JPH10233691A - Encoding system and decoding system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve encoding efficiency even in the case that the amplitude distribution of frequency components is not uniform by providing respectively one or more quantization parts for encoding all the frequency signals of a block and quantization parts for encoding only the frequency signals of a part of the block in the plural quantization parts. SOLUTION: Digitized audio signals are inputted from an input terminal 1 and the signals are band-divided and converted to 32 pieces of the frequency signals every time the 32 pieces of audio signal samples are inputted in a signal conversion part 2. The 32 pieces of the frequency signals are turned to one set and the 12 sets of the frequency signals are stored. In an analysis part 3, by using the inputted audio signals and the frequency signals, the size of an allowable error is calculated for the respective frequency signals. A quantization part selection part 4 performs adjustment so as to make the size of the quantization error of the respective frequency signals be proportional to the size of the allowable error and decides the respective quantization parts so as to house a bit number required for encoding all the blocks within an allowable range determined from a transmission rate.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for efficiently encoding and decoding voice, audio, and the like.

[0002]

2. Description of the Related Art In a digital encoding system for speech and audio signals, the amount of transmission required for signal transmission is reduced by utilizing signal redundancy. Here, IS O / I EC (International Org)
announcement forStandardisat
ion / International Electro
Technical Commission International Standard (International Standard) 1
A description will be given of an example of an encoding method defined by the layer 1 of Layer 1172-3 with reference to FIGS.

A conventional encoder comprises an input terminal 1, a signal conversion unit 2, an analysis unit 3, a quantization unit selection unit 4, a quantization unit, a multiplexing unit 6, and an output terminal 7. A plurality of quantization units having different numbers of quantization steps are prepared. In this example, for simplicity of explanation, the three-step quantization units 51 and 7 are used.
Step quantizer 52, 15 step quantizer 53-3
The case where types are prepared will be described.

A digital audio signal is input from an input terminal 1. The signal converter 2 divides the input audio signal into bands and converts it into 32 frequency signals every time 32 audio signal samples are input. Further, these 32 frequency signals are set as one set, and
Store two sets of frequency signals. A set of 12 frequency signals for each frequency is called one block.

[0005] The analyzer 3 calculates an allowable error when quantizing each frequency signal. For example, when an objective signal-to-noise ratio is used as an evaluation of coding quality, the allowable error is made constant for each frequency signal. In addition, in audio coding, etc., not only objective numerical values such as the signal-to-noise ratio, but also subjective evaluations in trial listening are emphasized, so the coding noise is controlled to minimize the perceived quality deterioration of the reproduced sound. There is a need to. For this reason,
An allowable error is obtained based on at least one of the input audio signal and the frequency signal.

[0006] As will be described later, the quantization unit selection unit 4 determines a quantization unit to be used when quantizing a frequency signal for each block.

The quantization section obtains amplitude value information of the frequency signal in the block, encodes the frequency signal using the information, and outputs the amplitude value information and the sign of the frequency signal. The amplitude value information is obtained with an accuracy of 2 dB as shown in Table 1. Specifically, a frequency signal in one block stored in the buffer is inspected to detect a maximum amplitude value, and a value rounded up by 2 dB is set as an amplitude value. In this standard, 63 types of amplitude value information are prepared, and transmission of the amplitude value information requires 6 bits.

[0008]

[Table 1]

The quantization characteristic is linear in this standard. Assuming that the magnitude of the frequency signal is C, the amplitude value is L, and the number of quantization steps is S, using the coefficients A and B obtained from the number of quantization steps S according to Table 2, (A × C / L + B) × ( S +
1) / 2 is calculated, the upper N bits obtained by rounding off the decimal portion of the calculation result are extracted, and the sign of the frequency signal is obtained by inverting the most significant bit of the N bits. The inverse quantization unit in the decoding device obtains Q by inverting the most significant bit of the code, and calculates 2 × (Q + 1) / S ×
By calculating L, an inverse quantization signal of the frequency signal is obtained.

[0010]

[Table 2]

The specific operations of the quantization and the inverse quantization are as follows. The magnitude of the frequency signal of one block is 0.10, -0.15,
−0.03, 0.20, 0.05, 0.44, 0.0
5, -0.11,0.32, -0.40,0.92
An example in which it is 0.04 is shown.

In this block, the maximum amplitude value is 0.92
Therefore, the amplitude value is 1.0 (index is 3)
Select The codes obtained when the 15-step quantization is applied are 8, 6, 7, 9, 7, 7,
10, 7, 6, 9, 4, 14, and 7. The inverse quantization unit decodes these codes to obtain 0.133, -0.13
3,0.0,0.267,0.0,0.4,0.0,-
0.133, 0.267, -0.400, 0.933,
0.0 is obtained.

In the 15-step quantizer 53, four bits are required to transmit the code of one frequency signal. Therefore, 48 bits are used to transmit the codes of all 12 frequency signals in the block. It takes 6 bits to transmit the value information, for a total of 54 bits. When the quantization unit selection unit 4 selects the three-step quantizer, the obtained code is 1,1,1,1,1,2,1,1,2.
1,1,0,2,1. The dequantized value is 0.
0, 0.0, 0.0, 0.0, 0.0, 0.667,
0.0, 0.0, 0.0, -0.667, 0.667,
0.0.

Therefore, the number of bits required to transmit one block is a total of 30 bits, i.e., 12 samples of 2 bits representing the three-level quantization value of each frequency signal and 6 bits representing the amplitude value of the block. .

As shown by the inverse quantization calculation, the magnitude of the quantization error is proportional to (amplitude value L) / (quantization step number S) of each block. , The quantization accuracy of the frequency signal can be improved. However, when the number of quantization steps S increases, the number of bits N required to represent each code increases, and the transmission rate increases. Therefore, the quantization unit selection unit 4 adjusts the magnitude of the quantization error of each frequency signal so as to be proportional to the magnitude of the allowable error determined by the analysis unit 3 and encodes all the frequency signals. The quantization unit is determined so that the required number of bits falls within a range determined by the transmission rate.

The multiplexing unit 6 multiplexes the quantization unit selection information of each block and the quantization unit output to form a bit stream, and outputs the bit stream from an output terminal 7.

The conventional decoder comprises an input terminal 11, a separating unit 12, a three-step inverse quantizing unit 81, a seven-step inverse quantizing unit 82, a 15-step inverse quantizing unit 83, as shown in FIG. It comprises a signal inversion section 13 and an output terminal 14.

On the decoding side, the multiplexed signal is received at the input terminal 11 and separated by the decomposer 12 into quantizer selection information and quantizer output. According to the quantization unit selection information, an inverse quantization unit corresponding to the quantization method on the encoding side is selected from the three-step inverse quantization unit 81, the seven-step inverse quantization unit 82, and the 15-step inverse quantization unit 83. In each inverse quantization unit, the output of the quantization unit is separated into amplitude value information and the code of the frequency signal, and as described above, the code of the frequency signal is inversely quantized using the amplitude value information to obtain the frequency signal of each block. Reproduce. The signal inverse converter 13 inversely converts the frequency signal into a time domain signal and outputs the signal from the output terminal 14.

[0019]

In the prior art, one common quantization unit is used in a block. Therefore,
When the amplitude distribution of the frequency components is not uniform, for example, when only a minimum number of frequency signals have a prominent amplitude in a block, the quantized code is biased, and the coding efficiency is degraded.

[0020]

SUMMARY OF THE INVENTION The present invention provides a signal converter for converting an input signal into a frequency signal to form a block by combining a plurality of signals, and an analyzer for analyzing the input signal and the frequency signal to obtain an allowable error. A quantization unit selection unit that outputs quantization unit selection information for selecting a quantization unit to be used when quantizing the frequency signal of each block from the plurality of quantization units while referring to the allowable error; A plurality of quantization units for quantizing the frequency signal according to the quantization unit selection information to calculate the amplitude information and the code, and a multiplexing unit for multiplexing the amplitude information, the code, and the quantization unit selection information. In the quantization method, the plurality of quantization units include one or more quantization units for encoding all frequency signals of the block and one or more quantization units for encoding only some of the frequency signals of the block. I do.

Also, the present invention provides a separation unit for separating an input signal into amplitude information, a code, and quantization unit selection information, and a plurality of units for dequantizing the code using the amplitude information according to the quantization unit selection information. Inverse quantization in which a plurality of inverse quantizers decode all the frequency signals of a block in a decoding method including an inverse quantizer and a signal inverse transformer that performs signal conversion on outputs of the inverse quantizer collectively. And a dequantizer for decoding only a part of the frequency signal of the block.

Quantization and inverse quantization are performed on the frequency signal having a prominent amplitude using amplitude information, polarity information, and a frequency index.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to FIGS. Although the embodiment of the present invention describes a case where four types of quantization units and inverse quantization units are used, the number of quantization units and inverse quantization units is not limited.

The encoder according to the present invention comprises an input terminal 1, a signal conversion unit 2, an analysis unit 3, a quantization unit selection unit 4, a quantization unit,
It comprises a multiplexing unit 6 and an output terminal 7. As the quantization unit, a pulse quantization unit 50 is prepared in addition to the three types of three-step quantization unit 51, 7-step quantization unit 52, and 15-step quantization unit 53 corresponding to the conventional example.

A digitized audio signal is input from an input terminal 1. The signal converter 2 divides the input audio signal into bands and converts it into 32 frequency signals every time 32 audio signal samples are input. Furthermore, these 32 frequency signals are set as one set,
Twelve sets of frequency signals are stored. Twelve frequency signals are collected for each frequency to form one block. As the signal conversion, Fourier transform, discrete cosine transform, or the like may be used instead of band division, and a plurality of transform coefficients having continuous frequency indices may be combined into one block.

In the analysis unit 3, the same method as in the prior art is used.
Using the input audio signal and frequency signal,
The magnitude of the permissible error is calculated for each frequency signal.

The quantization section selection section 4 adjusts the magnitude of the quantization error of each frequency signal so as to be proportional to the magnitude of the permissible error, and adjusts the number of bits necessary for encoding all the blocks. The quantization unit used in each block is determined so as to fall within an allowable range determined from the transmission rate.
The determined quantization unit encodes the frequency component of each block as described later.

The multiplexing unit 6 multiplexes the quantization unit selection information and the quantization unit output of each block to form a bit stream, and outputs the bit stream from an output terminal 7.

The decoder according to the present invention has an input terminal 1
1, a separation unit 12, a pulse inverse quantization unit 80, a three-step inverse quantization unit 81, a seven-step inverse quantization unit 82, a 15-step inverse quantization unit 83, a signal inverse transformation unit 13, and an output terminal 14. You.

The multiplexed signal is received at an input terminal 11 and separated by a separation unit 12 into quantization unit selection information and a quantization unit output. According to the quantization unit selection information, the inverse quantization unit corresponding to the quantization method on the encoding side is set to the pulse inverse quantization unit 8.
A selection is made from the 0, 3 step inverse quantization section 81, the 7 step inverse quantization section 82, and the 15 step inverse quantization section 83. The selected inverse quantization unit reproduces the frequency signal of each block using the output of the quantization unit, as described later. The signal inverse converter 13 inversely converts the frequency signal into a time domain signal and outputs the signal from the output terminal 14.

A plurality of specific examples of the operation of quantization and inverse quantization will be described below.

As a first example, when a frequency signal of each block is encoded, a frequency signal having a prominent amplitude value is encoded by a pulse quantization unit, and other frequency signals are encoded by a conventional quantizer. I do. In the example used in the description of the conventional technology, the frequency signal 0.92 is encoded by a pulse quantizer,
Other frequency signals 0.10, -0.15, -0.0
3, 0.20, 0.05, 0.44, 0.05, -0.
11, 0.32, -0.28, and 0.04 are encoded by the same quantizer as the conventional one.

The pulse quantizer obtains an amplitude value, a polarity, and a frequency index for one frequency signal, and outputs it. In the example of processing the frequency signal 0.92, when applying the relationship between the index and the amplitude value information shown in Table 1 same as the related art,
1.0 (index is 3) is output as amplitude value information, and positive is output as polarity information. Since the frequency signal 0.92 is the eleventh in the block, the frequency index information is eleven. As described above, in the output of the pulse quantization unit, 6 bits for the amplitude information, 1 bit for the polarity information, and 1 bit in the block
4 for frequency index information identifying either of the two signals
A bit is needed. By excluding the salient signals, the maximum amplitude value is less than half from 0.92 to 0.44. Therefore, for these signals, the amplitude value is 0.5 (index is 6). The magnitude of the quantization error is (amplitude L) / (quantization step number S) of each block.
Therefore, in order to perform quantization with a quantization error of the same size as that in the conventional example using the 15-step quantization unit and the inverse quantization unit, the required number of steps is reduced from 15 steps to 7 steps. The number of bits required for transmitting each code can be reduced from four to three. Codes obtained by quantizing 11 frequency signals using a 7-step quantizer are 4, 2, 3, 4, 3, 6, 3, 2, 5,
0,3. On the decoding side, these codes are decoded,
0.143, -0.143, 0.0, 0.143, 0.
0, 0.429, 0.0, -0.143, 0.286,
-0.429, 0.0 is obtained.

Finally, when the present invention is used, the codes of all the frequency signals in one block can be transmitted with a total of 50 bits as shown in Table 3, which is compared with 54 bits when the conventional technique is used. , The required number of bits can be reduced.

[0035]

[Table 3]

Next, for the case where the number of quantization steps is 3, an example of using the pulse quantization and the inverse quantization of the present invention will be described with reference again to the example of the input signal shown in the prior art.

In a block having three steps, the magnitude of the quantized frequency signal is of three types: 0 and ± (amplitude value indicated by amplitude value information). In other words, by arranging positive and negative pulses around 0,
The frequency signal in this block can be represented. The information to be transmitted is one piece of amplitude value information common to the signals in the block, a frequency index of a frequency signal simulated by a pulse, and a polarity code.

In the example of the input signal shown in the prior art, since the frequency signal of the block can be simulated by three pulses, an example of bit allocation when the number of frequency components that can be expressed in one block is limited to four Are shown in Table 4. The frequency signal simulated by the pulse is 6, 10,
Since it is the eleventh signal, the information of 6, 10, 11 is encoded as a frequency index. Each polarity information is positive,
Negative and positive. The amplitude value information is 1.0 (index = 3) based on the maximum amplitude value, or 0.630 (index) based on the average value of the amplitude values of the three simulated frequency signals.
x = 5) is used.

The number of bits required for the amplitude value information, the frequency index, and the polarity code are 6, 1, and 4 bits, respectively. The total number of bits required is 23 bits, less than 30 bits according to the prior art example.

When the number of bits for the number information of the frequency signal to be encoded is changed to 1, 2, 3, or 4 bits, up to 2, 4, 8, and 16 frequency signals can be simulated, respectively. The required number of bits is M, where M is the number of frequency signals to be simulated.
6 + 1 + M × (4 + 1) and 6 + 2 + M × (4
+1), 6 + 3 + M × (4 + 1), 6 + 4 + M × (4+
1). Therefore, if the number of frequency components that can be represented in one block is limited to four, the required number of bits is always smaller than in the case of using the conventional three-step quantization.

[0041]

[Table 4]

When pulse quantization and inverse quantization are used, the number of bits can be reduced by limiting the selection of frequency signals to be encoded. For example, there is a restriction that 12 frequency signals in one block are divided into three groups of four groups, and only one frequency signal is encoded from each group. As an example of how to divide a group, the first group is 1, 4, 7, 1
The 0th frequency signal is assigned to the second group,
The eighth and eleventh frequency signals are assigned to the third group,
The sixth, ninth, and twelfth frequency signals are assigned. The magnitude of the amplitude value is used as a criterion for selecting one frequency signal from each group.

In the example of the frequency signal shown in the section of the prior art, the tenth frequency signal (−0.
40), the eleventh frequency signal (0.92) in the second group and the sixth frequency signal (0.44) in the third group have the largest amplitude in the group. In other words, among the frequency signals assigned to each group, the fourth, fourth, and second frequency signals are large. Therefore, if in addition to the amplitude value information, 4, 4 and 2 are transmitted as frequency indices of these selected frequency signals and negative, positive and positive are transmitted as polarity information, the frequency signal is reproduced on the decoding side. can do. At this time, as in the previous example,
The amplitude value is 1.0 (index) based on the maximum amplitude value.
Is 3) or 0.630 (index is 5) based on the average value of the amplitude values of the three frequency signals to be simulated.

As shown in Table 5, the number of bits required to transmit one block of frequency signals is 6 bits for block amplitude value information, and one of four frequency signals is used in each group. Two bits are used as an index for selection, and one bit is used to indicate the polarity of the amplitude of the selected frequency signal.
Requires a bit. Therefore, the total number is 15 bits, and the bit amount can be further reduced as compared with the above-described embodiment.

[0045]

[Table 5]

In the division of the frequency signals of one block into groups, as an example of how to divide the groups, first, second, third and fourth frequency signals are assigned to the first group, and 5th signals are assigned to the second group. , 6, 7, and 8 frequency signals, and the ninth, tenth, eleventh, and twelfth frequency signals to the third group.

[0047]

The number of bits required for quantization is reduced by adding a pulse quantizer 50 and a pulse inverse quantizer 80 capable of encoding and decoding only a part of frequency signals in a block. can do. Therefore, for a signal source in which some frequency signals have a larger amplitude than other frequency signals in the block, the same coding quality can be achieved with a smaller amount of bits than in the conventional example, and coding efficiency can be improved. .

[Brief description of the drawings]

FIG. 1 is a diagram showing an encoding method according to the present invention.

FIG. 2 is a diagram showing a decoding method according to the present invention.

FIG. 3 is a diagram showing an encoding method according to the related art.

FIG. 4 is a diagram showing a decoding method according to the related art.

[Explanation of symbols]

 Reference Signs List 1 input terminal 2 signal conversion unit 3 analysis unit 4 bit allocation unit 50 pulse quantization unit 51 3 step quantization unit 52 7 step quantization unit 53 15 step quantization unit 6 multiplexing unit 7 output terminal 11 input terminal 12 separation unit Reference Signs List 80 pulse inverse quantizer 81 3-step inverse quantizer 82 7-step inverse quantizer 83 15-step inverse quantizer 13 signal inverse transform unit 14 output terminal

Claims (2)

[Claims] 1. A signal conversion unit for converting an input signal into a frequency signal to collectively form a plurality of blocks to form a block, an analysis unit for analyzing the input signal and the frequency signal to obtain a permissible error, and referring to the permissible error. A quantization unit selection unit that outputs quantization unit selection information for selecting a quantization unit used when quantizing a frequency signal of each block from a plurality of quantization units; and a frequency signal according to the quantization unit selection information. A plurality of quantization units for quantizing the amplitude and calculating the amplitude information and the code, and a multiplexing unit for multiplexing the amplitude information, the code and the quantization unit selection information. A coding method, wherein the unit includes at least one quantization unit for coding all frequency signals of the block and one or more quantization units for coding only a part of the frequency signals of the block. 2. A demultiplexer for separating an input signal into amplitude information, a code and quantization section selection information, and a plurality of inverse quantization sections for dequantizing the code using the amplitude information according to the quantization section selection information. And a signal inverse transform unit for converting the output of the inverse quantizer collectively and a signal inverse transform unit, wherein a plurality of inverse quantizers decodes all frequency signals of the block and a block. And a dequantizer for decoding only a part of the frequency signals.