JPH08186525A - Echo suppresser circuit - Google Patents

Abstract

PURPOSE: To provide an echo suppresser circuit with which a head cutting phenomenon is not generated. CONSTITUTION: LPC analyzing circuits 10 and 20 perform linear predictive encoding to audio data to be inputted. Based on the encoded result of the LPC analyzing circuits 10 and 20, LPC calculating circuits 12 and 18 generate LPC kepstrums L1kj and L2k and (n) pieces of the latest LPC kepstrums L1kj are stored in a ring buffer 14. A square distance calculating circuit 16 calculates a square sum Sj of the LPC kepstrums L1kj and L2k . When echo is generated in a communication line, the square sum Sj gets less than a threshold value. In this case, the square distance calculating circuit 16 judges the audio data on the reception side as the echo of audio data on the transmission side and enlarges the amount of attenuation to audio data from the communication line to the reception side by controlling an attenuation circuit 22 so that audio data can not leak from the communication line to the reception side.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo suppressor circuit used for suppressing echo generated in a communication line.

[0002]

2. Description of the Related Art Conventionally, an echo suppressor circuit 8 shown in FIG. 2 has been used to suppress an echo generated in a communication line for transmitting voice data. As shown in FIG.
The echo suppressor circuit 8 includes attenuators 80 and 84 and a level comparison circuit 82. The level comparison circuit 82 compares the level of voice data input from the receiving side with the level of voice data input from the line. However, it is configured to suppress the echo by increasing the attenuation amount of the audio data having the lower audio data level.

That is, the level comparing circuit 82 increases the attenuation amount of the attenuator 80 when the level of the voice data input from the receiving side is lower than the level of the voice data input from the line, and the level comparing circuit 82 from the receiving side. When the level of the input voice data is higher than the level of the voice data input from the line, the attenuators 80 and 84 are controlled so as to increase the attenuation amount of the attenuator 84.

[0004]

However, when the echo suppressor circuit 8 is used, a so-called "cut-out" phenomenon is uttered when double talk occurs in which both parties in the call utter simultaneously. In other words, the voice output portion of the caller whose speech is slightly delayed becomes lower than the voice level of the other party in the echo suppressor circuit 8 of the other party, so that the attenuation amount becomes large and the other party cannot hear it. There is.

The echo suppressor circuit according to the present invention has been made in view of the above-mentioned problems of the prior art,
An object of the present invention is to provide an echo suppressor circuit that can eliminate the phenomenon of head loss during double talk.

[0006]

In order to achieve the above object, an echo suppressor circuit according to the present invention includes time domain first voice data transmitted in a predetermined direction in a communication line and frequency domain first voice data. A frequency data storage circuit for sequentially converting the first frequency data into a predetermined number, and a frequency data storage circuit for storing a predetermined number of the first frequency data from the latest one in the predetermined direction in the communication line. A second frequency data generation circuit for sequentially converting second time data in the time domain, which is transmitted in the opposite direction and is included in a predetermined frame, into second frequency data in the frequency domain; and the frequency data storage circuit. A similarity discriminating circuit for discriminating the similarity between each of the first frequency data stored in the circuit and the second frequency data, and a similarity discriminating circuit for similarity. While identifying, and an attenuation circuit to increase the attenuation amount of the second audio data.

Preferably, the first frequency data generation circuit performs LPC analysis on the first audio data included in a predetermined frame to sequentially generate a first LPC cepstrum, and the first LPC cepstrum is generated. Each cepstrum is the first frequency data, and the second frequency data generation circuit performs LPC analysis on the second audio data included in a predetermined frame to sequentially generate second LPC cepstrum, The second LPC cepstrum is used as the second frequency data, and the similarity identifying circuit corresponds to each of the first LPC cepstrum stored in the frequency data storage circuit and each of the first LPC cepstrum. When the sum of squares of the differences from each of the second LPC cepstrums is sequentially calculated, and the sum of squares becomes equal to or less than a predetermined value, Identify the similarity to.

[0008]

The first frequency data generation circuit sequentially performs linear predictive coding (LPC analysis) on the audio data transmitted from the transmission circuit of the telephone in the direction of the communication line and included in a predetermined frame for conversion. Then, the first frequency data (acoustic parameter; LPC cepstrum) is generated by converting to frequency domain data. The frequency data storage circuit stores the first frequency data for 50 frames, for example, from the latest frequency data to the frequency data 50 frames before. The second frequency data generation circuit sequentially converts the voice data, which is transmitted from the communication line toward the receiving device of the telephone and included in a predetermined frame for conversion, into LPC analysis to convert it into frequency domain data. , Second frequency data is generated.

The similarity identifying circuit includes the first frequency data stored in the frequency data storage circuit and the second frequency data stored in the frequency data storage circuit.
Identify the similarity with the frequency data of. Specifically, the first sum of squares of the difference between the first frequency data, that is, each of the first LPC cepstrum and each of the corresponding second LPC cepstrum, is sequentially calculated, and the sum of squares is equal to or less than a predetermined value. If it is, it is determined that there is a similarity between the two, and if there are a predetermined number of consecutive frames in which the sum of squares is a predetermined value or more, it is determined that there is no similarity. If the first frequency data and the second frequency data are similar to each other, the second frequency data can be determined to be an echo component of the first frequency data. While identifying the similarity, the attenuation amount of the second audio data is increased to prevent the audio data from passing to the receiving side.

[0010]

Embodiments of the present invention will be described below. Figure 1
It is a figure which shows the structure of the echo suppressor circuit 1 which concerns on this invention. The echo suppressor circuit 1 is used to suppress the echo on the digital communication line. As shown in FIG. 1, the echo suppressor circuit 1 includes an LPC analysis circuit 10,
20, LPC calculation circuits 12 and 18, ring buffer 1
4. The square distance calculation circuit 16 and the attenuation circuit 22.

The LPC analysis circuit 10 sequentially performs linear predictive coding (LPC analysis) on the voice data input from the transmission circuit (transmission side) of the telephone for each 5 msec frame used for encoding, and encodes the voice data. The result is output to the LPC calculation circuit 12 and the voice data is sent to the communication line. The LPC analysis circuit 20 linearly predictively encodes the voice data input from the communication line for each 5 msec frame used for encoding, and outputs the encoding result as an LP.
The audio data is output to the C calculation circuit 12 and also transmitted to the reception circuit (reception side) of the telephone.

The LPC calculation circuit 12 is an LPC analysis circuit 1
12th-order LPC cepstrum L _1kj from the coding result of 0
(K = 1 to 12, j is 1 to n) is calculated and sequentially output to the ring buffer 14. LPC calculation circuit 18
Is a 12th-order LP from the encoding result of the LPC analysis circuit 20.
The C cepstrum L _2k is calculated and sequentially output to the square distance calculation circuit 16. The order of the LPC cepstrum calculated by the LPC calculation circuits 12 and 18 is set to 12 because the order is a value necessary and sufficient for the analysis processing of voice data.

The ring buffer 14 is an LPC calculation circuit 1
The memory content is updated each time a new LPC cepstrum is input from 2, and the LPC calculated by the LPC calculation circuit 12 is generated.
Store n (eg, 50) cepstrum from the last frame to the frame n frames before. The squared distance calculation circuit 16 performs the calculation shown below to calculate the sum of squared differences S between the LPC cepstrum L _{1k j} and the LPC cepstrum L _2k.
Calculate _j . The sum of squares S _j is the LPC cepstrum L
It calculates the distance between the _1kj and LPC cepstrum L _2k.

[0014]

[Equation 1]

Further, the squared distance calculation circuit 16 judges all the values of the sum of squared S _j , and if they are equal to or less than a predetermined threshold value, they are similar to the voice data on the transmitting side and the voice data on the receiving side in the frame j. If the sum of squares S _j is larger than a predetermined threshold value, it is determined that the voice data on the transmitting side and the voice data on the receiving side have no similarity. That is, when the square distance calculation circuit 16 identifies that there is similarity between the voice data on the transmitting side and the voice data on the receiving side, the voice data on the receiving side is an echo of the voice data on the transmitting side. If it is determined that there is no similarity between the voice data of the transmitting side and the voice data of the receiving side, it is determined that the voice data of the receiving side is not an echo of the voice data of the transmitting side.

In the ring buffer 14, 50
LPC cepstrum L for frame _1kjRemember
Then, the echo suppressor circuit 1 delays up to 250 msec.
Is available. The square distance calculation circuit 16 makes the sound on the receiving side
Judge that the voice data is not an echo of the voice data of the sender
In this case, the square distance calculation circuit 16 uses the attenuation circuit 22.
The audio data that is controlled and output to the receiving side is passed through without being attenuated.
To pass. On the contrary, the square distance calculation circuit 16 causes the reception side sound
Judge that the voice data is an echo of the voice data of the transmitting side.
However, if that state continues for a predetermined number of frames,
If it is detected that the connection has continued, the audio output to the receiving side
Attenuate the data sufficiently to prevent leakage from the communication line to the receiving side.
Try to stay

The operation of the echo suppressor circuit 1 will be described below. When the echo above a certain level is generated on the communication line When the echo is generated on the communication line, LP from the transmission side
A part of the voice data sent to the communication line via the C analysis circuit 10 will be returned to the LPC analysis circuit 20. The LPC analysis circuit 10 sequentially linearly predictively encodes the audio data input from the transmission side for each frame described above, and outputs the encoding result to the LPC calculation circuit 12. The LPC calculation circuit 12 generates the LPC cepstrum L _1kj for each frame described above based on the coding result of the LPC analysis circuit 10, and stores the latest n (50) L _ring cepstrums in the ring buffer 14.

On the other hand, the LPC analysis circuit 20 sequentially outputs the voice data input from the communication line for each frame described above.
Linear prediction coding is performed, and the coding result is output to the LPC calculation circuit 18. The LPC calculation circuit 18 generates the LPC cepstrum L _2k for each frame described above based on the coding result of the LPC analysis circuit 20, and the squared distance calculation circuit 1
Output to 6.

The squared distance calculation circuit 16 calculates the sum of squares S _j by performing the calculation shown in Expression 1. Since an echo is generated in the communication line, the sum of squares S _j remains below the threshold for a predetermined number of frames or more when the other party is not speaking. Therefore, the square distance calculation circuit 16 determines that the voice data on the transmitting side and the voice data on the receiving side have similarities, and determines that the voice data on the receiving side is an echo of the voice data on the transmitting side. Become. The square distance calculation circuit 16, which has determined that the voice data on the reception side is an echo of the voice data on the transmission side, controls the attenuation circuit 22 to increase the attenuation amount for the voice data from the communication line to the reception side. Prevent audio data from leaking to the receiving side.

Echoes above a certain level occur on the communication line
If the communication partner is not speaking, or if an echo is not generated on the communication line, or if the communication partner is speaking, the LPC analysis circuit 1 from the transmitting side.
The voice data sent to the communication line via
The voice data input to the analysis circuit 20 will be different.

Therefore, since the value of the sum of squares S _j calculated by the squared distance calculation circuit 16 is equal to or larger than the threshold value, the squared distance calculation circuit 16 has no similarity between the voice data on the transmitting side and the voice data on the receiving side. And it is determined that the voice data on the reception side is not an echo of the voice data on the transmission side. The squared distance calculation circuit 16, which has determined that the voice data on the reception side is not an echo of the voice data on the transmission side, controls the attenuation circuit 22 to pass the voice data from the communication line to the reception side without attenuation.

When double talk is occurring For example, when the call partner speaks while the user of the telephone connected to the communication line via the echo suppressor circuit 1 speaks, Before or after
The voice data sent from the transmitting side to the communication line via the LPC analysis circuit 10 remains different from the voice data input to the LPC analysis circuit 20. Therefore, the value of the sum of squares S _j calculated by the square distance calculation circuit 16 is equal to or more than the threshold value, as in the case where the echo above a certain value is not generated in the communication line or the communication partner is speaking. Therefore, the square distance calculation circuit 16 determines that the voice data of the transmitting side and the voice data of the receiving side have no similarity, and determines that the voice data of the receiving side is not an echo of the voice data of the transmitting side. Become.

The squared distance calculating circuit 16, which has determined that the voice data on the receiving side is not an echo of the voice data on the transmitting side, controls the attenuating circuit 22 to pass the voice data from the communication line to the receiving side without attenuating. Therefore, the user of the telephone to which the echo suppressor circuit 1 is connected does not hear the voice of the other party as cut off. If the other party is using a telephone connected to the communication line via the echo suppressor circuit 1, even if one of the telephones starts speaking while the other is speaking, the head of both telephones will be cut off. The phenomenon does not occur.

As described above, according to the echo suppressor circuit 1 of the present invention, the head-cut phenomenon does not occur during double talk. By connecting the echo suppressor circuit 1 to the communication line via the digital / analog conversion circuit and the analog / digital conversion circuit, echo on the analog communication line can be effectively suppressed.

Further, the LPC analysis circuits 10 and 20 are circuits for analyzing voice data by another analysis method, for example, FF.
Even if the circuit is replaced with a T circuit, the same effect can be obtained. Further, the echo suppressor circuit 1 has a so-called half echo suppressor configuration in which only the audio data from the communication line is attenuated by the attenuating circuit 22, but the echo suppressor circuit 1 is configured as a transmitting side and a receiving side. A so-called full echo suppressor configuration may be added by adding an echo suppressor circuit which is the reverse of the above. Moreover, it does not matter whether each part of the echo suppressor circuit 1 is constituted by independent hardware or is realized in the form of software on a computer.

[0026]

As described above, according to the echo suppressor circuit of the present invention, it is possible not only to effectively suppress the echo generated in the communication line, but also to cause a double problem which has been a problem in the conventional echo suppressor circuit. It is possible to eliminate the phenomenon of head loss during talk.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of an echo suppressor circuit according to the present invention.

FIG. 2 is a diagram showing a configuration of a conventional echo suppressor circuit.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Echo suppressor circuit, 10, 20 ... LPC analysis circuit, 12, 18 ... LPC calculation circuit, 14 ... Ring buffer, 16 ... Square distance calculation circuit, 18 ... LPC calculation circuit

Claims (2)

[Claims] 1. A first frequency data generation circuit for sequentially converting first voice data in the time domain transmitted in a predetermined direction in a communication line into first frequency data in the frequency domain, and the first frequency data generation circuit. A frequency data storage circuit that stores a predetermined number of frequency data from the latest one, and a second time domain portion that is transmitted in a direction opposite to the predetermined direction in the communication line and is included in a predetermined frame. A second frequency data generation circuit that sequentially converts audio data into second frequency data in the frequency domain, each of the first frequency data stored in the frequency data storage circuit, and the second frequency A similarity identifying circuit for identifying a similarity with data; and an attenuating circuit for increasing an attenuation amount of the second audio data while the similarity identifying circuit identifies a similarity. Echo suppressor circuit. 2. The first frequency data generation circuit performs LPC analysis on the first audio data included in a predetermined frame to sequentially generate a first LPC cepstrum, and the first LPC cepstrum is generated. The second frequency data generation circuit performs LPC analysis on the second audio data included in a predetermined frame to sequentially generate second LPC cepstrum, respectively. Two
LPC cepstrum is used as the second frequency data, and the similarity identifying circuit includes the first LPC cepstrum stored in the frequency data storage circuit and the second LPC cepstrum corresponding to the second LPC cepstrum. 2. The echo suppressor circuit according to claim 1, wherein the sum of squares of the differences from the respective LPC cepstrums are sequentially calculated, and the similarity is identified when the sum of squares becomes a predetermined value or less.