JPS5848113B2 - Onseito Shinboshiki - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a voice communication system that compresses voice information and utilizes lines efficiently.

In the transmission of voice and other signals, it is required to transmit as much information as possible over a certain communication line by band compression.

When considering voice communication, there are methods for band compression that reduce the redundancy of the voice signal itself, and methods for voice communication that take advantage of the fact that the proportion of voice on a line is relatively low and send multiple lines together using a small number of lines. There is a method to do this.

The former includes DPCM (DifferentialPC
M) t△M (De 1 ta M) , △ΣM(D
There is a method for reducing the pit rate using methods such as eltaSigma M), but this does not result in a large reduction in pits in voice communication, and there is a problem in that signals with large amplitudes and high frequency portions cause overload.

Another data compression method is the pocoder, which is highly effective in reducing pits, but has the problems of complicating the device and leaving unsatisfied sound quality.

Furthermore, with the recent advances in digital technology, communication systems using predictors have become possible, and various types have been proposed.

The human voice generator tube consists of a sound cord and a sound tract, and voice signals, especially vowels, are generated by the vibration of the vocal cords and are emitted through the vocal tract.

Approximately, the vocal cords can be regarded as an impulse generator and the vocal tract as a resonator, so if a voice signal is passed through a filter with appropriate characteristics (inverse characteristics of the vocal tract), an impulse can be obtained.

The vocal tract can be thought of as having several-order poles, and realizing the opposite characteristics of the vocal tract creates a zero circuit.

The zero circuit can be constructed using a transpersal filter,
Functionally, this can be thought of as consisting of a linear predictor and a subtraction circuit.

Compressing the amount of information is not difficult if it is possible to convert the audio signal into an impulse, and various methods have been proposed as described below.

One of the methods proposed by Attar et al. predicts the audio signal from past data using a transversal filter with variable coefficients (hereinafter referred to as VTF), and transmits the difference between the predicted value and the true signal (prediction error). .

Further, the VTF coefficient (prediction coefficient) is determined from the correlation coefficient of the audio signal and is transmitted simultaneously with the prediction error.

On the receiving side, the receiving side VTF coefficients are set from the sent prediction coefficients, and the sent prediction error is used as input to regenerate the original audio signal (B.S.Atal, M.
．． R. Schroeder t BSTJ.
Oct. 1970, vol49, 482, pp 1
973-1986).

Still another method is basically the same as the above, but extracts the pitch of the speech without sending the prediction error, and sends only information about the pitch (B.S., Atal
．． S. L. Hanauer t JASAA
ug. 1 97 1, vol50, A2, pp
637-655).

Although it is possible to reduce pits by sending prediction coefficients at regular intervals or every pitch period, the receiving side cannot stably reproduce the original audio signal unless the prediction coefficients are sent with extremely high accuracy.

Furthermore, these methods require an extremely large amount of calculations such as calculation of prediction coefficients, resulting in a large-scale circuit.

Another method using a predictor is the method proposed by Moye (L. S. Moye:
Self-Adaptive Filters:UK
Patent 41 1 8 4 6 5 3, J
A.N. ．． 1 9 6 7).

This is a self-adaptive transpersal filter (SATF) that automatically corrects prediction coefficients using prediction errors.
) is used.

In this method, the same SATF is used on both the transmitting and receiving sides, and both SAT
This method uses the fact that if the initial state of F is exactly the same, the original audio signal can be faithfully reproduced on the receiving side by transmitting only the prediction error.

This method can be implemented relatively easily in terms of circuits, but as described above, it is necessary that both SATFs match completely at all times, and even the slightest transmission error or calculation error is not allowed.

In other words, synchronization and complete error correction on both the transmitting and receiving sides are important issues.

Unlike the methods described above, TASI (
Time Assignment SpeechIn
terpolation) system.

(K. Bullington tM
．． M. Fraser tBSTJvo1.3
8, Mar. 1959, pp353-364) TASI
The system monitors a large number of lines, extracts only the audio section, and transmits it over a small number of lines.The receiving side regenerates the sent signal and sends it to the destination line.It is already used for long-distance telephone calls using submarine cables. It is commercially available on the line.

With the recent development of satellite communications, application to TDM lines has been proposed, and this has been proposed as TASI likeDS I (
Digital Speech Interpola
tion) system.

Although this method allows compression by 1/2, it has the serious drawback that the probability of the existence of voice increases and if the transmission capacity is exceeded, that is, if the overload point is reached, the voice will be cut off. .

Furthermore, as an attempt to reduce audio redundancy, SPE
C (Speech Predictive Enc.
odingCommunication) system has been proposed.

This is a voice sample sequence, and if samples of the same value or a small difference continue, no signal is transmitted, and on the receiving side, the same value continues when no transmission occurs.

In the SPEC system, when an overload point is reached, the signal to noise deteriorates, but the audio disconnection as in the TASI system can be avoided.

However, since the probability that adjacent sample values in an audio signal will be the same is not very high, a high compression rate cannot be expected.

An object of the present invention is to provide a voice communication system that enables high quality and large band compression with a simple device.

According to the present invention, an adaptive predictor predicts the next sample value from a past input audio sample sequence and corrects its own coefficients according to the prediction error; a transmitter having means for stopping data transmission when the prediction error is small and means for transmitting a mode signal indicating whether or not data has been transmitted; A receiver having an adaptive predictor and means for outputting received data when data is transmitted according to the received mode signal and outputting an adaptive predictor output signal when data transmission is stopped. A voice communication method is obtained which is characterized by a command from the beginning.

In the present invention, adaptive predictors capable of correcting coefficients are provided on both the transmitting and receiving sides.

When an audio signal is input on the transmitting side, the adaptive predictor continues to successively modify the coefficients so that the prediction error is minimized.

When an appropriate number of samples of the audio signal are input, the predictor has characteristics completely inverse to that of the vocal tract, and the prediction error appears in an impulse manner.

If the prediction error is large, the original audio sample is sent, and if the prediction error is small, the transmission is stopped.The prediction error is large, so transmission is stopped until the adaptive predictor has appropriate characteristics. The adaptive predictor on the receiving side also continues its adaptive operation, and approaches the inverse characteristics of the vocal tract to the same extent as the adaptive predictor on the transmitting side.

When the inverse characteristic is approached sufficiently, only unpredictable signals are sent in a Cainhals-like manner.

Any other signals? It is possible to predict with JllJ machine.

By the above-described operation, it is possible to reproduce the current signal without transmitting the coefficients of the predictor and even when a considerable portion of the audio signal is lost.

This will be explained in detail below using the drawings.

FIG. 1 shows a first embodiment according to the present invention.

Sample sequence XI of the audio signal input to the transmitter 107
, X2...xi are input to the adaptive predictor.

A transpersal filter type predictor is most suitable for use here.

In other words, the estimated value xi of the sample value xj can be obtained at j times.

Here, Ci (i-1 to N) are coefficients of the predictor, and N is the number (dimension) thereof.

Furthermore, the algorithm for modifying the coefficients is required to have high adaptation speed and simple calculations.

Examples of correction algorithms include the following.

Modify the coefficient C1 at time j as follows. Ru.

Specifically, at sampling time j, the adaptive predictor 110 calculates the predicted value Aj of the input sample xj at time j based on the sample sequence input up to time j-1.

The subtraction circuit 120 calculates the difference (prediction error) ej between the input sample xj and the predicted value j.

Prediction error ej is applied to adaptive predictor 110 to modify the coefficients.

The prediction error ej is further applied to a comparator 130.

If the prediction error is smaller than a certain value, the gate circuit 140 is driven to output the input sample xj as it is from the transmission terminal 102.

If the prediction error is smaller than a certain value, the gate circuit 140 blocks the input sample Xji and stops the signal output.

At this time, modification of the coefficients of the adaptive predictor is prohibited.

Also, the output of the comparator 130 is sent to the terminal 10 as a transmission mode signal.
Output from 3.

As shown in FIG. 2, only the unpredictable parts (indicated by black circles) of the digitized input sample sequence a of the original audio signal are transmitted, and the predictable parts (indicated by open circles) are not transmitted.

In the receiving section 108, the transmitted sample value sequence is input to an adaptive predictor 150.

At time j as well as at the transmitter, the adaptive predictor 150
calculates the estimated value j' of xj, and the difference (prediction error) ej from the true value xj is added to the adaptive predictor 150.

The transmission mode signal is input from terminal 105 and sent to register 17.
Stored at 0.

When audio sample xj is transmitted, gate 180 is driven to output xj and at the same time the coefficients of adaptive predictor 150 are modified.

If the audio sample xj has not been transmitted, the adaptive predictor 150 output j', which is the predicted value of xj, is sent to the output terminal 10.
Output from 6.

At this time, the adaptive predictor 150 is not modified.

Also, j' is stored as an input to the adaptive predictor 150 instead of xj.

Therefore, by stopping correction of coefficients in the adaptive predictor on the transmitting side when data transmission is stopped, it becomes easy to match the predictor coefficients on both the transmitting and receiving sides.

Further, the same effect can be obtained by inputting the transmitting side predicted value j to the transmitting side predictor instead of xj.

If an unpredictable signal is transmitted in this manner, the original audio signal can be reproduced as shown in FIG. 2d even if a predictable signal is not transmitted.

In the present invention, unlike Moye's method, it is not necessary to synchronize the coefficients of the adaptive predictors on the transmitting side and the receiving side.

Even if the coefficients of both predictors deviate, the receiving side will make its own and successive corrections based on the signals sent, so it will match the transmitting side in a short time.

If the threshold value in the comparator 130 is set larger than the line noise, no signal is sent when there is no voice.

Therefore, even if the average probability of signal transmission in a voice section is 0.5, the proportion of voice present in the voice communication line is about 0.3, so it is only necessary to transmit about 15 of the total samples.

It is also clear from this that the present invention has the ability to eliminate line noise.

Furthermore, the threshold value in comparator 1.30 is set to the input sample xj
Even greater compression ratios are possible without degrading quality if designed according to the amplitude of .

Input samples undergo nonlinear transformation (A-law or M-law, etc.)
If so, the threshold value of the comparator 130 can be easily set using that segment number.

As described above, according to the present invention, data compression is possible with almost no loss in audio quality.

FIG. 3 shows a second embodiment according to the present invention, in which PCM/T
This was used for the DM system.

The N-channel time-multiplexed PCM signal has a format as shown in FIG. 4e.

S is a synchronization signal, and D1 is a sample value of channel 1, which has 8 nonlinear pits.The most common format is to have D1 to DN as one frame and have a frame period of 125 μsec.

When data Dn of channel n is input at sampling time j, the adaptive predictor 310 reads data of channel n up to j-1 time from the auxiliary memory 311, calculates a predicted value, and corrects the predictor coefficients. .

The transmission control unit 320 obtains a prediction error from the predicted value and the input sample value, determines its magnitude, and stores it as a transmission mode signal.

After performing the same processing for channels 1 to N, the transmission control unit 320 outputs a signal indicating the transmission mode of each channel as shown in A in FIG. A sample is output as shown in B in Fig. 4f.

On the receiving side, the received signal f is demultiplexed by a demultiplexer 380.
The signal is divided into two, A and B, and input to the reception control unit 370 and buffer memory 360, respectively.

The reception control unit 370 corrects the adaptive predictor 350 for the channel to which the audio sample was sent according to A of the received signal f, reads out the transmitted audio sample from the buffer memory 360, and outputs it to the terminal 306 through the multiplexer 381.

In the case of a channel on which no transmission was performed, the output of the adaptive predictor 350 is selected by the multiplexer 381 and output.

Through the above operations, an N-channel multiplexed PCM signal is restored at the terminal 306 as shown in FIG. 4g.

By performing multiple processing as in this embodiment, the cost per channel can be significantly reduced.

As explained above, according to the present invention, it is possible to compress data at low cost, with simple hardware, and maintaining high quality.

[Brief explanation of drawings]

FIG. 1 shows a first embodiment of the present invention, in which 107 is a transmitter, 108 is a receiver, 101 is an audio input terminal, and 102 is a receiver.
is an audio signal transmission terminal, 103 is a transmission mode output terminal, 1
10 is an adaptive predictor, 120 is a subtracter, 130 is a comparator, 140 is a gate circuit, 104 is an audio signal receiving terminal, 1
05 is a transmission mode input terminal, 106 is an audio signal output terminal, 150 is an adaptive predictor, 160 is a subtracter, 170 is a register, and 180 is a gate circuit. Figure 2 shows the waveforms a, b, c, and d in Figure 1, where a is the original audio signal, b is the signal to be transmitted, C is the transmission mode signal, and d is the signal reproduced by the receiver. . FIG. 3 shows a second embodiment of the present invention used in a TDM system, in which 307 is a transmitter, 308 is a receiver,
301 is an N-channel time-multiplexed audio signal input terminal, 302 is a transmission terminal, 310 is an adaptive predictor, 311
3 is an auxiliary memory, 320 is a transmission control unit, 340 is a buffer memory IJ, 341 is a multiplexer, 304 is a receiving terminal, 3
06 is an N-channel time-multiplexed audio signal output terminal, 350 is an adaptive predictor, 351 is an auxiliary memory, 360
370 is a reception control unit, 380 is a demultiplexer, and 381 is a multiplexer. FIG. 4 shows the data format at e, f, and g in FIG.

Claims (1)

[Claims] 1. An adaptive predictor that prepares the next sample value from a past input audio sample sequence and corrects its own coefficients according to the prediction error, and transmits the input sample as data when the prediction error is large, and a transmitting unit having means for stopping the transmission of data if the data is transmitted, and means for transmitting a mode signal indicating whether data has been transmitted; and an adaptive predictor having the same characteristics as the adaptive predictor of the transmitting unit. and means for outputting the received data when data is transmitted, and outputting an adaptive predictor output signal when data transmission is stopped, according to the received mode signal. A voice communication method characterized by: