JPH03132216A - Encoder for digital data - Google Patents

Abstract

PURPOSE:To heighten data compression efficiency and to reduce the number of times of search for vector quantization by selecting first and second code books corresponding to the detected output of level change in block data with prescribed time interval. CONSTITUTION:The level change of the block data with the prescribed time interval is detected in an encoder 50, and at least the first or second code book 60, 61 is selected corresponding to the detected output, and the vector quantization of parameter data in plural blocks are performed by referring to a selected code book 60 or 61. Those first and second code books 60, 61 can be set as the code books 60, 61 in which the vectors of floating coefficients used in attack time and recovery time are stored. Thereby, it is possible to heighten the data compression efficiency by utilizing the correlation of respective change pattern, and also, to reduce the number of times of search for the vector quantization.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a digital data encoding device that compresses input digital data by performing block floating processing.

[Summary of the Invention] The present invention provides a digital data encoding device that blocks input digital data and quantizes the digital data of each block based on parameter data necessary for quantization for each block. At least the first and second codebooks are selected according to the detection output of a level change in block data at a predetermined time interval, and the parameter data of the plurality of blocks are vector quantized with reference to the selected codebooks. This provides a digital data encoding device capable of efficient parameter compression.

[Conventional technology]

For example, as a high-efficiency encoding technology for digital data based on audio signals, etc., input digital data is divided into blocks, floating coefficients are calculated for each block, data in each block is normalized by the floating coefficient, and then quantized. A so-called block floating technique is known. Input digital data that is subjected to this blocking includes sample data in the time axis direction that is obtained by digitizing analog signals such as audio signals, so-called ordinary PCM data, as well as various other types of encoded data. Data can be considered. In other words, these various encoding techniques include, for example, band division encoding (sub-band coding) in which a signal on the time axis, such as an audio signal, is divided into multiple frequency bands and encoded.
5BC) or converting a time axis signal to a frequency axis signal (
So-called adaptive transform coding (ATC), which divides into multiple frequency bands using (orthogonal transform) and adaptively encodes each band.
Alternatively, the above SBC and so-called adaptive predictive coding (APC) may be used.
), the time axis signal is divided into bands, each band signal is converted to a baseband (low band), and then multi-order linear predictive analysis is performed to perform predictive coding (so-called adaptive bit allocation (APC-AB)). etc.

Among these various encoding techniques, for example, in the adaptive transform encoding described above, audio signals on the time axis are processed using fast Fourier transform (FFT) or discrete cosine transform F14 (DCT).
), etc., to an axis (frequency axis) orthogonal to the time axis, and then divides it into multiple bands, and adaptively quantizes the FFT coefficients, DCT coefficients, etc. of each of these divided bands ( As an example of requantization in adaptive transform encoding of the fast Fourier transform,
As shown in the figure, blocks (blocks B1 to B1
2...), a floating coefficient is calculated for each block, and each block data is normalized (normalized and then quantized) using this floating coefficient, thereby performing the block floating process. In this case, the floating coefficient used is the peak value or average value of each block multiplied by a coefficient, and the data of each block is divided by the floating coefficient corresponding to the block (or the floating coefficient is Normalization is performed by multiplying when setting as the reciprocal of the peak value etc. Also, the floating coefficients are quantized and sent.When quantizing the floating coefficients, each floating coefficient is quantized using a so-called scalar quantization technique.In addition, -a, the above-mentioned scalar quantization refers to, for example, quantizing individual independent sample values, etc.
So-called one-dimensional quantization is collectively called scalar quantization.

Furthermore, in band division coding, which is one of the high-efficiency coding methods described above, for example, as shown in FIG.
...), and the signals S for each of these multiple frequency bands are quantized. These signals S are sampled in the time axis direction and divided into blocks for each predetermined sample (blocks BLI, BL2, etc.).
), and it is also possible to perform floating processing similar to the above for each of these blocks. Block floating processing in this band division coding is similar to the block floating processing in adaptive transform coding described above.
Normalization of the program data using floating coefficients and scalar quantization of the floating coefficients are performed.

[Problem to be solved by the invention]

As described above, in both the adaptive transform coding and the band division coding, the floating coefficients of the block floating process as the parameters are quantized using a scalar quantization method. In this scalar quantization, floating coefficients are quantized for each block, so for example, when transmitting this quantized output, the bit rate for transmitting the quantized floating coefficients must be In other words, a large number of bits are required to transmit the parameters.

Therefore, the present invention was proposed in view of the above-mentioned circumstances, and an object of the present invention is to provide a digital data encoding device that can significantly reduce the number of transmission bits for parameters. .

[Means to solve the problem]

The digital data encoding device of the present invention has been proposed to achieve the above-mentioned object, and it divides input digital data into blocks, calculates parameter data necessary for quantization for each block, and converts the input digital data into blocks. In a digital data encoding device that quantizes the digital data of each block based on, detecting a level change of the block data at a predetermined time interval,
At least the first and second codebooks are selected according to the detection output, and parameter data of a plurality of blocks are vector quantized with reference to the selected codebooks.

[Effect]

According to the present invention, since codebooks are provided separately according to the level change patterns of block data, data compression efficiency can be improved by utilizing the correlation between each change pattern, and vector Quantization searches can also be reduced.

〔Example〕

Embodiments to which the present invention is applied will be described below with reference to the drawings.

Input digital data such as audio and voice supplied to the input terminal 41 of the encoding device 50 of this embodiment is divided into blocks and quantized based on parameter data necessary for quantization, which will be described later, and calculated for each block. is performed, and is outputted from the output terminal 42 as the encoded output of the device 50 of this embodiment. Therefore, the input digital data is first sent to the blocking circuit 51,
The blocking circuit 51 blocks the input digital data in predetermined time units. This unit time block is transmitted to the book floating circuit 53 via the delay circuit 52.
3, input digital data is quantized for each unit time block using parameter data necessary for quantization calculated for each unit time block. Here, the block floating process performed by the block floating circuit 53 means that, for example, the amplitude information detection circuit 5
This is a process in which normalization is performed for each unit time block using a floating coefficient expressed by, for example, a peak value of each data for each unit time block from 5. That is, by dividing the input digital data for each unit time block by the value of the floating coefficient, for example, the input digital data is normalized. The thus normalized data for each unit time block is further requantized by the quantization circuit 54 and then outputted from the output terminal 42 of the device 50 of this embodiment as an encoded output.

Here, generally, in a device that performs such block-by-block encoding, as described above, quantization is performed for each unit time block, and the quantized data is output, and the above parameter data is also quantized. I'm trying to transmit it. However, as described above, in the past, each parameter data (floating coefficients, etc.) is scalar quantized and transmitted, resulting in a high transmission bit rate for these parameter data.

For this reason, in the encoding device 50 of this embodiment,
A so-called vector quantization method is used to quantize the parameter data (floating coefficients, etc.). Vector quantization involves comparing a code vector from a codebook composed of multiple code vectors (representative vectors) with an input vector, and selecting the code that is most similar (closest in distance) to the input vector. This is a quantization method that compresses data by reading an identification code (index) corresponding to a vector and obtaining this identification code. Therefore, in vector quantization of the floating coefficients, as shown in FIG. The identification code corresponding to the code vector is obtained as the vector quantization output of the floating coefficient.This vector quantization output (!a separate code) is
It is output from the output terminal 44 of.

By the way, for example, in audio signals, the rise of a signal is generally called an attack, and the fall or attenuation of a signal is called a recovery.

Further, the time from attack to the peak of the signal, for example, is called attack time, and the time from recovery until the signal attenuates to a predetermined level is called recovery time.

Furthermore, in general, an effect called temporal masking, which is masking on the time axis, is known in audio signals and the like. The temporal masking effect is related to the characteristics of human hearing, and when there is audio at an arbitrary time on the time axis, the audio before and after the audio is masked and cannot be heard by the human ear. It is an effect that disappears. Therefore, even if the S/N ratio of the signal is not improved during the time period during which this temporal masking effect is in effect, the human ear will not perceive it as a deterioration in sound quality. Furthermore, in the above temporal masking, masking due to the preceding voice (masking after the voice) is called anterograde masking, and masking before the voice is called retrograde masking. At the time of the above-mentioned attack, the backward masking effect will work, and at the time of the above-mentioned recovery, the anterograde masking effect will work. In the above-mentioned anterograde masking, the preceding voice is masked for a time of, for example, 100 ■S to 30 fJas, whereas in the backward masking, the masking effect is about 15 ■S.

In this way, in the above attack where the masking time is short, signal S/N degradation cannot be tolerated.
At the time of recovery, there is no audible problem even if the S/N is poor for a period of 100ss to 300ss from the start of the recovery.

Here, since the floating coefficient is normalized by dividing the input digital data as described above, the floating coefficient is used to normalize the input digital data having an amplitude value as shown in A in FIG. The value of
Generally, the value corresponds to the amplitude value at the attack start time t, and the recovery start time t! The value corresponds to the amplitude value. Therefore, by using the temporal masking effect, the value of the floating coefficient can be changed to the recovery start time t! from 100
This means that it is not necessary to change the m5 to 300 values faithfully in response to changes in the amplitude value.

In other words, when normalizing the input digital data of the amplitude value as shown in 8 in Fig. 3 using the floating coefficient,
The value of the floating coefficient is expressed as if IF(+lI
As shown in the floating coefficient value curve shown in F in Figure 3, the S/N is the highest at the end of the time period T (100ms to 300+a) during which the anterograde masking effect acts. It is sufficient to normalize with a floating coefficient that improves (changes slowly). In this case, the S/N value of the decoded output obtained by the decoding device 70, which will be described later, changes as shown by R in FIG. As described above, in this anterograde masking time T8, time 1. No matter what the amplitude value of the floating coefficient is, it becomes possible to have a similar pattern of change (how it decreases) in the value of the floating coefficient, which increases the correlation and improves the compression efficiency. becomes possible. In other words, vector quantization of floating coefficients as described above does not require a sufficient number of code vectors to correspond to the pattern of changes in the values of the floating coefficients, and therefore the search for vector quantization is reduced (operation I ).

For this reason, the device 50 of this embodiment detects the level change of the block data at a predetermined time interval, selects at least the first and second codebooks according to the detection output, and selects the selected codebook. Parameter data of multiple blocks is vector quantized by referring to a codebook. Here, at least the first and second codebooks may be codebooks that store vectors of floating coefficients used during the attack time and recovery time. The attack mode codebook is made up of a group of vectors in which the final word of the floating coefficient vector is larger than the first word, and the codebook made up of other vector groups is the recovery mode codebook. In addition, the first and second codebooks are two for attack mode and recovery mode, but at this time, for the attack mode.

When vector quantizing floating coefficients in a steady state other than recovery, this code vector for the steady state can also be included in the codebook for the recovery mode. Alternatively, a separate codebook for steady state conditions may be provided. Furthermore, each codebook can be made into a variable length code by varying the amount of data within the codebook.

That is, returning to FIG. 1 again, the unit time block from the block level circuit 51 is transmitted to the amplitude information detection circuit 55.
It is also transmitted to The amplitude information detection circuit 55 is configured to obtain an envelope connecting, for example, peaks of each data for each unit time block, and this envelope data is transmitted to the mode discrimination circuit 59. The mode discrimination circuit 59
The method detects the level change of the block data (envelope data) at the predetermined time interval. For example, data corresponding to one unit of vector is obtained from a memory, etc., and the data for each unit of vector is detected. By analyzing it, it is determined whether the vector data is present during the attack or during recovery as described above. This mode discrimination information is outputted from the output terminal 43 of the apparatus 50 of this embodiment, and serves as information for mode switching control in each section within the encoding apparatus 50.

Further, the envelope data is transmitted to low pass filters (LPF) 57 and 58 via a delay circuit 56. The low-pass filter 57 has a very small time constant and is designed to pass the data at the time of attack almost unchanged. The low-pass filter 58 has a large time constant, and is used to gently change the amplitude of the data during recovery and pass the data.

The outputs of the low-pass filters 57 and 58 are connected to selected terminals 63a and 63b of the selection switch 63.

This selection switch 63 is controlled tII based on the mode discrimination information from the mode discrimination circuit 59, and when the mode discrimination information indicates the attack mode, the selected terminal 63a is selected and the recovery mode is selected. When it is shown, the selected terminal 63b is selected.

The vector-based envelope data selected by the selection switch 63 based on the mode discrimination information is transmitted to the optimum vector selection circuit 62. The optimum vector selection circuit 62 also includes an attack mode codebook 60, which is the first and second codebook described above.
The output from the recovery mode code book 61 is supplied via a selection switch 64.

The selected terminal 64a of the selection switch 64 is connected to the codebook 60, and the selected terminal 64b is connected to the codebook 61. These two codebooks 6
0.61 stores the code vectors for attack mode and recovery mode as described above. Here, the selection switch 64 switches according to the mode discrimination information! 111, the code vector from the codebook 60 is transmitted to the appropriate vector selection circuit 62 in the attack mode, and the code vector from the codebook 61 in the recovery mode. Therefore, the optimum vector selection circuit 62 vector-quantizes the parameter data of a plurality of blocks by referring to the codebook selected based on the mode discrimination information. That is, the vector-based envelope data (
A comparison is made between the input vector) and each code vector from each codebook 60 or 61, and the closest one (
similar) code vectors are selected. In addition, when selecting this optimal vector, for example, to prevent overflow, each element of the floating coefficient vector is determined by the distance from the code vector, which does not exceed each element of the floating coefficient of the actual data. It is also possible to select the one with the smallest value. In addition,
Vector quantization based on characteristics such as the waveform of the amplitude value of input digital data as described above is generally called shape-gain vector quantization. This code vector selection information is sent to the code book 60 or 61, and a code vector based on this code vector selection information and an identification code corresponding to the code vector are read from the code book 60 or 61, respectively. It looks like this.

- Here, the identification code from the codebook 60 is transmitted to the selected terminal 66a of the selection switch 66, and the identification code from the codebook 61 is transmitted to the selected terminal 66b. This selection switch 66 is controlled based on the mode discrimination information, and the selected identification code is output from the output terminal 44 of this embodiment.

Further, the code vector read out from the code book 60 based on the code vector selection information is sent to the selected terminal 65a of the selection switch 65, and the code vector read out based on the code vector selection information from the code book 61 is sent to the selected terminal 65a of the selection switch 65. The signal is transmitted to the selected terminal 65b. The selection switch 65 is switched and controlled based on the mode discrimination information, and the 70-ting coefficient of the selected code vector is sent to the block floating circuit 53 as parameter data necessary for quantization for each unit time block. It is now being transmitted.

That is, in this embodiment, the floating coefficients as each parameter data are sent to the optimal vector selection circuit 62.
．． Since vector quantization is performed using codebook 60.61, the transmission bit rate can be reduced compared to the case of scalar quantization. Further, the code vector based on the mode discrimination information transmitted to the block floating circuit 53 is a code vector selected corresponding to the attack mode or recovery mode, and this code vector is based on the floating coefficient (parameter data). ) In this way, by switching the floating coefficients (code vectors) used for quantization in attack mode or recovery mode, quantization can be performed using floating coefficients that do not degrade S/N in attack mode. It can be performed.

Furthermore, in the recovery mode, the amount of search in vector quantization can be reduced. In other words, according to the device 50 of this embodiment, when the overall data compression rate increases, or for example, because the processing time block length is short, the number of σ bits in the processing time block is small. This method is highly effective in compressing parameters with a large information content ratio, such as in cases such as the following.

Note that when performing block floating processing in the block floating circuit 53 described above, for example, input digital data is divided into a plurality of frequency bands, and the number of bits allocated to each band is decreased as the energy within the band increases and the frequency decreases. It is also possible to do so.

Here, the configuration of the decoding device 70 is shown in FIG.

The encoded output of the encoding device 50 is transmitted to the input terminal 45 of the decoding device 70, the mode discrimination information is transmitted to the input terminal 46, and the identification code is transmitted to the input terminal 47.

The encoded output is transmitted to a block floating cancellation circuit 71, and the block floating cancellation circuit 71 performs processing opposite to the quantization in the block floating circuit 53, thereby canceling block floating.

At the same time, a process opposite to that of the quantization circuit 54 is performed, and the obtained decoded data is output from the output terminal 48.

Here, during the book floating cancellation process in the block floating cancellation circuit 71, the code from the attack mode codebook 72 and the recovery mode codebook 73 is selected via the selection switch 81 controlled by the mode discrimination information. Vectors are now being transmitted. The codebook 72 is connected to the selected terminal 81a of the selection switch 81, and the codebook 73 is connected to the selected terminal 81b. That is,
Code vectors corresponding to the above-mentioned identification codes are read out from these code books 72 and 73, and block floating cancellation processing is performed by using these code vectors. Note that the code vectors in the code books 72 and 73 are stored in a reciprocal relationship with the code vectors in the code book 60 and 61. That is,
Although it is possible to set the codebook small in the codebook 72.73 to be a codevector with the same content as the codebook 60.61, it is also possible to use the codevector with the same content to perform the block floating cancellation process. In this case, the encoded output must be divided by the code vector having the same content, and in this case, the circuit configuration becomes complicated. However, if a code vector having the above-mentioned reciprocal relationship is used, the block floating cancellation process can be performed by multiplying the encoded output by the code vector, thereby simplifying the circuit configuration. Furthermore, the S/N of this decoded output depends on the masking time T during recovery, as shown in Figure 3 above.
1, but there is no adverse effect on human auditory perception.

In addition to dividing input digital data into unit time blocks as in the embodiment shown in FIG. It can also be applied to a device that performs divisional encoding.

In other words, when applied to a device that performs band division encoding, input digital data on the time axis is divided into bands, and the time axis signals for each band are divided into blocks and encoded using mode discrimination and vector quantization. can do. Also,
When applied to a device that performs adaptive transform encoding, it converts input digital data on the time axis to data on the frequency axis,
This data on the frequency axis can be divided into blocks in predetermined units (frequency units) and encoded using mode discrimination and vector quantization. In this case as well, the number of assigned focuses may be reduced as the frequency range increases, as described above.

As described above, the same effects as described above can also be obtained in a coding apparatus in which the present invention is applied to band division coding or adaptive transform coding.

Although the parameters in the above-mentioned embodiments were floating coefficients, the above parameters are not limited to these floating coefficients, but may also be, for example, amplitude information obtained by log conversion (
d8) can also be used as a parameter.

Further, for example, bit allocation information indicating the number of allocated bits for each block can also be used as the parameter. In such a case, the transmission bit rate can also be reduced.

〔Effect of the invention〕

In the digital data encoding device of the present invention, at least first and second codebooks are separately provided according to the pattern of level changes of block data, and detection outputs of level changes of block data at predetermined time intervals are provided. Since at least the first and second codebooks are selected according to becomes possible. Furthermore, by vector quantizing multiple blocks of parameter data with reference to the selected codebook, it is possible to significantly reduce the number of parameter transmission bits, that is, the overall data compression rate. This is effective in compressing parameters with a large information content ratio, such as when the processing time block length is short and the number of bits in the processing time block is small. big.

[Brief explanation of the drawing]

FIG. 1 is a block circuit diagram showing a schematic configuration of a digital data encoding device according to an embodiment of the present invention, FIG. 2 is a diagram schematically showing vectors composed of unit time blocks, and FIG. 3 is a diagram showing amplitude values, 4 is a block circuit diagram showing a schematic configuration of a decoding device, FIG. 5 is a diagram schematically showing a block of adaptive transform coding, and FIG. 6 is a diagram showing floating coefficient values and S/N values. FIG. 3 is a diagram showing blocks of band division encoding. 51...Blocking circuit 52.56...Delay circuit 53...Block floating circuit 54
......Quantization circuit 55...Amplitude information detection circuit 5758...Low pass filter 59...Mode discrimination circuit 60... Attack mode code book 6
1...Recovery mode code link 62...Optimum vector selection circuit 63-66
・・Selection switch

Claims (1)

[Claims] Digital data encoding in which input digital data is divided into blocks, parameter data necessary for quantization is calculated for each block, and digital data in each block is quantized based on the parameters. The device detects a level change of the block data at a predetermined time interval, and detects at least the first and second level changes according to the detection output.
1. A digital data encoding device, wherein a codebook is selected, and parameter data of a plurality of blocks are vector quantized by referring to the selected codebook.