JPS63124636A - Pseudo signal insertion system in voice semiconductor system - Google Patents

Abstract

PURPOSE:To improve the voice quality of listener and keep the naturaility of talking by using a synthesis filter driven by a coded parameter of a frame before missing and noise so as to synthesize a pseudo voice and interpolate the missing part. CONSTITUTION:When the silence state is detected from a signal at a terminal 3, a changeover switch 5 selects a D/A converter 6 to the position of a power adjuster 14. An LPC analyzer 10 analyzes an envelope of a spectrum of background noise at the voice presence before the voice signal is silent and obtains a prediction coefficient for a predicting device 12 for a short time in the short period synthesis filter 11 to set the result for a short time prediction device 12. Then the noise from the noise generator 4 is inputted to the filter 11 and a noise having a spectrum similar to the spectrum of the background noise is outputted and given to the power adjuster 14. As a result, the pseudo noise having a spectrum and power similar to the background noise is outputted from the power adjuster 14.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to an audio transmission system that transmits an audio signal to a transmission path only in a time period in which the audio signal exists, and particularly relates to a time period in which the signal is not transmitted. A pseudo signal insertion method in a voice transmission system that is effective for improving the perceptual quality of calls (%) In conversation-type communications such as voice lines, when one subscriber is talking, the other subscriber's Since the subscriber acts as a listener, the time efficiency of one-way lines is generally significantly lower.

Therefore, various techniques have been proposed to improve line usage efficiency.

A typical method for improving line usage efficiency is the time allocation multiplexing method (hereinafter referred to as
There is also a digital audio insertion method (hereinafter referred to as rTAsI method) used in digital lines.
DsI method) or one channel, one carrier method (r5CPCJ method) in which a carrier is allocated only when a voice signal is present (hereinafter referred to as "speech state").
There is also a packet audio transmission method that transmits audio in packet units. These TAS I systems, DS1 systems, etc. multiplex lines by efficiently using transmission lines (output lines) that are smaller in number than the number of input lines connected to the subscriber side.

By the way, DSI J? In the 3SCPC system, when the audio from the transmitting side is interrupted and there is no sound, that is, when the voice state changes from a sound state to a silent state, after a certain hangover time (usually about 15 seconds), the input card link corresponding to the transmission path is connected. The connection is terminated, and no signal is transmitted to the receiving side, resulting in an unnatural conversation. To solve this problem, conventionally, when there is silence, the receiving side inserts white noise from a white noise generator into the corresponding trunk, making it appear as if the transmission line is still connected, and transmitting it to the receiver. We try to keep the call natural.

FIG. 1 is a block diagram of an example of a noise insertion method in a conventional digital voice transmission system. Based on the sound/silence state signal sent to the terminal 3, if there is a sound, the selector switch 5 is controlled, the D/A converter 6 is connected to the decoder 2, and the signal is transmitted from the encoder on the transmitting side. The transmitted encoded signal (for example, a digital signal encoded by PCM, ADPCM, APC (adaptive predictive coding), etc.) is input to the receiving side input terminal 1. This received encoded signal is decoded by a decoder 2, then converted to an analog signal via a D/A converter 6, and then converted to an analog signal by an analog filter 7.
The signal is restored to the original analog audio signal and output from the output terminal 8.

On the other hand, when the sound/silence state signal indicates silence, the noise generator 4 and the D/A converter 6 are connected via the changeover switch 5, and the white noise from the noise generator 4 is converted into a D/A converter. Vessel 6
The naturalness of the call is maintained by transmitting it to the receiver from the terminal 8 via the analog filter 7.

However, the signal transmitted during the hangover time, which is the active state immediately before changing to the silent state, is background noise such as room noise. For this reason, in these conventional methods, the white noise inserted on the receiving side is different from the background noise level and its spectrum, so even if white noise is inserted, it gives a strange feeling to the receiver, and the unnaturalness of the call is not so great. Not improved.

In addition, in audio transmission systems such as the DSI method and the packet audio transmission method, a part of the audio signal is not received on the receiving side due to a significant increase in the audio operation rate, which locks out the input quadratic rank, or because packets are dropped during transmission. occurs. Conventionally, when a part of the audio signal is missing, methods such as making silence or inserting white noise during the missing audio signal, or simply manipulating the missing audio packet from the previous frame, etc. method was used. However, both methods have resulted in significant auditory deterioration and have not been able to significantly improve call voice quality. In particular, when an audio packet from one frame before is used repeatedly, a memory is required to store audio information from one frame before.

As described above, in the past, there was a problem in that the pseudo signals inserted when there was silence or when a voice packet was missing caused audible deterioration, making it impossible to maintain the naturalness of the conversation.

(Objects and Features of the Invention) The present invention has been made to solve the problems of the prior art described above. The purpose of this invention is to provide a pseudo signal insertion method for a voice transmission system that can easily create a pseudo signal.

A feature of the present invention is that when an audio signal is missing or when there is no sound, the power included in both the audio signal in the active state immediately before the silent state and the signal within the hangover time, or at least a part of the hangover time, and the input signal on the transmitting side. The present invention uses spectrum information to adjust the level and spectrum of the noise output from the noise generator on the receiving side to create and insert a pseudo signal on the receiving side.

(Structure of the invention) The present invention will be described in detail below using the drawings.

In the following description, the same components as those of the conventional example (FIG. 1) are given the same reference numerals to avoid redundant description.

(Embodiment 1) FIG. 2 is an embodiment of the present invention, and is a schematic diagram of a pseudo noise signal insertion method in the voice state of a voice transmission system. This embodiment shows an example in which information on the receiving side is processed based on decoded digital audio. 9 is a power calculation circuit that calculates the background noise power for all or part of the hangover time in a sound part when changing from a sound state to a silent state;
10 is L that analyzes the spectrum of background noise within that interval.
PG (Linear Predictive Codi)
ng) The analyzer 11 is a short-time synthesis filter that adjusts the envelope of the noise spectrum based on the short-term prediction coefficients obtained by the LPC analyzer 10, and is composed of a short-time predictor 12 and an adder 13. The output of the synthesis filter 11 is input to the power regulator 14, which adjusts the noise power to be equal to the background noise power based on the background noise power information from the power calculation circuit 9. Convert to analog signal via.

Next, the operation will be explained. A digital signal input to an input terminal 1 is decoded by a decoder 2 to become a digital audio signal. When the sound/silence state signal from the terminal 3 indicates that there is sound, the decoder 2 is connected to the D/A converter 6 via the changeover switch 5, and outputs analog audio from the terminal 8 via the analog filter. is restored. On the other hand, if a silent state is detected from the signal at the terminal 3, the changeover switch 5 connects the D/A converter 6 to the power regulator 14.

The LPC analyzer 10 analyzes the envelope of the spectrum of the background noise at the time of speech before the decoded speech signal becomes silent, that is, during the hangover time, and makes a prediction for the short-term predictor 12 in the short-term synthesis filter 11. Find the coefficient,
Set short-term prediction to 12. At this time, coefficients obtained by weighting the coefficients obtained by these calculations may be set. Third
The figure shows an example of the configuration of the fourth-order short-time synthesis filter 11. D I"" in the short-term predictor 12 of FIG.
D < means 1 sample delay memory, α1 to α4
is the prediction coefficient. 15 is an adder which provides a predicted value as a composite value of the outputs of the memories D, -D loaded with coefficients α1 to α4. The noise from the noise generator 4 is input to the short-time synthesis filter 11, which outputs noise having a spectrum similar to the spectrum of background noise, and inputs it to the power regulator 14.

The power regulator 14 adjusts the power based on information from the power calculation circuit 9 that calculates the power of background noise so that the power of the signal input from the short-time synthesis filter 11 is the same as the power of the background noise.

As a result, pseudo noise having a spectrum and power similar to background noise is output from the power regulator 14, and the changeover switch 5
is sent to the D/A converter 6 via the analog filter 7.
The signal is converted into an analog signal via the terminal 8 and transmitted to the receiver as noise inserted during no call.

By generating and inserting pseudo-noise having the same spectrum and power as the transmitted background noise on the receiving side as shown in the method described above, it is possible to significantly improve the sense of discomfort during a silent state.

In addition, in the above-mentioned Example 1, the signal power and spectrum information of the background noise were extracted based on the decoded audio signal.
Next, adaptive predictive coding (hereinafter referred to as rAPC coding)
The case where this method is used will be explained. As for the noise insertion method on the receiving side in the APC encoding method, the power calculation circuit 9 and the LPC analyzer 10 are not required as in the first embodiment described above, and the structure inside the APC decoder is used in combination, resulting in a very simple structure. We will show what can be achieved with

Further, in the next embodiment, a configuration function for a pseudo-speech interpolation method in the case where one frame of encoded signal is completely lost due to packet loss can also be provided.

(Embodiment 2) FIG. 4 is a second embodiment according to the present invention, and is a schematic diagram of a pseudo noise signal insertion method when using the APC encoding method.

In the APC encoding method, the encoder on the transmitting side processes the long-term correlation of the input signal, for example, the correlation corresponding to the pitch period due to the repetition of a similar waveform for each pitch period of voiced sounds in the audio signal. The short-time correlation representing the short-time spectral envelope is removed using a time predictor, and the short-time correlation representing the short-time spectral envelope is removed using a short-time predictor.The resulting signal, called a residual signal, is quantized and encoded before being transmitted.

The encoded signal is received at the terminal 1, and the residual signals encoded with the encoding parameters are demultiplexed by the demultiplexer circuit 16. The APC decoder on the receiving side determines the step size of the inverse quantizer 18 based on the information on the residual signal power from the residual signal power decoder 17, and converts the inverse quantization of the received residual signal into an inverse quantizer. This is done using a converter 18. Sound from terminal 3/
When it is determined from the silence state signal that the current frame is a sound frame, the adder 22 and the inverse quantizer 1 are connected to each other via the changeover switch 5.
8 is connected, and the switch 23 is further closed. As a result, the dequantized residual signal is input to a long-term synthesis filter 26 consisting of a long-term predictor 21 and an adder 22, and then a short-term synthesis filter 26 consisting of a short-term predictor 24 and an adder 25. It is input to the synthesis filter 27. These configurations are A
The reconstructed digital voice is obtained as the output of the short-time synthesis filter 27, which operates as a PC decoder. Terminal 8 via D/A converter 6 and analog filter 7
Analog audio is sent out. The long-term prediction coefficient decoder 19 decodes the transmitted long-term prediction coefficients, and sets the decoded coefficients in the long-term predictor 21.

Further, the short-time prediction coefficient decoder 20 similarly decodes the transmitted short-time prediction coefficients, and transfers each coefficient to the short-time prediction coefficient 24.
Set each. Here, each parameter of residual signal power, long-term prediction coefficient, and short-term prediction coefficient is transmitted for each frame or subframe.

If it is then determined at terminal 3 that the current frame is silent, no encoded signal is normally sent to terminal 1. Therefore, the noise generator 4 and the adder 22 are controlled to be connected via the changeover switch 5 411, and the switch 23 is opened. The short-time predictor 24 is operated with the prediction coefficients in the voice state of the previous frame set as they are. Further, noise having the same power is sent from the noise generator 4 using the information from the residual signal power decoder 17 regarding the power of the residual signal of the background noise in the active state of the previous frame. do. This noise is considered a residual signal and is added to the long-term synthesis filter 26. However, in this example, it is assumed that background noise does not have a pitch structure like speech, and the output of the long-term predictor 21 is not added, but is directly input to the short-term synthesis filter 27 and synthesized as background noise. This output becomes a suspicious noise signal with power and spectrum close to the background noise on the transmitting side. Background noise is encoded and transmitted in the frame of the voice state when the voice state changes from the voice state to the silent state, and the residual signal power decoder 17 outputs the power of the residual signal and the short-term prediction coefficient. The short-term spectral envelope is obtained from the decoder 20. Furthermore, the residual signal is generally decorrelated and has properties similar to noise, and the short-term spectrum of background noise and the residual signal power are thick (do not change) from frame to frame, so the above-mentioned Depending on the configuration, background noise can be synthesized on the receiving side.

In this example, a switch 23 is provided so that the long-term predictor 21 does not operate, because it is assumed that the background noise does not have a long-term correlation such as a pinch. Generally, even if the switch 23 is closed and the long-term predictor 21, which has been set with the long-term prediction coefficient of the previous frame, is operated, there is no major deterioration because the long-term prediction coefficient of background noise is close to 0.

Here, the residual signal power and the prediction coefficient are constant in frames in which a silent state continues. However, if the silent state continues for a long time, the noise power may be gradually reduced. Furthermore, when synthesizing background noise, the prediction coefficients of the synthesis filter may be weighted with decoded values.

When using the APC encoding method in this way, each function is prepared as explained in Example 1 by using the synthesis filter in the APC decoder and the power information of the residual signal when synthesizing background noise. If the background noise can be synthesized without any noise and is transmitted to the receiver when there is no call, the unnaturalness of the call can be eliminated.

In this way, in the present invention, a pseudo-noise signal (background noise) is easily created using power information and spectrum information during the hangover time immediately before the silent state, and this created pseudo-noise signal is inserted during the silent state. This allows the naturalness of the call to be maintained.

Next, when audio is transmitted in packet units, such as in a packet audio transmission system, an audio interpolation method will be described when a packet is lost on the transmitting side or during transmission.

(Embodiment 3) FIG. 5 is another embodiment according to the present invention, and is a schematic diagram showing an interpolation method of a pseudo voice signal when a packet voice transmission system is used. Note that Example 2 of the present invention described above (
4) is that a voice packet loss information signal is obtained from the packet voice transmission system and used instead of the voice/silence state signal, and the long-term predictor 21 in which voice pinch information is set is used. Always use it. Therefore, in the following description, operations different from those in the second embodiment will be described.

If the audio packet loss information signal does not come from the terminal 28,
As in the sound state of the second embodiment, the changeover switch 5 is connected to the inverse quantizer 18 side, and the packet audio received from the input terminal 1 is decoded.

On the other hand, when the voice packet loss information signal is received from the terminal 28, the selector switch 5 is switched to the noise generator 4 side, and the noise generator 4 and the adder 22 are connected. at the same time,
If the output of the long-term predictor 21 is not connected to the adder 22, it is connected by a switch 23. In this state, the noise generator 4 is set with the residual signal power from the residual signal power decoder 17 in the frame or subframe immediately before the packet loss, and similarly the short-term prediction coefficient and long-term prediction coefficient of the frame immediately before the packet loss. Time prediction coefficients are set in the short-time predictor 24 and the long-term predictor 21, respectively.

Therefore, the noise power output from the noise generator 4 is adjusted to be equal to the power of the frame immediately before the packet loss, and then
A pseudo speech signal in which the pitch information of the long-term synthesis filter 26 and the spectral envelope of the short-term synthesis filter 27 are adjusted is created and sent to the receiver. Therefore, in the present invention, even if a packet is lost, a pseudo voice signal is created and interpolated using noise as a driving sound source, so that the call quality can be significantly improved. Furthermore, the present invention does not require a memory for storing residual signals as in the prior art, and the apparatus can be simplified because a pseudo voice signal is created using noise as a driving sound source.

(Effects of the Invention) As described above, when a voice signal is missing, the present invention synthesizes pseudo speech using a synthesis filter driven by the coding parameters of the frame before the loss and noise, and interpolates the missing part. It is possible to significantly improve the voice quality of the receiver and maintain the naturalness of the call. Furthermore, in the silent state when there is no talk, background noise is synthesized using a synthesis filter driven by the encoding parameters and noise in the last frame of the active state, and this is inserted into the silent state to improve the naturalness of the call. This is extremely effective in improving call quality.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a schematic structure of a conventional pseudo-noise insertion method in a silent state; FIG. 2 is a block diagram showing an embodiment of the pseudo-noise insertion method for a voice transmission system according to the present invention; FIG. 3 is a connection diagram showing an example of the configuration of a short-time synthesis filter used in the present invention, FIG. 4 is a block diagram showing an embodiment of the pseudo-noise insertion method for APC encoding according to the present invention, and FIG. FIG. 2 is a block diagram showing an embodiment in which a pseudo voice signaling system for a packet voice transmission system is implemented according to the present invention.

Claims (1)

[Claims] (1) The receiving side discretely receives the signal within the sum of the length of time during which the audio signal to be transmitted from the transmitting side to the audio line exists and the predetermined hangover time length as transmission information, and the transmission In a pseudo signal insertion method in an audio transmission system that inserts a pseudo signal created on the receiving side and outputs it to the audio line during a time period in which no information is received, the power and spectrum of the noise signal generated on the receiving side are is configured to create the pseudo signal by adjusting the signal power information and spectrum information included in all of the transmission information received immediately before or at least a part of the hangover time. A method for inserting pseudo signals in audio transmission systems.