter is added. Cut the echo e _i to 1024 samples of the frame. A 1024-point hamming window is applied to the sample cut in the frame.

Ew _i = e _i × [0.5−0.5 cos (2π (i−0.5) / 1024)]
The sample with the hamming window is transformed to the frequency domain using FFT. The polar coordinate representation of the value obtained by the FFT is calculated, and the amplitude component and the phase component are set as er _w and ef _w . d-2: Calculate energy in each region where threshold calculation is performed.

E _b = Σer _w ² (w = wlow _b to whig
h _b) (optionally echo suppression amount calculation) e-1: taking the difference in masked previously determined threshold thr _w and echo energy e _b, it calculates the echo detection amount EDV. _{edv = e b -thr w e-} 2: finally determine the desired echo suppression amount dl of the present frame from the echo detection amount.

Dl = 0.8 × edv + 8.4 The above calculation is performed for each frame, and the maximum value of the statistically reliable section of the obtained data set is set as the desired echo suppression amount for each frequency band. To perform such calculations,
By substituting the parameters of the acoustic characteristics and the transmission characteristics according to the assumed voice communication environment, it is possible to obtain a desired echo suppression amount in various environments. Further, since the calculation is performed using known data of human hearing, the desired echo suppression amount obtained has generality and is an appropriate value for various users.

[0024]

As described above, according to the present invention, a masking threshold value is obtained from all masker signals for masking an echo signal, and a desired echo suppression amount is derived from the masking threshold value and the echo signal. Then, the voice communication environment is substituted as a parameter, and the desired echo suppression amount is derived using the data of the auditory characteristics. Therefore, it is possible to derive a general desired echo suppression amount suitable for various loudspeaker communication environments and for various users, and apply the derived desired echo suppression amount to an echo suppression method and an echo suppression device. Echo can be effectively suppressed.

[Brief description of the drawings]

FIG. 1 is a schematic diagram of a voice communication system.

FIG. 2 is a schematic diagram of an echo canceller.

FIG. 3 is a configuration diagram of the echo canceller of the present invention.

FIG. 4 is a diagram showing a flow of a method for deriving a desired echo suppression amount according to one embodiment of the present invention.

FIG. 5 is a table summarizing region values for threshold calculation.

FIG. 6 is a table summarizing audible energy thresholds.

[Explanation of symbols]

 1,3 Transmitting microphone 2,4 Receiving speaker 5,7 Transmitting signal amplifier 6,8 Receiving signal amplifier 10,11 Transmitter 12 Receiver 21 Echo canceller 22 Adaptive filter processing unit 23 Loss insertion processing unit 30 Masker And masking discriminating means 31, 32 Frequency analyzing means A, B 33 Desired echo suppression amount determining unit 34 Loss inserting means 35 Frequency synthesizing means

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Yutaka Kaneda F-term (reference) in Nippon Telegraph and Telephone Corporation 2-3-1 Otemachi 2-chome, Chiyoda-ku, Tokyo 5D020 AC01 5K027 BB03 DD10 5K046 AA01 BB00 CC29 HH01 HH79

Claims (9)

[Claims] 1. A masking threshold value for an echo signal is obtained from all masker signals for masking an echo signal, and a desired echo suppression amount is derived from the masking threshold value and the echo signal. A reverberation suppression method comprising suppressing an echo signal based on the echo signal. 2. The echo suppression method according to claim 1, wherein the echo signal used for deriving the desired echo suppression amount includes acoustic characteristics in a room where loudspeaker communication is performed and in the vicinity of an evaluator;
And a signal to which transmission characteristics of a communication network are added. 3. The echo suppression method according to claim 1, wherein all of the masker signals used for deriving the desired echo suppression amount are acoustic characteristics in a room where loudspeaker communication is performed and in the vicinity of an evaluator, and transmission characteristics of a communication network. The echo suppression method is characterized by using a signal to which an echo is added. 4. The echo suppression method according to claim 1, wherein the masking threshold for the echo signal is calculated as
A reverberation suppression method characterized by a masking threshold determined by a simultaneous masking phenomenon and a successive masking phenomenon by all maskers and an audible level determined by ambient noise and human auditory characteristics. 5. The echo suppression method according to claim 1, wherein the calculation of the desired echo suppression amount is obtained from an echo detection level determined from the echo signal and the masking threshold. . 6. A reverberation suppression apparatus comprising a desired echo suppression amount determining section and loss insertion means, wherein the desired echo suppression amount determining section determines a masking threshold value for an echo signal from all masker signals for masking the echo signal. The echo suppressor derives a desired echo suppression amount from the masking threshold and the echo signal, and the loss insertion means suppresses the echo signal based on the desired echo suppression amount. 7. The echo suppressor according to claim 6, wherein the echo signal used for deriving the desired echo suppression amount is a sound characteristic in a room where loudspeaker communication is performed and in the vicinity of an evaluator;
And a signal to which transmission characteristics of a communication network are added. 8. The echo suppressor according to claim 6, wherein all the masker signals used for deriving the desired echo suppression amount are acoustic characteristics in a room where loudspeaker communication is performed and in the vicinity of an evaluator, and transmission characteristics of a communication network. A reverberation suppressor characterized by using a signal added with a symbol. 9. The echo suppression apparatus according to claim 6, wherein the calculation of the masking threshold for the echo signal comprises: a masking threshold determined by a simultaneous masking phenomenon and a successive masking phenomenon by all maskers;
A reverberation suppressor, which is obtained from an audible level determined by ambient noise and human auditory characteristics. 10. The reverberation suppression apparatus according to claim 6, wherein the desired echo suppression amount is calculated from an echo detection level determined from the echo signal and the masking threshold.