JP3856328B2 - Speech encoding method and speech decoding method - Google Patents

Description

  The present invention relates to a speech encoding method for predictively encoding and compressing a speech signal, and a speech decoding method for decoding a speech signal encoded by this speech encoding method.

  As a method for predictively encoding a speech signal, the present inventor has used a plurality of predictors having different characteristics for the original digital speech signal of one channel (channel) in the previous application (Japanese Patent Application No. 9-289159). A plurality of linear prediction values of the current signal are calculated from past signals in the region, a prediction residual for each predictor is calculated from the original digital speech signal and the plurality of linear prediction values, and the prediction residuals are calculated. A method for selecting the minimum value is proposed.

  However, in the above method, when the original digital audio signal has a sampling frequency = 96 kHz and the number of quantization bits = 20 bits, it is possible to obtain a certain degree of compression effect. Since the frequency (= 192 kHz) is used and the number of quantization bits tends to be 24 bits, it is necessary to improve the compression rate.

  Accordingly, the present invention provides a speech coding method capable of improving the compression rate when predictive coding a speech signal, and a speech decoding method for decoding a speech signal encoded by this speech coding method. With the goal.

In order to achieve the above object, the present invention comprises the following means 1) and 2).
That is,

1) For a stereo 2-channel audio signal, a head sample value is obtained in units of frames for a predetermined time in response to an input audio signal for each channel as it is or for each channel correlated with each other . A linear prediction method in which the linear prediction value of the current signal is predicted from the past in the time domain by the linear prediction method, and the prediction residual obtained from the predicted linear prediction value and the speech signal is minimized. Selecting and predictively encoding sub-frame units obtained by further dividing the frame;
The data structure includes header information and user data including a compressed PCM access unit, and the compressed PCM access unit is provided for each subframe, and each channel for each subframe selected in the step is selected. Predictive encoded data including a prediction residual and a linear prediction method is stored in a subpacket arranged in the compressed PCM access unit according to the number of samples corresponding to the sampling frequency of the audio signal, and the prediction residual is bit information Packing the number of bits based on the number of bits and storing the head sample value if the compressed PCM access unit is at the head of the frame; and
A speech encoding method comprising:
2) A speech decoding method for decoding an original speech signal from data encoded by the speech encoding method according to claim 1,
Extracting a compressed PCM access unit from the user data;
Retrieving the head sample value from the compressed PCM access unit and prediction encoded data in subframe units including a prediction residual of the number of samples according to a sampling frequency and a linear prediction method;
Decoding the prediction residual with the number of bits based on bit information, and calculating a prediction value based on the decoded prediction residual, the leading sample value, and a linear prediction method;
And restoring the original audio signal from the calculated prediction value,
A speech decoding method comprising:

  As described above, according to the present invention, it is possible to perform speech encoding with a compression rate improved more than before, and to perform speech decoding of the encoded speech signal without any inconvenience.

  Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a first embodiment of a speech encoding apparatus and a speech decoding apparatus corresponding to the speech encoding apparatus according to the present invention, FIG. 2 is a block diagram showing the encoder of FIG. 1 in detail, and FIG. 3 is multiplexed by the multiplexer of FIG. FIG. 4 is an explanatory diagram showing the format of the DVD pack, FIG. 5 is an explanatory diagram showing the format of the DVD audio pack, and FIG. 6 is a block diagram showing the decoder of FIG. 1 in detail. FIG.

  The channel correlation circuit A shown in FIG. 1 has an addition circuit 1a and a subtraction circuit 1b. The adding circuit 1a calculates a sum signal (L + R) of stereo 2ch signals L and R with each channel (hereinafter referred to as ch) having a sampling frequency of 192 kHz and the number of quantization bits of 24 bits, for example, and a 1ch lossless encoder 2D1 for the sum ch The subtraction circuit 1b calculates the difference signal (LR) and outputs it to the 1ch lossless encoder 2D2 for the difference channel. As shown in detail in FIG. 2, the encoders 2D1 and 2D2 predictively encode the differences Δ (L + R) and Δ (LR) of the sum signal (L + R) and the difference signal (LR), respectively, to record and communicate media Is transmitted through.

  On the decoding side, as shown in detail in FIG. 6, the decoders 3D1, 3D2 respectively decode the prediction encoded data of each channel into a sum signal (L + R) and a difference signal (LR), and then the channel correlation circuit B The sum signal (L + R) and difference signal (LR) are restored to stereo 2ch signals L and R.

  The encoders 2D1 and 2D2 will be described in detail with reference to FIG. The sum signal (L + R) and the difference signal (LR) are stored in one frame buffer 10 for each frame. Then, each sample value (L + R), (LR) of one frame is applied to the difference calculation circuits 11D1, 11D2, respectively, and the difference Δ (L + R), Δ (LR), that is, the difference PCM ( DPCM) data is calculated. In addition, the head sample values (L + R) and (LR) of each frame are applied to the multiplexer 19.

  The difference Δ (L + R) calculated by the difference calculation circuit 11D1 is applied to a plurality of predictors 12a-1 to 12a-n and subtractors 13a-1 to 13a-n having different prediction coefficients. Each of the predictors 12a-1 to 12a-n calculates each prediction value of the difference Δ (L + R) based on each prediction coefficient, and each of the subtractors 13a-1 to 13b-n calculates each prediction value and the difference Δ. Each prediction residual of (L + R) is calculated. The buffer / selector 16 </ b> D <b> 1 temporarily stores the plurality of prediction residuals, selects the minimum prediction residual for each subframe specified by the selection signal generator 17, and outputs the selected prediction residual to the packing circuit 18. Note that this subframe has a sample length of about one tens of frames, and one frame is 80 subframes as an example. Here, the predictors 12a-1 to 12a-n and the subtractors 13a-1 to 13a-n constitute a prediction circuit 15D1 for the sum signal ch, and the prediction circuit 15D1 and the buffer / selector 16D1 are the sum signal ch. The prediction encoding circuit is configured.

  Similarly, the difference Δ (LR) calculated by the difference calculation circuit 11D2 is applied to a plurality of predictors 12b-1 to 12b-n and subtractors 13b-1 to 13b-n having different prediction coefficients. The predictors 12b-1 to 12b-n calculate the predicted values of the difference Δ (LR) based on the respective prediction coefficients, and the subtracters 13b-1 to 13b-n respectively Each prediction residual of the difference Δ (LR) is calculated. The buffer / selector 16D2 temporarily stores the plurality of prediction residuals, selects the minimum prediction residual for each subframe specified by the selection signal generator 17, and outputs it to the packing circuit 18. The predictors 12b-1 to 12b-n and the subtractors 13b-1 to 13b-n constitute a difference signal ch prediction circuit 15D2, and the prediction circuit 15D2 and the buffer / selector 16D2 are prediction codes for the difference signal ch. Circuit.

  The selection signal generator 17 applies a prediction residual bit number flag (5 bits) to the packing circuit 18 and the multiplexer 19, and also predictor selection flags (the number n) indicating the predictor having the smallest prediction residual. 2 to 9 and 3 bits) is applied to the multiplexer 19. The packing circuit 18 packs the prediction residuals for 2ch selected by the buffers / selectors 16D1 and 16D2 with the designated number of bits based on the bit number flag designated by the selection signal generator 17.

The succeeding multiplexer 19 has a frame header (40 bits) for one frame as shown in FIG.
The first sample value (25 bits) of one frame of the sum signal ch (L + R),
The first sample value (25 bits) of one frame of the difference signal ch (LR),
A predictor selection flag (3 bits × 80) for each subframe of the sum signal ch (L + R);
A predictor selection flag (3 bits × 80) for each subframe of the difference signal ch (LR);
A bit number flag (5 bits × 80) for each subframe of the sum signal ch (L + R);
A bit number flag (5 bits × 80) for each subframe of the difference signal ch (LR);
A prediction residual data string (number of variable bits) of the sum signal ch (L + R);
-The prediction residual data string (number of variable bits) of the difference signal ch (LR) is multiplexed as an access unit and output as a variable rate bit stream. The prediction residual data string constitutes a subpacket. According to such predictive coding, when the original signal is, for example, sampling frequency = 192 kHz, number of quantization bits = 24 bits, and 2 channels, a compression rate of 59% can be realized.

  When this variable rate bit stream data is recorded on a DVD audio disk, it is packed into an audio (A) pack of compressed PCM shown in FIG. This pack consists of 2034 bytes of user data (A packet, V packet), 4 bytes of pack start information, 6 bytes of SCR (System Clock Reference) information, and 3 bytes of Mux rate ( rate) information and a 1-byte stuffing total 14-byte pack header are added (1 pack = total 2048 bytes). In this case, the time of the A pack in the same title can be managed by setting the SCR information as a time stamp as “1” in the first pack in the ACB unit and continuing in the same title.

As shown in detail in FIG. 5, the compressed PCM A packet is composed of a 17, 9 or 14 byte packet header, a private header, and audio compressed PCM data of 1 to 2015 bytes in the format shown in FIG. The compressed PCM private header is
A 1-byte substream ID,
-2-byte UPC / EAN-ISRC (Universal Product Code / European Article Number-International Standard Recording Code) number and UPC / EAN-ISRC data;
-1 byte private header length,
A 2-byte first access unit pointer;
-4 bytes of audio data information (ADI),
・ With stuffing byte of 0-7 bytes,
It is made up of.
Thus, the ADI of the compressed PCM A packet is selected to be 4 bytes, which is 4 bytes shorter than the ADI of the normal uncompressed PCM A packet. Therefore, the audio data can be increased by 4 bytes.

  Next, the decoders 3D1 and 3D2 will be described with reference to FIG. The variable rate bit stream data in the format shown in FIG. 3 is separated by the demultiplexer 21 based on the frame header. Then, the head sample values of one frame of the sum signal ch (L + R) and the difference signal ch (LR) are respectively applied to the accumulation arithmetic circuits 25a and 25b, and the sum signal ch (L + R) and the difference signal ch (LR) are applied. ) Predictor selection flags are applied as selection signals of the predictors (24a-1 to 24a-n) and (24b-1 to 24b-n), respectively, and the sum signal ch (L + R) and difference signal ch (L- The bit number flag (R) and the prediction residual data string are applied to the unpacking circuit 22. Here, the predictors (24a-1 to 24a-n) and (24b-1 to 24b-n) are the predictors (12a-1 to 12a-n) and (12b-1 to 12b-) on the encoding side, respectively. The same characteristics as those in n) are selected by the predictor selection flag.

  The unpacking circuit 22 separates the prediction residual data strings of the sum signal ch (L + R) and the difference signal ch (LR) based on each bit number flag, and outputs them to the adder circuits 23a and 23b, respectively. In addition circuits 23a and 23b, current prediction residual data of sum signal ch (L + R) and difference signal ch (LR) from unpacking circuit 22, and predictors (24a-1 to 24a-n), Of the (24b-1 to 24b-n), the previous predicted value predicted by each one selected by the predictor selection flag is added to calculate the current predicted value. The current predicted values are the differences Δ (L + R) and Δ (LR) calculated by the difference circuits 11a and 11b shown in FIG. 2, that is, DPCM data, and predictors (24a-1 to 24a-n). , (24b-1 to 24b-n) and the cumulative calculation circuits 25a and 25b.

  The cumulative calculation circuits 25a and 25b respectively add the differences Δ (L + R) and Δ (LR) for each sample with respect to the first sample value of one frame, and add the sum signal ch (L + R) and difference signal ch (L -R) of each PCM data is output. As for the sum signal (L + R) and the difference signal (LR), a 2L signal is calculated by the adder circuit 4a and a 2R signal is calculated by the subtractor circuit 4b as shown in FIG. Then, the 2L signal and the 2R signal are respectively divided by 1/2 by the dividers 5a and 5b, and the original stereo two-channel signals L and R are restored.

  Next, a 2nd form is demonstrated with reference to FIG. 7, FIG. In the above embodiment, each difference Δ (L + R) and Δ (LR) of the sum signal (L + R) and the difference signal (LR), that is, the DPCM data is configured to be predictively encoded. In the second embodiment, the sum signal (L + R), the difference signal (LR), that is, PCM data, or each difference Δ (L + R), Δ (LR), that is, DPCM data is selectively predictively encoded. It is configured.

  Therefore, in the encoding device shown in FIG. 7, prediction circuits 15A and 15S and a buffer / selector for predictively encoding the sum signal (L + R) and the difference signal (LR) with respect to the configuration shown in FIG. 16A and 16S are added. In addition, the selection signal generator 17 includes a sum signal (L + R) and a difference signal (LR) selected by the buffers / selectors 16A and 16S, and a difference Δ (respectively selected by the buffers / selectors 16D1 and 16D2. Based on the minimum value of each prediction residual of (L + R) and Δ (LR), it is determined which of the PCM data and DPCM data has the higher compression rate, and the higher data is selected. At this time, the PCM / DPCM selection flag (prediction circuit selection flag) is added and multiplexed.

  Here, the prediction circuit 15A for the sum signal (L + R) and the prediction circuit 15D1 for the difference Δ (L + R) shown in FIG. 7 have the same configuration, and the difference Δ (L−R) is different from the prediction circuit 15S. LR) prediction circuit 15D2 has the same configuration, the decoding apparatus does not need to provide both PCM data and DPCM data prediction circuits as shown in FIG. . Based on the prediction circuit selection flag transmitted from the encoding device, the selectors 26a and 26b select the outputs of the cumulative arithmetic circuits 25a and 25b in the case of DPCM data, and in the case of PCM data, the addition circuit 23a, Select the output of 23b.

  In the third embodiment, as shown in FIG. 9, the original signals L and R (PCM data), the sum signal (L + R), the difference signal (LR) (PCM data), and their respective differences Δ (L + R), One of the three groups of Δ (LR) (DPCM data) is selectively predictively encoded.

  For this reason, in the encoding apparatus shown in FIG. 9, prediction circuits 15L and 15R and buffer / selectors 16L and 16R for predictively encoding the original signals L and R, respectively, are added to the configuration shown in FIG. . The selection signal generator 17 includes the original signals L and R selected by the buffers and selectors 16L and 16R, the sum signal (L + R) and the difference signal (LR) selected by the buffers and selectors 16A and 16S. Then, a group of data with a high compression ratio is selected based on the minimum value of the prediction residuals of the differences Δ (L + R) and Δ (LR) selected by the buffer / selectors 16D1 and 16D2. At this time, the selection flag (prediction circuit selection flag) is added and multiplexed.

  In addition, when the three groups of prediction circuits shown in FIG. 9 have the same configuration, the decoding apparatus does not need to provide prediction circuits for three groups as shown in FIG. Based on the prediction circuit selection flag transmitted from the encoding device, the output of the cumulative arithmetic circuits 25a and 25b is selected in the case of DPCM data, and the output of the adder circuits 23a and 23b is selected in the case of PCM data. Then, the original signals L and R are restored by the channel correlation circuit B. Further, the outputs of the adder circuits 23a and 23b are selected by the selectors 27a and 27b in the case of the group of the original signals L and R, and the output of the channel correlation circuit B is selected in the other cases.

  Also, when variable rate bitstream data predictively encoded by the encoding side is transmitted via the network, the encoding side packetizes it for transmission as shown in FIG. 11 (step S41), and then packet header (Step S42), and then the packet is sent out on the network (step S43). As shown in FIG. 12, the decoding side removes the header (step S51), then restores the data (step S52), then stores this data in the memory and waits for decoding (step S53).

FIG. 13 shows an aspect different from FIG. 3 of the audio (A) packet of the compressed PCM (PPCM) shown in FIG. In this different mode, the audio data area in the compressed PCM (PPCM) audio (A) packet is composed of a plurality of PPCM access units as shown in FIG. 13, and the PPCM access unit is composed of PPCM sync information and subpackets. ing. The subpacket in the first PPCM access unit is composed of a directory, a bitstream BS0, CRC, and extra information, and the bitstream BS0 is composed only of PPCM blocks. Sub-packets in the second and subsequent PPCM access units are composed of a bit stream BS0, CRC, and extra information except for the directory. The sub-stream BS0 at the head of the frame has a restart header and a PPCM block (frame head sample value). Included).

The PPCM sync information (hereinafter also referred to as synchronization information) includes the following information.
-Number of samples per packet: 40, 80 or 160 is selected according to the sampling frequency fs.
Data rate: “0” in the case of VBR (an identifier indicating that the data in the subpacket is compressed data)
-Sampling frequency fs and number of quantization bits Qb
Channel allocation information The restart header has information specifying the channel correlation circuit A (having an addition circuit and a subtraction circuit) for each frame. The variable rate bit stream data in the format shown in FIG. 13 is decoded into the original 2-channel audio signal by the decoders 3D1 and 3D2 having the configuration below the demultiplexer 21 in FIG.

  FIG. 14 is a block diagram showing a second mode of the speech encoding apparatus and speech decoding apparatus according to the present invention. The channel correlation circuit A-1 shown in FIG. 14 includes an addition circuit 1a and a subtraction circuit 1b. The adder circuit 1a calculates the sum signal (L + R) of the stereo 2ch signals L and R, divides this sum signal (L + R) by ½ by the divider 5a, and then outputs the result to the lossless encoder 2D. 1b calculates a difference signal (LR), divides this difference signal (LR) by 1/2 by a divider 5b, and then outputs it to the lossless encoder 2D. The lossless encoder 2D multiplexes these signals using 1/2 (L + R) and 1/2 (LR) to create a multiplexed signal 250. The multiplexed signal 250 is decoded by the lossless decoder 3D to obtain the original ½ (L + R) and ½ (LR), which are the addition circuit 4a constituting the channel correlation circuit B-1. Are supplied to the subtracting circuit 4b, and stereo 2ch L and R signals are obtained as output signals. Note that a series of operations in the lossless encoder 2D and the lossless encoder 2D includes calculation of a difference, calculation of a prediction value, selection of a minimum prediction residual, calculation of a prediction value using the minimum prediction residual, and the like. This is performed in the same manner as in the above. The variable rate bit stream data in the format shown in FIG. 13 is identified by the identifier stored in the restart header of the PPCM access unit, for example, whether the channel correlation circuit of FIG. 1 or the channel correlation circuit of FIG. 14 is used. Therefore, it can be decoded reliably. In addition, although the lossless compression for every frame was demonstrated to the example, it is not fixed and an area may be made into variable length.

It is a block diagram which shows the 1st form of the audio | voice encoding apparatus to which this invention is applied, and the audio | voice decoding apparatus corresponding to it. It is a block diagram which shows the encoder of FIG. 1 in detail. It is explanatory drawing which shows the format of 1 frame multiplexed by the multiplexer of FIG. It is explanatory drawing which shows the format of the pack of DVD. It is explanatory drawing which shows the format of the audio pack of DVD. It is a block diagram which shows the decoder of FIG. 1 in detail. It is a block diagram which shows the encoder of a 2nd form. It is a block diagram which shows the decoder of a 2nd form. It is a block diagram which shows the encoder of a 3rd form. It is a block diagram which shows the decoder of a 3rd form. It is a flowchart which shows the audio | voice transmission method. It is a flowchart which shows the audio | voice transmission method. FIG. 6 is a format explanatory diagram showing a mode different from FIG. 3 of the audio (A) packet of the compressed PCM (PPCM) shown in FIG. 5. It is a block diagram which shows the 2nd form of the audio | voice encoding apparatus to which this invention is applied, and the audio | voice decoding apparatus corresponding to it.

Explanation of symbols

1a, 4a Adder circuit 1b, 4b Subtractor circuit 5a, 5b Divider 11D1 Difference operation circuit 11D2 Difference operation circuit 12a-1 to 12a-n Predictor 12b-1 to 12b-n Predictor 13a-1 to 13a-n, 13b -1 to 13b-n Subtractor 16D1, 16D2, 16A, 16S, 16L, 16R Buffer / selector 15A Prediction circuit 15S Prediction circuit 15L Prediction circuit 15R Prediction circuit

Claims (2)

A stereo two-channel audio signal is obtained for each channel as it is or in correlation with each other , and a head sample value is obtained in units of frames for a predetermined time in response to the input audio signal, and a plurality of linear predictions having different characteristics A linear prediction method in which a linear prediction value of a current signal is predicted from the past in the time domain by the method, and a prediction residual obtained from the predicted linear prediction value and the speech signal is minimized. Selecting for each subframe unit further divided and predictively encoding;
The data structure includes header information and user data including a compressed PCM access unit, and the compressed PCM access unit is provided for each subframe, and each channel for each subframe selected in the step is selected. Predictive encoded data including a prediction residual and a linear prediction method is stored in a subpacket arranged in the compressed PCM access unit according to the number of samples corresponding to the sampling frequency of the audio signal, and the prediction residual is bit information Packing the number of bits based on the number of bits and storing the head sample value if the compressed PCM access unit is at the head of the frame; and
A speech encoding method comprising: A speech decoding method for decoding an original speech signal from data encoded by the speech encoding method according to claim 1,
Extracting a compressed PCM access unit from the user data;
Retrieving the head sample value from the compressed PCM access unit and prediction encoded data in subframe units including a prediction residual of the number of samples according to a sampling frequency and a linear prediction method;
Decoding the prediction residual with the number of bits based on bit information, and calculating a prediction value based on the decoded prediction residual, the leading sample value, and a linear prediction method;
And restoring the original audio signal from the calculated prediction value,
A speech decoding method comprising:

Images