JPH09152899A - Voice coding changeover system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the generation of an abnormal sound by providing specific control means to eliminating the generation of the sense of noise to be caused by the discordance of internal states between an encoder and a decoder at the time of changing over voice coding systems. SOLUTION: A coding system changeover data detecting part 7A detects that a coding system is changed over from the transmission data on a transmission line T. Then, suppression circuits 81A, 82A, 91A, 92A monitor the control signal from the coding system changeover data detecting part 7A and when the control signal is turned to a low level, the circuits become silent states for several 10 msec and perform such suppressions that suppression amounts are gradually reduced with the passage of time. Consequently, since silent voices are inputted to encoders and coded data are generated and they are made to be decoded in decoders, the discordance of the internal states between the endcoder and the decoder is corrected. Thereafter, since level risings are gradually controlled, the discontinuity of the decoded voice disapears to eliminate sense of incongruity.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding switching system, and more particularly to a codec having a codec which is a voice coding / decoding means adapted to a plurality of voice coding systems and adapted according to the voice coding system. The present invention relates to a voice coding switching system for switching and selecting.

[0002]

2. Description of the Related Art A voice signal, which is an analog signal, is converted into a digital signal at a transmitting side terminal, the digital signal is subjected to required signal processing and then transmitted through a digital transmission line, and the receiving side terminal converts this. A digital communication technique of a voice signal which is decoded again and converted into an analog signal to be received has been developed.

Further, in this case, VSE is used as an encoding method.
LP (Vector-Sum Excited Lin)
Ear Predictive Coding) and PS
I-CELP (Pitch Synchronous)
Innovation Code Excited L
There is also proposed a method in which one audio processing system has a CODEC of a different encoding method so as to be compatible with a different encoding method, such as an inner prediction.

For example, a method described in Japanese Patent Laid-Open No. 2-84830 is used as a voice coding switching system, and this conventional example is shown in FIG. In FIG. 10, terminals A and B are connected via a digital transmission line T,
One terminal A is provided with a telephone 1A, a hybrid 2A, a first CODEC 31A, a second CODEC 32A, A / D converters 41A and 42A, and D / A converters 51A and 52A.

Similarly, the other terminal B has a telephone set 1B, a hybrid 2B, a first CODEC 31B, and a second CODE.
C32B, A / D converters 41B and 42B, and D / A converters 51B and 52B are provided.

Further, the encoding system switching data detecting section 7
A, 7B and D / A converter temporary stop sections 10A, 10B are provided.

The hybrids 2A and 2B are telephones 1A and 1
The B transmission / reception signal is distributed.

Further, changeover switches 6A and 6B are provided between the terminals A and B and the transmission line T, and the changeover switches 6A and 6B are such that the encoding method is changed over from the transmission data on the transmission line. Based on the determination results from the determination units 7A and 7B, the switching is performed in conjunction with the CODEC side of the same type.

Here, the encoding system switching data detection unit 7
A and 7B detect that the coding method is switched from the transmission data on the transmission path, and the coding method switching data detection units 7A and 7B detect the coding method determination data from the data on the transmission path T. And a detection unit for detecting encoding method switching data.

The determination data is the first CODEC 31
It is assumed that A and 31B are "0" and the second CODECs 32A and 32B are "1". And when there is no change in the judgment data (C
When the ODEC is not switched), a high level output is output, and when the judgment data changes from 0 → 1 or 1 → 0 (when the CODEC is switched), a certain period of time (for example, during call) It outputs a low level for a few hundreds of microseconds, which is not a problem.

Further, the changeover switches 6A and 6B switch the CODEC to be used based on the coding system determination data from the coding system switching data detecting units 7A and 7B so that the same type of CODECs are connected. And the receiver side works together.

Further, the D / A converter temporary stop unit 10A, 1
OB temporarily suspends the operation of the D / A converters 51A, 52A, 51B, 52B on the receiving side at the time of switching by the changeover switches 6A, 6B.

When there is no change in the judgment data, the conversion clock to the D / A converter is not interrupted and the operation of the D / A converter is not interrupted. If there is a change in the judgment data, the output signal of the encoding method switching data detection unit becomes low level, the conversion clock to the D / A converter is temporarily kept at low level, and the D / A converter Temporarily stop the operation of. As a result, the receiving side temporarily becomes silent.

[0014]

However, in such a speech processing system which switches the conventional speech coding method, the residual driving LPC (LiP) such as CELP is used as the coding method.
In the case of near predictive coding, there is a problem that an abnormal noise may occur due to the internal state mismatch between the encoder and the decoder after the voice coding system is switched.

For example, in the CELP CODEC, the encoder and the decoder have the same codebook, and the encoder searches for a strongly correlated pattern by using the pattern of the input voice data, the internal state of the encoder and the codebook. The index is transmitted to the decoder, and the decoder generates a composite signal from the internal state of the decoder and the codebook based on the received index. When the internal states match, the audio signal input to the encoder can be reproduced.

However, immediately after switching the encoding system, the internal state is different because the encoded data of the encoder is not decoded. If the decoding process is performed in a different state, there is a drawback that a feeling of noise occurs and abnormal noise occurs in the worst case.

It is an object of the present invention to eliminate the occurrence of noise caused by the inconsistency of the internal states of the encoder and the decoder at the time of switching the voice encoding system and reduce the generation of abnormal noise. Is to provide.

[0018]

According to the present invention, the speech coding / decoding means having speech coding / decoding means adapted to a plurality of speech coding systems and adapted according to the speech coding system is provided. Is a voice coding switching system for switching and selecting the voice coding method, wherein each input / output data of the voice coding means is put in a silent state for a certain period in response to the switching of the voice coding method, and thereafter the level gradually rises. (EN) A voice coding switching system characterized by including a control means for controlling.

Further, according to the present invention, instead of controlling the output data of each decoding unit of the speech encoding unit by the control unit, the data of each frame to be decoded is to be decoded in each of the decoding units. Error detection and correction means for performing error detection and correction, bad frame masking means for performing bad frame masking processing on frame data that cannot be corrected by this error detection and correction means, and decoding for data after this bad frame masking processing A voice coding switching system is provided which further comprises a decoding means for performing a bad frame masking process by operating the bad frame masking means in response to the switching of the voice coding method.

[0020]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The operation of the present invention will be described. At the encoder input and the decoder output, a silent state is set for a certain period immediately after the voice coding system is switched, and thereafter, the voice level is controlled to gradually increase with the passage of time. By doing so, silent voice is input to the encoder, encoded data is generated, and this encoded data is decoded by the decoder, so that the mismatch between the internal states of the encoder and the decoder is corrected. Then, since the level is gradually increased, the discontinuity of the decoded speech is eliminated and the discomfort is eliminated.

Embodiments of the present invention will be described in detail below with reference to the drawings.

1 and 2 are block diagrams of an embodiment of the present invention, in which the same parts as in FIG. 10 are designated by the same reference numerals. 1 shows a block diagram of the terminal A on one side, and FIG. 2 shows a block diagram of the terminal B on the other side. Also, CO
This is a case where there are two audio encoding methods of DEC, but the present invention is not limited to this.

Referring to FIGS. 1 and 2, terminals A and B are connected via a digital transmission line T, and one terminal A is connected.
Has a telephone 1A, a hybrid 2A, a first CODEC
31A, second CODEC 32A, A / D converter 41A,
42A, D / A converters 51A and 52A are provided, and the other terminal B has a telephone 1B and a hybrid 2
B, first CODEC 31B, second CODEC 32B, A
/ D converters 41B and 42B, D / A converters 51B and 52
B is provided. Furthermore, the point that the encoding method switching data detection units 7A and 7B are provided is similar to the conventional example of FIG.

However, compared with the conventional example, COD
Suppression circuits 81A, 81B, 82A, 82B, 91A, between the EC and A / D converter and between the CODEC and D / A converter.
91B, 92A, and 92B are provided respectively, and the D / A converter temporary stop portion is removed, which is a structural difference.

Encoding method switching data detectors 7A, 7B
Detects the switching of the encoding method from the transmission data on the transmission path, and the encoding method switching data detection units 7A and 7B detect the encoding method determination data and the encoding method from the data on the transmission path T. A detection unit that detects system switching data is provided.

The determination data is the first CODEC 31
It is assumed that A and 31B are "0" and the second CODECs 32A and 32B are "1". And when there is no change in the judgment data (C
Output a high level when ODEC is not switched) and output a low level when the judgment data changes from 0 → 1 or 1 → 0 (when CODEC is switched). It has become.

Further, the changeover switches 6A and 6B switch the CODEC to be used based on the coding system determination data from the coding system switching data detecting units 7A and 7B so that the same type of CODECs are connected. And the receiver side works together.

Suppression circuits 81A, 82A, 91A, 92A
An output signal of the encoding system switching data detection unit 7A is input to the input terminal, and audio data suppression control is performed. Similarly, the suppression circuit 81B, 82B, 91B, 92B inputs the output signal of the encoding method switching data detection unit 7B as a control signal to perform suppression control of voice data.

For example, these suppression circuits monitor the control signal from the encoding system switching data detection section, and when this control signal becomes low level, in response to this, the silence circuit becomes silent for several tens of msec. Therefore, the suppression is performed such that the suppression amount is gradually reduced with the passage of time. As an example of the suppression processing after this silent section, it is proposed to perform i / 80 weighting processing (smoothing processing) on each sample data i of the audio signal. In addition, the same effect can be obtained by performing weighting that can be expressed by a function of exp (-τt).

A specific example of the suppression circuit will be described below. FIG. 3 is a block diagram of a specific example thereof, and FIG. 4 shows an example of an operation time chart thereof. When the switching of the voice encoding system is detected, assuming that the switching timing signal a is generated in the form of a trigger pulse in synchronization with the timing, the address generation circuit 201 responds to the trigger pulse a and the data ROM (read Only memory) 20
The read address of 2 is generated while sequentially incrementing from the initial value “0”.

A multiplication coefficient is stored in advance in the data ROM 202, and the read multiplication coefficient and the input data b are latched by the latches 204 and 203, respectively.
It is input to 05. The multiplication output is derived as the output signal c via the latch 206 and becomes the suppression circuit output.

For example, by preliminarily storing in the ROM 202 the multiplication coefficients as shown in FIGS. 5 and 6 for each address, the characteristic that the level increases at a constant inclination of 1/80, and 1-
It is possible to realize a suppression circuit having a characteristic that increases with exp (-t / 80).

Since the duration of occurrence of abnormal noise varies depending on the encoding method, the storage address of the value 0 of the multiplication coefficient of the ROM is changed in order to change the length of the silent period t0 according to the encoding method to be switched. Try to change the number.
Further, the level increasing characteristic can be selected variously by the multiplication coefficient.

As an example of the suppression circuit, a software configuration (CPU configuration) other than the hardware configuration shown in FIG. 3 can be used. In this case as well, the silent period t0 and the level increase characteristic can be controlled as desired. And can be easily implemented.

The encoder encodes the voice data in the silent section t0 and transmits the encoded data. The internal state is almost initialized by encoding the silent data of several tens of msec. The decoder can almost initialize the internal state by decoding the coded data in the silent section, and can correct the mismatch between the internal states of the encoder and the decoder. However, immediately after the encoding system is switched, there is a possibility that abnormal noise may occur because it is a transitional period of the correction of the internal state mismatch. Therefore, the decoded output of the transition period is muted by performing the suppression process on the decoder output signal. As a result, it is possible to suppress the occurrence of temporary abnormal noise.

Considering the case where the suppression processing is performed only on the decoder side, it is possible to suppress the occurrence of a temporary abnormal sound immediately after the switching, but since the internal states of the encoder and the decoder do not match, a silent interval is reached. The decoded speech with noise added is output. It is important to perform suppression processing on the encoder side in order to correct the internal state mismatch.

However, even when the suppression processing is performed only on the encoder side or when the suppression processing is performed only on the decoder side,
Abnormal noise can be suppressed by setting the time in a silent section.

When the suppression processing is performed only on the decoder side, a silence detecting means and a suppression processing control means are added so as to control so as to suppress the decoded speech until the silence of the decoded speech is detected after switching the coding method. Is desirable.

In addition to the configuration shown in FIG. 1, it is also possible to input a control signal from the encoding system switching data detection section to each CODEC and operate as an internal process of the CODEC. The encoder performs the suppression process as a pre-process of the encoding process, and the decoder performs the suppression process as a post-process of the decoding process.

FIGS. 7 and 8 are block diagrams of another embodiment of the present invention, in which the same parts as those in FIGS. 1 and 2 are designated by the same reference numerals. In the present embodiment, the CO in the embodiment of FIGS.
Instead of using the suppression circuit for the output data on the decoding unit side of the DEC, the function of replacing the suppression circuit is added to the line of the decoding unit in each CODEC.

Each CODEC will be described. On the terminal A side, coding units 311A and 321A, error detection and correction code (CRC) generation units 312A and 322A, error detection and correction units 313A and 323A, bad frame masking units 314A and 324A, and decoding unit 315A,
And 325A.

The decoded data of the decoding units 315A and 325A are directly passed through the D / A converter 51A, without passing through the suppression circuit.
52A is input respectively. Then, the output signal from the encoding method switching data detection unit 7A is input as a control signal to the bad frame masking units 314A and 324A.

The other terminal B has the same configuration as the terminal A.

Here, the bad frame masking process in the bad frame masking section will be described.
The bad frame masking technique is one of the typical error control techniques in voice coding, and is disclosed in, for example, NTTR & D Vol. 43, N
o. 4,1994 p. 377.

The error detection / correction unit 313A, 323A, 313B, 323B using the error correction code (CRC) may not be able to completely correct the error on the transmission line, and as a result, the error on the transmission line remains. Happens. A large amount of distortion may occur in the voice decoded from the received data in which an error remains, and the audible characteristic deteriorates. In order to improve the auditory characteristics in this case, when a residual error is detected, the degradation of the auditory characteristics is suppressed by waveform processing that gradually attenuates the level of the decoded speech or smoothes the connection with the decoded speech before and after. I am devising.

This technique is bad frame masking, which is an interpolation process in decoding when a transmission line error that cannot be corrected by an error correction code occurs, and is a bad frame masking unit 314A, 324A, 314B, 324.
This processing is performed in B respectively.

Further, the operation of the bad frame masking process will be described in detail with reference to the state transition diagram of FIG. This process is based on a 9-state (0-8) machine as shown in FIG. 9, and transitions from each state to the state indicated by the arrow.
A 1 on the path indicates a case where a CRC error is detected in the current frame. The state changes each time the received data is decoded. The normal state is 0, which means that there is no CRC error in the CRC determination. State 6 is entered when a CRC error occurs at least 6 times consecutively. Different processing is performed in each state. State 8 is forcibly transitioned when the voice coding method is switched.

In state 0, no processing is performed. In state 1 and state 2, frame repetition is simply performed. In states 3, 4 and 5, speech is repeated and attenuated. States 6, 7 and 8
Now, completely silence the voice.

The state number is CRC, with three exceptions.
Indicates the number of consecutive frames of CRC error in the judgment. For example,
State 5 indicates that 5 consecutive frames (including the current frame) are in error. The exceptions are states 6, 7, and 8.

In the states 6 and 7, CRC errors of 6 frames or more occur continuously. In any of the states except State 6 and State 8, the receive CRC and the regenerated CRC
When and match, the state transits to state 0, which is the initial state. To return from state 6 to state 0, it is necessary to continue two frames in which no CRC error is detected.

Under a bad condition for a long time, as a result of CRC judgment,
Since it may be mistakenly indicated as valid voice data, the additional protection described above is added. Even when the state 8 transits to the state 0, two frames without CRC error must be consecutive.

State 0, which is an error-free state, is the initial state of the system. The state machine remains in this state except when no CRC error is detected and in case of a coding scheme switch. Here, a CRC error is defined as occurring when the regenerated CRC does not match the received CRC. The parity polynomial is generated based on the decoded input polynomial that may contain channel errors.

In successive speech frames where an error is detected, the state machine transitions to the next number respectively. State 8
Changes from all states when the coding method is switched, and changes to state 7 with a CRC error.

If no CRC error is detected in the voice frame, the state machine returns to state 0 except for states 6 and 8. Unlike the other cases, the transition in the state 6 transits to the state 7 in one frame in which no error is detected, or returns to the state 0 in two consecutive frames in which no error is detected. less than,
The processing in each state is shown.

State 0: No CRC error detected. The received decoded voice data is used as it is.

State 1: A CRC error has been detected in the frame. Replace each encoding parameter value with the corresponding value in the last frame that was in state 0. The remaining decoded bits in this frame are sent to the speech encoder as they are.

State 2: The same processing as in state 1 is performed.

State 3: Subtract 2 from R0 (parameter indicating frame energy). Otherwise, the frame is iterated in the same manner as in the states 1 and 2.

State 4: The same processing as in state 3 is performed. R0
Is reduced by 2 and the current power level is 8 dB lower than the original value.

State 5: Subtract 2 from the value of R0.

State 6: Frame or iterate,
At the same time, the parameter R0 is set to 0. Alternatively, comfort noise is inserted instead of voice.

State 7: The voice signal remains silent R0 = 0.

State 8: The same processing as in state 7 is performed.

In the states 3 to 5, after subtracting 2 from the value of R0, it is checked that the value is 0 or more. If the value is negative, R0 is set to 0.

In the above-described bad frame masking process, in the present embodiment, the bad frame masking units 314A, 324A, 3 are switched at the switching timing of the coding method.
14B and 324B are activated. That is, the state of the path 2 (code switching state) in the state transition diagram of FIG. 9 is set.
As a result, the state 8 is forcibly changed.

From the state 8, R0 = 0 is set for two consecutive frames, and then the state is returned to the state 0 for the first time. Therefore, the silence data (R0 = 0) is input to the decoding unit for two frames after the encoding method is switched, and then the reception data is decoded, so that the generation of abnormal noise is suppressed.

[0067]

As described above, the first effect of the present invention is a difference that may occur due to the internal state mismatch between the encoder and the decoder when the encoding method is a residual drive LPC such as CELP. The generation of sound can be reduced.

The reason for this is that by performing suppression processing on a voice signal at the time of switching the encoding system, a voice signal starting from a silent state and exhibiting a gentle rising characteristic can be input to the encoder. By inputting the suppressed audio signal, the internal state of the encoder is initialized, and by inputting the audio signal exhibiting a gradual rising characteristic, the discontinuity of the signal is alleviated. Also, by decoding the decoder that encoded these suppression signals with the decoder, the internal state of the decoder was initialized in the same manner as the encoder, and the internal state mismatch between the encoder and the decoder could be corrected.

Further, by suppressing the decoded voice output signal of the decoder, the decoded voice output when the internal states of the encoder and the decoder do not match (immediately after switching) is suppressed, so that the generation of abnormal noise can be suppressed.

The second effect is that it is possible to remove the temporary noise generated when the coding method is switched.

The reason is that the decoder output signal is muted by suppressing the audio signal when the coding method is switched.

[Brief description of the drawings]

FIG. 1 is a partial block diagram of an embodiment of the present invention.

FIG. 2 is a partial block diagram of an embodiment of the present invention.

FIG. 3 is a diagram showing an example of a suppression circuit according to an embodiment of the present invention.

4 is a timing chart showing the operation of the suppression circuit of FIG.

5 is a diagram showing a part of data stored in a data ROM of FIG.

6 is a diagram showing a part of data stored in a data ROM of FIG.

FIG. 7 is a partial block diagram of another embodiment of the present invention.

FIG. 8 is a partial block diagram of another embodiment of the present invention.

FIG. 9 is a state transition diagram showing an operation of bad frame masking processing according to another embodiment of the present invention.

FIG. 10 is a block diagram of a conventional speech coding switching system.

[Explanation of symbols]

 1A, 1B Telephone 2A, 2B Hybrid 6A, 6B Changeover switch (switching means) 7A, 7B Coding method changeover data detector 31A, 31B First CODEC 32A, 32B Second CODEC 41A, 41B A / D converter 51A, 51B D / A converter 81A, 82A, 81B, 82B 91A, 92A, 91B, 92B Suppression circuit A, B Terminal T Digital transmission line

Claims (6)

[Claims] 1. A voice encoding system comprising voice encoding / decoding means adapted to a plurality of voice encoding systems, wherein the voice encoding / decoding means adapted to the voice encoding system is switched and selected. A switching system, comprising control means for controlling each input / output data of the speech coding means to be in a silent state for a certain period in response to the switching of the speech coding method, and controlling the characteristic so that the level gradually increases thereafter. Voice coding switching system. 2. The voice coding switching system according to claim 1, wherein the control means controls the level to rise at a constant slope after the constant period. 3. The voice coding switching system according to claim 1, wherein the control means controls the level to rise at a predetermined time constant after the fixed period. 4. The memory for storing a multiplication coefficient according to the level rising characteristic in advance, and the address generator for sequentially generating read addresses of the memory in response to switching of the voice coding method. 4. The voice coding switching system according to claim 1, further comprising: a multiplication unit that multiplies each input / output data of the voice encoding unit by a multiplication coefficient read from the memory. 5. An error for error detection and correction of data of each frame to be decoded in each of the decoding units instead of controlling the output data of each decoding unit of the voice encoding unit by the control unit. A detection / correction unit, a bad frame masking unit for performing a bad frame masking process on the frame data that cannot be corrected by the error detection / correction unit, and a decoding unit for decoding the data after the bad frame masking process. 5. The voice coding switching system according to claim 1, wherein said bad frame masking means is operated in response to the switching of said voice coding system to perform a bad frame masking process. 6. The speech coding according to claim 5, wherein said bad frame masking means operates in response to the switching of the speech coding method to minimize the decoded speech level for at least two frames. Switching system.