JP2006171353A - Voice decoding system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To make compatible temporal continuity of a voice signal and lowering of delay. <P>SOLUTION: An arrival time recorder 50 records arrival time, when a voice code packet arrives and a jitter counter 60 calculates the jitter value. A readout controller 70 sends an indication for a skip read of the packet or a read stop to a jitter buffer 10, based on the jitter value. A voice decoder 20 decodes the voice code packet supplied from the jitter buffer. A voice signal generating section 30 for compensation generates a voice signal for compensation for the start of packet skip read control and a voice signal generating section 40 for repetition generates a voice signal for repeated reproduction for the start of packet skip read control. A selector 80 receives notice from the readout controller 70 and selects and outputs a voice signal to be reproduced. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a speech decoding system that inputs a speech signal encoded by a speech coder from a transmission line, decodes it, and outputs it.

  In a voice communication system that receives and decodes a packet of a voice signal (voice code) that has been compressed and encoded via a transmission line, the arrival time interval of the packet is disturbed due to congestion of the transmission line, and jitter occurs. is there. In this case, the decoding timing in the decoder is not constant, and as a result, the decoded audio signal becomes temporally discontinuous.

  As a countermeasure, a storage device called a jitter buffer is arranged between the transmission line and the decoder, and while holding a certain amount of speech code packets in the jitter buffer, the speech code packets in the jitter buffer are sent to the decoder at regular time intervals. In general, the speech code is decoded by a decoder.

  However, when the jitter is large, if all the speech code packets stored in the jitter buffer are lost, the speech code to be decoded by the decoder is lost before the arrival of the next speech code packet, and speech communication is broken. As a countermeasure, as the capacity of the jitter buffer is increased and the number of voice code packets stored in the jitter buffer is increased, the tolerance to jitter becomes stronger. However, storing too many speech code packets in the jitter buffer is not preferable because it increases the transfer delay of speech code packets and impairs real-time performance of speech communication.

  In order to solve such a problem, conventionally, when the silent part of the received audio signal is extracted and the buffer amount is reduced, the silent part of the received audio signal is deleted and the buffer amount is increased. An example is known in which noise generation of a decoded audio signal is suppressed while a buffer amount is dynamically controlled by inserting a silence pattern immediately after the silence portion (see, for example, Patent Document 1).

  The voice code stored in the fluctuation absorbing buffer is provided with a fluctuation absorbing buffer for accumulating voice code packets supplied from the transmission path and a voice signal accumulation buffer for accumulating voice signals obtained by decoding the voice code by decoding processing. If the amount falls below the lower limit, the audio signal to be stored in the audio signal storage buffer is interpolated and at the same time the reading of the audio code from the fluctuation absorbing buffer is delayed to maintain the amount of audio code in the fluctuation absorbing buffer. When the amount of audio code stored in the absorption buffer exceeds the upper limit, the audio signal stored in the audio signal storage buffer is thinned out, and at the same time, the audio code in the fluctuation absorption buffer is accelerated so that the audio code in the fluctuation absorption buffer is accelerated. An example in which the amount is reduced is also known (see, for example, Patent Document 2). Thereby, it is said that deterioration of voice quality can be suppressed even in continuous voice transmission in which the frequency of appearance of silent sections is low.

JP 2002-271391 (2nd page-3rd page, FIG. 2) JP2003-050598 (2nd page-4th page, FIG. 1)

  However, in the technique described in Patent Document 1, since the buffer control is activated under the condition that the received audio signal includes a non-voice part, a call in an environment where background music is flowing, or a noise level However, there is a problem in that there is a case where a silent part does not occur for a long time in a call under a high environment, and the voice communication may fail without the buffer amount control being activated.

  Further, in the technique described in Patent Document 2, by performing linear interpolation processing in the interpolation processing, an average value between the past audio signal and the latest audio signal is calculated, and the average value is calculated as an audio signal for interpolation. Although the temporal continuity of the decoded audio signal is ensured by performing the decimation process, the decoding audio signal is simply thinned out by an arbitrary number of samples in the decimation process. Is not kept.

  Furthermore, the fluctuation absorption buffer control in this technology suppresses the jitter buffer overflow or underflow caused by the monotonous increase or decrease of the jitter buffer accumulation caused by the synchronization of the operation clock between transmission and reception. However, the failure of the buffer due to jitter caused by congestion of the transmission path is not considered. In other words, the fluctuation absorbing buffer in this technique is premised on having a sufficiently large capacity so as not to cause overflow or underflow due to jitter caused by congestion in the transmission path.

  Further, both of the techniques described in Patent Document 1 and Patent Document 2 are not control for actively reducing the voice code packet in the jitter buffer in order to reduce the delay of the voice communication, but preventing jitter buffer overflow or underflow. There is only control for.

  After all, in any of the above prior arts, in a voice communication environment where the arrival time interval of packets is disturbed due to congestion of the transmission path, ensuring continuity of voice communication by jitter buffer control and low delay of voice communication It is impossible to achieve both.

  Accordingly, an object of the present invention is to ensure the continuity of voice communication by absorbing jitter and to reduce the delay of voice communication by controlling jitter buffer in a voice communication environment in which the arrival time interval of packets is disturbed due to congestion of the transmission path. An object of the present invention is to provide a speech decoding system that realizes the above.

  In order to achieve the above object, the present invention imprints arrival time on a packet, calculates jitter from the time stamped, and supplies the jitter buffer based on the calculated jitter value and the jitter buffer. Decoder for decoding a speech code packet to be output and outputting a speech signal, a storage device for temporarily storing the decoded speech signal, means for generating a prediction signal from the decoded speech signal, and a decoding held by the storage device Means for generating a compensation speech signal using the decoded speech signal and the predicted signal, and means for controlling to output either the decoded speech signal or the compensation speech signal in conjunction with the means for controlling the buffer memory It has.

  More specifically, the speech decoding system of the present invention is a speech decoding system that decodes and outputs a speech code packet input from a transmission line, and holds the speech code packet and outputs a speech code packet by controlling a clock. A buffer (10 in FIG. 1), an arrival time stamp (50 in FIG. 1) that sequentially stamps the arrival times of speech code packets, and the difference in the arrival time by the arrival time stamp. A jitter counter (60 in FIG. 1) that counts the jitter, and when the jitter amount indicated by the jitter counter is sufficiently smaller than the retention amount of the voice code packet in the jitter buffer, Jitter is skipped, and if the retention amount of voice code packets in the jitter buffer does not have sufficient margin for the jitter amount indicated by the jitter counter A read controller (70 in FIG. 1) that stops reading the voice code packet from the buffer, and after receiving the voice code packet supplied from the jitter buffer based on the control by the read controller, the voice code packet is decoded. A speech decoder (20 in FIG. 1) that outputs a decoded speech signal, and a compensation speech for suppressing speech noise at the time of packet skip control by the readout controller by the decoded speech signal output by the speech decoder An audio signal for duplication for suppressing audio noise at the time of packet read stop control by the read controller is determined by a compensatory audio signal generator (30 in FIG. 1) that generates a signal and a decoded audio signal output from the audio decoder. One of the duplicate audio signal generation unit (40 in FIG. 1) to be generated, the decoded audio signal output from the audio decoder, the compensation audio signal, and the overlap audio signal is read out. And a selector (40 in FIG. 1) for selecting and outputting based on the control by the control unit.

  The compensation audio signal generation unit (30 in FIG. 1) includes a buffer memory (31 in FIG. 1) that holds the audio signal one clock before output from the audio decoder, and 1 stored in the buffer memory. A prediction signal generator (20 in FIG. 1) that predicts a speech signal discarded by skipping using a speech signal before the clock, and a predicted speech signal generated by the prediction signal generator at the start of skipping control. An overlap calculator (33 in FIG. 1) that generates a compensation audio signal by performing an overlap calculation so that the weight coefficient is increased and the weight coefficient for the subsequent audio signal is increased at the end of the skipping control. Have

  Also, the duplicate audio signal generator (40 in FIG. 1) includes a buffer memory (41 in FIG. 1) that holds the audio signal one clock before output from the audio decoder, and the 1 stored in the buffer memory. A prediction signal generator (42 in FIG. 1) that predicts a speech signal that is discarded due to read stop using a speech signal before the clock, and a weight for the predicted speech signal generated by the prediction signal generator at the start of read stop control The overlap is generated by increasing the coefficient and performing the overlap calculation so as to increase the weighting coefficient for the audio signal one clock before stored in the buffer memory at the end of the read stop control. And an arithmetic unit (43 in FIG. 1).

  According to the present invention, when the number of packets stored in the jitter buffer is small based on the arrival time jitter of the voice code packet, the number of packets stored in the jitter buffer is reduced. Therefore, transmission delay can be reduced, and when the jitter is large, by increasing the number of packets stored in the jitter buffer, it prevents voice interruption due to jitter buffer underflow, Real-time performance of voice communication can be ensured.

  In addition, when adaptively reducing and increasing the number of packets stored in the jitter buffer, it is possible to reduce the temporal discontinuity of the audio signal associated with jitter buffer control. It is possible to obtain the effect of being large in terms of contributing to improvement.

  In the voice communication system of the present invention, the arrival time is stamped on the voice code packet supplied from the transmission line, and the jitter is calculated based on the stamping time. Based on the calculated jitter, if the jitter is small, instruct the jitter buffer to skip reading packets to reduce the transfer delay, and reduce the number of packets stored in the jitter buffer. In order to prevent interruption of voice communication due to packet depletion in the buffer, an instruction to stop packet reading is issued to maintain the number of packets stored in the jitter buffer.

  When the skip skip control in the jitter buffer is activated, the decoded audio signal becomes temporally discontinuous. On the other hand, the compensation audio signal generation unit stores it in the compensation audio signal generation unit. Using a speech signal obtained by decoding a packet one hour before the skipped packet stored in the apparatus and a prediction signal generated using a speech signal decoded from the packet one hour after the skipped packet Thus, the discontinuity of voice communication due to skipping of packets is reduced by generating and outputting the compensation voice signal.

  In addition, when the readout stop control of the jitter buffer is activated, the decoded audio signal becomes discontinuous in time. On the other hand, in the duplication audio signal generation unit, in the duplication audio signal generation unit, A packet by generating and outputting a duplicate audio signal using the audio signal decoded immediately before the read stop control activation stored in the storage device and the prediction signal generated using the decoded audio signal The discontinuity of the voice communication due to the stop of reading is reduced.

  FIG. 1 is a block diagram showing the configuration of an embodiment of a speech decoding system according to the present invention. This speech decoding system inputs a coded input speech signal from a transmission line, decodes it and outputs an output speech signal, and comprises a jitter buffer 10, a speech decoder 20, a compensation speech decoding unit 30, The audio signal generator 40 for duplication, an arrival time stamp 50, a jitter counter 60, a read controller 70, and a selector 80 are included.

  The jitter buffer 10 holds a predetermined number of packets of an encoded audio signal (hereinafter referred to as “audio code”) supplied from a transmission line, while receiving a predetermined number of packets from an external clock supply source (not shown). In response to the clock, it is sent to the speech decoder 20 at regular intervals.

  The arrival time stamper 50 monitors the arrival of a packet from the transmission path to the jitter buffer 10, and when the packet arrives at the jitter buffer 10, the arrival time is stamped and the time stamp is passed to the jitter counter 60. .

  The jitter counter 60 holds the time from the arrival time stamp 50 and waits for the supply of the time stamp for the next arrival packet from the arrival time stamp 50. When receiving the time stamp of the next arrival packet, the jitter counter 60 takes the difference in time stamp between the previous arrival packet and the current arrival packet and supplies it to the read controller 70 as jitter. Further, after being supplied to the read controller 70 as jitter, the time stamp for the old packet is discarded among the time stamps of the two held packets.

  The read control unit 70 holds the jitter value supplied from the jitter counter 60 and updates it to the latest jitter each time it is supplied. When the updated jitter is smaller than a predetermined value, the read controller 70 issues a packet skip instruction in order to reduce the amount of speech code packets stored in the jitter buffer 10. At this time, the selector 80 is notified that the skipping control has been activated.

  In addition, when the updated jitter is larger than a predetermined value, the read controller 70 reads the packet from the jitter buffer 10 to prevent the voice code packets stored in the jitter buffer 10 from being depleted. Give stop instructions. At this time, the selector 80 is notified that the read stop control has been activated.

  The audio decoder 20 sequentially decodes the audio code packet supplied from the jitter buffer 10 and outputs the decoded audio signal to the selector 80, the compensation audio signal generation unit 30, and the duplication audio signal generation unit 40.

  The compensation speech decoding unit 30 generates a compensation speech signal for suppressing speech noise during packet skip control by the read controller 70 and outputs the compensation speech signal to the selector 80. For this purpose, a buffer memory 31, a prediction signal generator 32, and an overlap calculator 33 are provided.

  The buffer memory 31 holds the audio signal output from the audio decoder 20, and supplies the held audio signal to the prediction signal generator 32 in response to the same clock as the clock for the jitter buffer 10.

  The prediction signal generator 32 generates a prediction signal based on the audio signal supplied from the buffer memory 31 and supplies it to the overlap calculator 33. The prediction signal here is an audio signal generated by predicting an input audio signal at the next time with respect to the audio signal supplied from the buffer memory 31, and in this case, used for audio signal prediction. The speech signal and the prediction signal have temporal continuity. The prediction signal generation method generally includes linear prediction, but it is only necessary to maintain temporal continuity between the speech signal and the prediction signal, and the prediction method does not affect the essence of the present invention. Does not specify.

  As an example of a prediction method for a speech signal, a method shown in “G.711 APPENDIX I” defined as an extension of the international standard ITU-T G.711 of a speech coding method is well known. This is because the sound signal periodically repeats a similar waveform, so that the waveform of the repetition interval is used as a unit for comparison with the past sound signal, and the sound signal is predicted using the waveform with the highest correlation. It is a method.

  The overlap calculator 33 overlaps the input of the speech signal output from the speech decoder 20 and the prediction signal from the prediction signal generator 32, generates a compensation speech signal, and outputs it to the selector 80.

The overlap calculation here is an operation that takes the sum of two audio signals (here, the audio signal X and the audio signal Y) while applying a window function that linearly increases from 0 to 1, and at time t If the value of the audio signal X at X is X (t), the value of the audio signal Y is Y (t), and the window function is W (t), the audio signal XY (t) after the overlap calculation at time t is expressed by Equation 1. Given in.
XY (t) = (1-W (t)) * X (t) + W (t) * Y (t) (0 ≦ t <N)
Here, the time t takes a value from 0 to N, the time 0 is the time for the first value of the audio signal, the time N is the time for the last value of the audio signal, and the window function W (t ) Is a function that linearly increases from 0 to 1 when 0 ≦ t <N.

  The duplication audio signal generation unit 40 generates an audio signal for duplication for suppressing audio noise at the time of packet read stop control by the read controller 70 and outputs it to the selector 80. As means for this, a buffer memory 41, a prediction signal generator 42 and an overlap calculator 43 are provided.

  The buffer memory 41 holds the audio signal output from the audio decoder 20, and responds to the same clock as the clock for the jitter buffer 10 in response to the audio signal held by the prediction signal generator 42 and the overlap calculator. Supply to 43.

  The prediction signal generator 42 generates a prediction signal based on the audio signal supplied from the buffer memory 41 and outputs the prediction signal to the overlap calculator 43. The prediction signal generation method is the same as that in the prediction signal generator 32.

  The overlap calculator 43 overlaps the audio signal output from the buffer memory 41 and the prediction signal from the prediction signal generator 42, and outputs the calculation result to the selector 80 as an audio signal for duplication. The method of overlap calculation is the same as that in the overlap calculator 33.

  In response to the same clock as the clock for the jitter buffer 10, the selector 80 selects either the audio signal from the audio decoder 20, the overlap calculator 33 or the overlap calculator 43 and outputs it as an output audio signal. .

Here, when the output switching instruction from the read controller 70 is not received at the output timing, the audio signal output from the audio decoder 20 is selected and output as the output audio signal. On the other hand, when an output switching instruction is received from the read controller 70, an output switching instruction is selected from the compensation audio signal from the overlap calculator 33 and the overlapping audio signal from the overlap calculator 43. Select and output the specified one.
[Description of operation]
In FIG. 1, the input voice signal supplied from the transmission path is in units of packets storing voice codes. In general, packets supplied via a transmission path arrive at indefinite time intervals. When a packet storing a voice code arrives from the transmission path, the voice code packet is stored in the jitter buffer 10. Also, the arrival time stamper 50 stamps the arrival time of the packet, and this time is passed to the jitter counter 60.

  First, the operation for skipping a packet will be described. To simplify the description, reference is made to FIG. 2 in which portions related to packet skipping are extracted from FIG. 1, FIG. 3 showing an image of processing, and FIG. 4 showing a timing chart. Note that the packet in the column of jitter buffer 10 in FIG. 4 represents a voice code packet output from the jitter buffer 10.

  Now, speech code packets that are temporally continuous from the transmission path, packet A, packet B, packet C, packet D, and packet E are stored in jitter buffer 10 in the order of arrival, and packet A and packet B are already It is assumed that the data is output to the speech decoder 20 by the normal operation (clocks 1 and 2 in FIG. 4), decoded by the speech decoder 20, and then the packet C is output. Also, audio signals obtained by decoding the packets A, B, C, D, and E by the audio decoder 20 are referred to as audio signal A, audio signal B, audio signal C, audio signal D, and audio signal E, respectively.

  Note that the normal operation here is an operation in a case where neither the packet reading skip nor the packet reading stop control is activated from the read controller 70 to the jitter buffer 10 and the selector 80. In this case, the selector 80 Outputs the audio signal from the audio decoder 20 as an output audio signal.

  Here, when a new voice code packet (assumed to be packet F) arrives from the transmission line, the arrival time of the packet F is stamped by the arrival time stamp 50, and the arrival time of the packet F is checked by the jitter counter 60. Then, the jitter value is calculated from the arrival time of the packet E that arrived one time before the packet F, and passed to the read controller 70.

  If this jitter is smaller than a predetermined value, the read controller 70 instructs the jitter buffer 10 to skip the next read packet, packet C (clock 3 in FIG. 4), and the selector 80 Is notified that packet skipping has been activated. The jitter buffer 10 responds to an instruction from the read controller 70, skips the packet C at the packet output timing, and outputs the packet D.

  When the packet D is supplied from the jitter buffer 10, the audio decoder 20 decodes the packet D, and the audio signal D is sent to the selector 80, the buffer memory 31 in the compensation audio signal generation unit 30, and the overlap calculator 32. Output.

  The buffer memory 31 outputs the audio signal B supplied from the audio encoder 20 one time before the audio signal D to the prediction signal generator 32 at the same clock as the clock for the jitter buffer 10, and then receives the audio signal D that arrives thereafter. Hold. With this operation, the buffer memory 31 always holds the audio signal one time before the supply packet from the jitter buffer 10 to the audio decoder 20.

  The prediction signal generator 32 generates a prediction signal (referred to as a prediction signal B ′ here) for the audio signal B one time before the audio signal D.

  The overlap calculator 33 overlaps the speech signal D supplied from the speech decoder 20 and the prediction signal B ′ supplied from the prediction signal generator 32 to obtain a compensation speech signal (D + B ′). And output to the selector 80. The compensation audio signal (D + B ′) is an audio signal in which temporal continuity between the past audio signal B and the subsequent audio signal E is ensured.

  This overlap calculation is performed by applying the prediction signal B ′ to W (t) and the audio signal D to Y (t) in the above-described equation 1, so that the weight for the prediction signal B ′ at the start of the skipping control. XY (t) is obtained as an audio signal in which the coefficient is increased and the weighting coefficient for the audio signal D is increased at the end of the skip-by-reading control, whereby the past audio signal B and the subsequent audio signal are obtained. Therefore, temporal continuity with E can be ensured.

  The selector 80 receives the audio signal D from the audio decoder 20 and the compensation audio signal (D + B ′) from the overlap calculator 33 (clock 4 in FIG. 4), but the read controller 70. The audio signal for compensation (D + B ′) is selected and output in accordance with the instruction from. However, as will be described later, the selector 80 also receives the overlapping audio signal (B + B ′) from the overlap calculator 43.

  As a result, the output audio signal is in the order of the audio signal B, the compensation audio signal (D + B ′), and the audio signal E in the clocks 3, 4, and 5, as shown in FIG. While the output is skipped, the temporal continuity of a series of output audio signals is ensured by the compensation audio signal (D + B ′).

  Next, the operation when packet reading is stopped will be described. In order to simplify the description, reference is made to FIG. 5 in which portions related to packet reading stop are extracted from FIG. 1, FIG. 6 showing an image of processing, and FIG. 7 showing a timing chart. Note that the packet in the column of jitter buffer 10 in FIG. 7 represents a voice code packet output from the jitter buffer 10.

  Now, voice code packets that are temporally continuous from the transmission path are stored in the jitter buffer 10 in the order of arrival in the order of packets A, B, and C (clocks 1 and 2 in FIG. 7). Assume that A and packet B are output to speech decoder 20 through normal operations, have been decoded by speech decoder 20, and packet C is next output. Also, audio signals obtained by decoding the packets A, B, and C by the audio decoder 20 are referred to as audio signal A, audio signal B, and audio signal C, respectively.

  Here, when a new voice code packet (assumed to be packet D) arrives from the transmission path, the arrival time of packet D is stamped by arrival time stamper 50, and the arrival time of packet D is stamped by jitter counter 60. Then, the jitter value is calculated from the arrival time of the packet C that arrived one time before the packet D, and is sent to the read controller 70.

  When this jitter is larger than a predetermined value, the read controller 70 instructs the jitter buffer 10 to stop reading the packet C (clock 3 in FIG. 7), and also reads the packet to the selector 80. Notify that the stop has been activated. The jitter buffer 10 responds to the instruction of the read controller 70, stops reading the packet C at the packet output timing, and outputs nothing to the speech decoder 20.

  The buffer memory 41 outputs the held audio signal B at the previous time with the same clock as that for the jitter buffer 10 to the overlap calculator 43 and the prediction signal generator 42. The prediction signal generator 42 generates a prediction signal for the audio signal B (herein referred to as a prediction signal B ′) and outputs it to the overlap calculator 43.

  The overlap calculator 43 overlaps the audio signal B supplied from the buffer memory 41 and the prediction signal B ′ supplied from the prediction signal generator 42 to generate an audio signal for duplication (B + B ′). And output to the selector 80. This duplication audio signal (B + B ′) is an audio signal in which temporal continuity between the past audio signal B and the subsequent audio signal C is ensured.

  This overlap calculation is performed by applying the past audio signal B to W (t) and the prediction signal B ′ to Y (t) in the above-described equation 1, so that the audio signal B is started at the start of the read stop control. XY (t) can be obtained as an audio signal in which the weighting factor is increased and the weighting factor for the prediction signal B ′ is increased at the end of the read stop control. Thus, temporal continuity with the subsequent audio signal C can be ensured.

  The selector 80 receives the no-signal from the speech decoder 20 and the duplicate speech signal (B + B ′) from the overlap calculator 43 (clock 4 in FIG. 7), but from the read controller 70. The audio signal for duplication (B + B ′) is selected and output according to the instruction. However, as described above, the selector 80 also receives the compensation audio signal (D + B ′) from the overlap calculator 33.

  As a result, as shown in FIG. 7, the output audio signal is in the order of the audio signal B, the duplication audio signal (B + B ′), and the audio signal C in the clocks 3, 4, and 5. While the output is once stopped, the temporal continuity of a series of output audio signals is ensured by the audio signal for duplication (B + B ′).

The block diagram which shows the structure of one Example of the audio | voice decoding system by this invention. The part extracted from FIG. 1 related to packet skipping in the present invention The figure which shows the operation | movement image of packet skipping in this invention Timing chart showing operation of skipping packets in the present invention The part extracted from FIG. 1 related to the packet reading stop in the present invention The figure which shows the operation | movement image of the packet reading stop in this invention Timing chart showing operation of stopping packet reading in the present invention

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Jitter buffer 20 Audio | voice decoder 30 Compensation audio | voice signal production | generation part 31 Buffer memory 32 Prediction signal generator 33 Overlap operation unit 40 Duplication audio | voice signal generation part 41 Buffer memory 42 Prediction signal generator 43 Overlap operation unit 50 Arrival time Stamper 60 Jitter counter 70 Read controller 80 Selector

Claims (5)

In a speech decoding system for decoding and outputting speech code packets input from a transmission path,
A jitter buffer that holds the voice code packet and outputs the voice code packet by controlling a clock;
An arrival time stamper that sequentially stamps the arrival times of the voice code packets;
A jitter counter that counts the arrival time jitter from the difference in time stamped by the time stamper;
When the jitter amount indicated by the jitter counter is sufficiently smaller than the retention amount of the voice code packet in the jitter buffer, the voice code packet is skipped from the jitter buffer, and the voice code packet in the jitter buffer is read. A read controller that stops reading voice code packets from the jitter buffer when the amount of dwell is not sufficient for the jitter amount indicated by the jitter counter;
A voice decoder that receives the supply of the voice code packet based on the control by the read controller from the jitter buffer, then decodes the voice code packet and outputs a decoded voice signal;
A compensation speech signal generator for generating a compensation speech signal for suppressing speech noise at the time of packet skip control by the readout controller based on the decoded speech signal output by the speech decoder;
An audio signal generator for duplication that generates an audio signal for duplication for suppressing audio noise at the time of packet read stop control by the read controller, based on the decoded audio signal output by the audio decoder;
A selector that selects and outputs one of the decoded audio signal output from the audio decoder, the compensation audio signal, and the duplication audio signal based on control by the read controller; Voice decoding system. The compensation audio signal generator is
A buffer memory for holding an audio signal one clock before output from the audio decoder;
A prediction signal generator that predicts an audio signal discarded by skipping using an audio signal one clock before stored in the buffer memory;
The overlap calculation is performed so that the weighting coefficient for the predicted speech signal generated by the prediction signal generator is increased at the start of the skipping control and the weighting coefficient for the subsequent speech signal is increased at the end of the skipping control. The speech decoding system according to claim 1, further comprising an overlap calculator that performs the compensation speech signal. The duplication audio signal generator is
A buffer memory for holding an audio signal one clock before output from the audio decoder;
A prediction signal generator that predicts an audio signal discarded due to stop of reading by using an audio signal one clock before stored in the buffer memory;
The weighting coefficient for the audio signal one clock before accumulated in the buffer memory at the start time of the read stop control is increased, and the weight coefficient for the predicted audio signal generated by the prediction signal generator at the end time of the read stop control. 2. The speech decoding system according to claim 1, further comprising an overlap computing unit that performs an overlap computation so as to increase the number of the speech signals for duplication. The compensation audio signal generator is
A first buffer memory for holding an audio signal one clock before output from the audio decoder;
A first prediction signal generator for predicting an audio signal discarded by skipping using an audio signal one clock before stored in the buffer memory;
The overlap calculation is performed so that the weighting coefficient for the predicted speech signal generated by the prediction signal generator is increased at the start of the skipping control and the weighting coefficient for the subsequent speech signal is increased at the end of the skipping control. Having an overlap calculator for generating the compensation audio signal,
The duplication audio signal generator is
A second buffer memory for holding an audio signal one clock before output from the audio decoder;
A smooth second predictive signal generator in which the next continuous discontinuity is suppressed by using the audio signal one clock before stored in the buffer memory;
At the start of read stop control, the weight coefficient for the audio signal one clock before stored in the buffer memory is increased, and at the end of read stop control, the weight for the predicted audio signal generated by the prediction signal generator. The speech decoding system according to claim 1, further comprising a second overlap computing unit that performs an overlap operation so as to increase a coefficient to generate the duplicate speech signal. 2. The overlap calculator applies a window function to an audio signal one clock before stored in the buffer memory and a prediction audio signal generated by the prediction signal generator. The speech decoding system according to claim 4.

Images