JPH07303070A - Double talk detecting method - Google Patents

Abstract

PURPOSE:To exactly detect the presence/absence of double talk even when transmission level difference between a far terminal speaker and a near terminal speaker is comparatively large. CONSTITUTION:Concerning the double talk detecting method to be applied to an echo canceler provided at a telephone line network for transmitting the voice of the far terminal speaker to be sent through a digital line and the voice of the near terminal speaker from an analog line, prescribed analytic parameters are extracted respectively from a first voice signal corresponding to the voice of the far terminal speaker to be transmitted from the echo canceler to a hybrid circuit and a second voice signal corresponding to the voice of the near terminal speaker to be inputted through the hybrid circuit to the echo canceler, the correlation value of fluctuation in both the analytic parameters is calculated and based on this correlation value, the presence/ absence of double talk is detected. In this case, when the correlation value between both the analytic parameters gets lower than a prescribed threshold value, the generation of double talk is judged.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a double-talk detecting method used in echo cancellation in a mobile communication network or a long distance telephone line network.

[0002]

2. Description of the Related Art In a long-distance telephone line passing through a submarine cable or a communication satellite, generally, a subscriber line connected to both ends is a two-line type, and an intermediate long-distance transmission part is used for signal amplification. It is a line type. Similarly, mobile phones (or cellular phones).
In the mobile communication network using e)), the subscriber line of the fixed-side analog telephone is a two-line type, and the part from the terminal of the mobile telephone to the exchange or the like is a four-line type. In this case, 2 lines and 4
A hybrid circuit for performing 4-line / 2-line conversion is provided in the connection portion with the line. This hybrid circuit is designed to match the impedance of the 2-wire line, but it is difficult to always obtain good matching, so the received signal that arrives at the 4-wire input side of the hybrid circuit is 4-wire output. It leaks to the side and so-called echo occurs. Such an echo has a lower level than the voice of the talker and reaches the talker with a delay of a certain time, resulting in a call failure. The call disturbance due to such an echo becomes more remarkable as the signal propagation time becomes longer. In particular, in the case of mobile communication using a mobile telephone, various kinds of processing are performed in a wireless communication section up to an exchange or the like, so that a signal delay amount is large and a call failure due to echo is a particular problem.

There are echo suppressors and echo cancellers as devices for blocking the above echo. FIG. 2 shows a schematic configuration of an echo canceller used in a mobile communication network.
The echo canceller 1 shown here is provided before the hybrid circuit 2. In this figure, a subscriber of an ordinary analog telephone is called a near-end speaker, and a subscriber of a mobile telephone or the like is called a far-end speaker. Further, the far-end voice signal input to the echo canceller 1 is Rin, the far-end voice signal output from the echo canceller 1 is Rout, and the near-end voice signal input to the echo canceller 1 is Sin, from the echo canceller 1. The output near-end audio signal is indicated by Sout.

The echo canceller 1 shown in FIG. 2 comprises an echo path estimation / pseudo echo generation circuit 3, a control device 4, an adder 5 and a non-linear processing circuit 6. Here, the echo path estimation / pseudo echo generation circuit 3 uses the far-end voice input R
hybrid circuit 2 based on in and near-end voice input Sin
Of the response characteristic of (1), and thus the echo path (that is, the line through which the echo propagates) is estimated. Next, an expected echo (that is, a pseudo echo) from the hybrid circuit 2 is generated by the convolution operation of the estimation result and the far-end voice input Rin. In the adder 5, this pseudo echo is subtracted from the near-end voice input Sin, thereby canceling the echo. As the above-mentioned echo path estimation method, a learning identification method, which has a relatively small amount of calculation among adaptive algorithms and has good convergence characteristics, is used.

The following conditions are required as conditions for the above learning. There is a far-end voice output Rout at a level sufficient for the echo to be returned as the near-end voice input Sin, in other words, the far-end speaker is in a transmitting state. The near-end voice input Sin is composed of only echo (or echo and white noise), in other words, the near-end speaker is not in the transmitting state.

On the other hand, when the far-end talker is in a non-sending state and when the far-end talker and the near-end talker are simultaneously talking (this state is called double talk), false learning of echo path estimation is performed. Therefore, it is necessary to turn off the learning function.

By the way, a digital signal is transmitted through the transmission line, and D / A conversion (generally, μ is performed between the echo canceller 1 for processing the digital signal and the hybrid circuit 2 for converting into the analog line. -LAW conversion) is performed. Therefore, a non-linear characteristic relationship is established between the far-end voice output Rout and the near-end voice input Sin, and the echo path estimation / pseudo echo generation circuit 3
Complete echo cancellation cannot be performed only by linear calculation using, for example. Therefore, an echo component that cannot be canceled is generated. In order to eliminate such an echo component (referred to as “residual echo”), the nonlinear processing circuit 6
Is provided. The non-linear processing circuit 6 performs a non-linear switching operation. That is, when the near-end voice output Sout is composed of only echoes, that is, when only the far-end speaker is in the transmitting state (this case is referred to as "far-end talker single talk"), the near end. Audio output Sou
A switching operation is performed to prevent the transmission of t, or an operation is performed to replace the near-end voice output Sout with pseudo noise.

The control device 4 controls the echo path estimation / pseudo echo generation circuit 3 and the non-linear processing circuit 6. That is, the far-end party's non-talking state is detected, or double-talk is detected, and the learning function for echo path estimation is turned ON / OF.
The F control is performed, the far-end talker's single talk is detected, and the switching operation of the nonlinear processing circuit 6 is controlled.

[0009]

By the way, as a method of detecting double talk performed by the control device 4, the power ratio between the far-end voice output Rout and the near-end voice input Sin is conventionally used, and this is expected. Echo level (eg
Maximum echo level -6d defined by CCITT standard
It was determined that double talk occurred when the value exceeded B). However, this conventional double-talk detection method has a problem that the detection is delayed. That is, when there is not a sufficient level difference at the beginning of the occurrence of double talk, double talk is not detected, and double talk is detected only when the level difference exceeds a certain value. It has the drawback of not being performed well. Further, there is also a problem that double talk cannot be effectively detected even when the far-end speaker and the near-end speaker have greatly different transmission levels. That is, when the transmission level of the near-end speaker is lower than that of the far-end speaker,
The difference between the echo level and the transmission level of the near-end speaker does not sufficiently appear, and in that case, double-talk detection becomes difficult.

The low accuracy of the above double talk detection is
Erroneous learning of echo path estimation may be caused. When such erroneous learning occurs, not only the echo canceling function is deteriorated, but also an erroneous sound is transmitted to the far-end speaker by generating an erroneous pseudo echo.

The present invention has been made in view of such a background, and an accurate determination is made by determining the presence or absence of double talk based on the degree of correlation between the analysis parameters included in the far-end voice signal and the near-end voice signal. It is an object of the present invention to provide a double talk detection method for detecting double talk.

[0012]

In order to solve the above-mentioned problems, the invention according to claim 1 provides a telephone network for transmitting between voices sent through four lines and voices sent through two lines. In the double-talk detection method applied to the provided echo canceller, analysis parameters are extracted / analyzed from each of the first voice signal corresponding to the voices on the four-line side and the second voice signal corresponding to the voices on the two-line side. Then, a correlation value of fluctuations of both analysis parameters is calculated, and the presence or absence of double talk is determined based on the correlation value.

Further, the invention according to claim 2 is a double-talk detecting method applied to an echo canceller provided in a telephone network for transmitting between a voice transmitted through four lines and a voice transmitted through two lines. In, the analysis parameters are extracted / analyzed from each of the first voice signal corresponding to the voices on the four-line side and the second voice signal corresponding to the voices on the two-line side, and the correlation value of the fluctuation of the two analysis parameters is calculated. It is characterized by calculating, comparing the correlation value with a predetermined threshold value, and judging the presence or absence of double talk based on the comparison result.

A third aspect of the invention is characterized in that, in the first or second aspect of the invention, the analysis parameter is a voice pitch.

[0015]

According to the present invention, the presence / absence of double talk is determined based on the correlation value of the fluctuations of the predetermined analysis parameters included in the voice of the far-end speaker and the voice of the near-end speaker. Since it is decided to do so, the presence / absence of double talk can be accurately determined even when the difference between the above-mentioned two transmission levels is relatively large.

If a voice pitch is adopted as an analysis parameter, it can be realized by a combination of well-known circuits (claim 3).

[0017]

1 is a block diagram showing a schematic configuration of an apparatus embodying a double-talk detection method according to the present invention.
This device is provided in the control device 4. In this figure, 11 indicates an analysis parameter extraction / analysis unit. In this embodiment, the voice pitch is adopted as the analysis parameter used when detecting the double talk. This voice pitch indicates the vibration cycle of the human vocal cords. For example,
In the case of voiced sound, the vocal cords vibrate, and a constant fundamental period (or fundamental frequency) appears in the vibration. The analysis parameter extracting / analyzing unit 11 extracts such a voice pitch, which requires distinction between voiced sound and unvoiced sound and operations such as detection of the fundamental period in the case of voiced sound. Since various techniques have been proposed and are known,
Here, the detailed description is omitted.

The analysis parameter extracting / analyzing unit 11 analyzes the analysis parameters Pr (i) and Ps (i) from the far-end voice output Rout and the near-end voice input Sin (where i represents time and i = 0, 1, ...) are extracted. In this case, the input voice signal is measured and accumulated for a fixed time (for example, about 10 ms), and then the analysis parameters are sequentially extracted. That is, it is 1 until the extraction of the first analysis parameter.
It takes 0 ms, but after that, the analysis parameter can be extracted at intervals according to the sampling pitch.

Next, the correlation value calculation unit 12 receives the time correlation value kj of the input analysis parameters Pr (i) and Ps (i).
Is calculated according to Equation 1.

[0020]

[Equation 1]

Here, N _D represents a sample value of echo delay, and Ns represents a section in which the above correlation calculation is performed. Further, G _R and G _S are expressed by the following equations 2 and 3, respectively.
The above correlation value kj is the analysis parameter Pr (i), Ps.
The correlation degree of variation in a predetermined section (i.e., i (= j-Ns) to j) of (i) is shown.

[0022]

[Equation 2]

[0023]

[Equation 3]

Next, the double-talk detecting section 13 compares the correlation value kj calculated as described above with a predetermined threshold value K _TH, and detects / determines the presence / absence of double-talk based on the comparison result. In this case, when the condition of the following Expression 4 is satisfied, that is, when the correlation value kj is equal to or less than the threshold value K _TH ,
Judge that double talk has occurred.

[0025]

[Formula 4] kj ≦ K _TH

Here, the threshold value is set within the range of 0 ≦ K _TH <1. The threshold value is set to an optimum value through various experiments or actual measurements.

The correlation value kj is the far-end voice output Rout.
And the correlation degree of the variation of the analysis parameter of the near-end voice input Sin. For example, when only the far-end speaker is transmitting and an echo is generated, Rout corresponds to the voice of the far-end speaker and Sin corresponds to the echo. In this case, the degree of correlation between the two is relatively high. During this time, that is, while the echo is occurring,
The learning function of the echo path estimation will work. However, if both the far-end speaker and the near-end speaker are transmitting at the same time, that is, if double talk occurs, R
out corresponds to the voice of the far-end speaker, while Sin corresponds to the voice of the near-end speaker. In this case, the voice pitches of the two are different, and therefore the correlation between the analysis parameters of the two becomes low. Therefore, when the low degree of correlation is below a certain value (that is, a threshold value), it can be determined that double talk is occurring. The present invention has been made based on such a principle.

(Modification) In the above-mentioned embodiment, the double talk is detected by comparing the correlation degree with the threshold value. However, the double talk is detected not by the comparison with the threshold value but by the change situation of the correlation degree. You may.

In the above-mentioned embodiment, the voice pitch is used as the analysis parameter, but the gist of the present invention is not limited to this, and other parameters can be applied. For example, formants, bandwidths, etc. can be adopted as analysis parameters. Various methods have already been proposed and publicly known as methods for analyzing such voices, and therefore detailed description thereof will be omitted in this specification.

Further, although the above-mentioned embodiment is an embodiment in which the present invention is applied to signal transmission between a mobile telephone and a fixed telephone, the application of the present invention is not limited to this, and two lines and four lines are used. It can be applied to all communication networks that perform signal transmission between devices.

[0031]

As described above, according to the present invention, the analysis parameters of the voice signals respectively corresponding to the voices of the far-end speaker and the near-end speaker are extracted, and the double talk of the double talk is performed based on the correlation degree of the variation of the analysis parameters. Since the presence / absence is determined, it is possible to accurately determine the presence / absence of double talk even when there is a large difference in the transmission level between the far-end speaker and the near-end speaker.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of an apparatus embodying a double-talk detection method according to the present invention.

FIG. 2 is a block diagram showing a configuration of a conventional echo canceller.

[Explanation of symbols]

 1 Echo Canceller 2 Hybrid Circuit 3 Echo Path Estimation / Pseudo Echo Generation Circuit 4 Controller 5 Adder 6 Nonlinear Processing Circuit 11 Analysis Parameter Extraction / Analysis Section 12 Correlation Value Calculation Section 13 Double Talk Detection Section

Claims (3)

[Claims] 1. A double-talk detection method applied to an echo canceller provided in a telephone network that performs transmission between a voice transmitted through four lines and a voice transmitted through two lines, which corresponds to voice on the four-line side. The analysis parameter is extracted / analyzed from each of the first voice signal and the second voice signal corresponding to the voices on the two-track side, the correlation value of the fluctuations of the two analysis parameters is calculated, and a double value is calculated based on the correlation value. A double-talk detection method characterized by determining the presence or absence of talk. 2. A double-talk detection method applied to an echo canceller provided in a telephone network that performs transmission between a voice transmitted through four lines and a voice transmitted through two lines, which corresponds to voice on the four line side. The analysis parameter is extracted / analyzed from each of the first audio signal and the second audio signal corresponding to the audio on the two-line side, and the correlation value of the fluctuations of both analysis parameters is calculated, and the correlation value and the predetermined value are calculated. A method for detecting double talk, which comprises comparing with a threshold value and judging the presence or absence of double talk based on the comparison result. 3. The double talk detection method according to claim 1, wherein the analysis parameter is a voice pitch.