JPH0758707A - Quantization bit allocation system - Google Patents

Abstract

PURPOSE:To offer a quantization bit allocation system which can effectively utilize the total number of bits in a composite frame and improve the quality of a playback signal. CONSTITUTION:The quantization bit allocation system of a system which quantizes and encodes signals of plural channels CH0-CHm-1, and then puts them together in a frame and sends the frame evaluates specific properties of the respective channel signals at a time and allocates bits adaptively to the quantization of the respective channels CH0-CHm-1 based on the evaluation result so that quantization errors of the respective channel signals are minimized without exceeding the number of sent frame bits of the system. The specific properties of the respective channel signals are preferably masking levels obtained in consideration of the amplitudes of input signals S0-Sm-1, the amplitudes of signals C0-Cm-1 after quantization, or the power spectra of the input signals S0-Sm-1 and noises and human auditory characteristics.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a quantized bit allocating method, and more particularly, a quantized bit allocating method in a system in which signals of a plurality of channels are respectively quantized and encoded, and then these are frame-synthesized and transmitted. Regarding In recent years, research on high-efficiency coding methods for stereo audio signals and the like has been advanced for the purpose of application to communication and storage media. Since a pair of stereo audio signals is one information source, it is preferable to combine the original high-efficiency code data for two channels into one frame signal and record it on a transmission or storage medium. Therefore, it is desired to provide a quantized bit allocation method that can effectively utilize the total number of bits of the combined frame in such a case and improve the quality of the reproduced signal.

[0002]

2. Description of the Related Art FIG. 7 is a diagram showing the structure of a conventional quantized bit allocation system, which is an example of the structure of a band division coding system in which two-channel stereo audio signals S ₀ and S ₁ are input. Is shown. In the figure, 1 ₀ and 1 ₁ are band division filters, 2 ₀ and 2 ₁ are high-speed Fourier transform units (FF).
T) 3 ₀ and 3 ₁ are auditory operation units, 4 ₀ and 4 ₁ are quantization units, 5 ₀ and 5 ₁ are coding units, 6 ₀ and 6 ₁ are bit allocation units,
Reference numeral 7 is a frame synthesis unit.

[0003] Focusing on the operation of the channel 0, the band division signals B ₀ of the band division filter 1 ₀ predetermined number of audio signal S ₀ which is sampled in the frame of several msec~ several tens msec (for example, 32) Split into. at the same time,
Fast Fourier Transform unit 2 audio signal S ₀ of the input at ₀
The the power spectrum PS ₀ obtained by fast Fourier transform, followed by auditory arithmetic unit 3 ₀ In the power spectrum P
A masking level M _0, which will be described later, is obtained based on the human psychoacoustic model based on S ₀ . Furthermore, the bit allocation unit 6 ₀
Based on the obtained power spectrum PS ₀ and the masking level M ₀ , adaptively allocates quantized bits so that the quantization error of the band-divided signal B _{0 in} the quantizer 4 ₀ is minimized, and the encoder 5 _{At 0} , the obtained quantized data Q ₀ is converted into transmission code data C ₀ . Then, the frame synthesizing unit 7 synthesizes (packs) a series of code data C ₀ and, if necessary, auxiliary data D ₀ representing the number of quantized bits thereof into an area of a predetermined size for channel 0 on the transmission frame, This is transmitted. The same applies to the operation of channel 1.

FIG. 8 is a diagram for explaining the principle of quantized bit allocation in the configuration of FIG. In the human ear, for each frequency, there is a minimum audible limit M _ORG that a sound below a certain level cannot be heard. Moreover, when a loud sound PS is present at a certain frequency, it is more difficult to hear sounds of frequencies around the PS (so-called masking effect). This can be understood as that the level of the minimum audible limit M _ORG around it is increased as a result of being affected by a loud sound, and this is represented by the masking level M in the figure.

Regarding the method of calculating the masking level,
Reference `` IEEE JOURNAL ON SELECTED AREAS IN COMMUNICAT
IONS, VOL.6 NO.2 FEBURUARY 1988 −Transform coding
ofAudio Signals Using Perceptual Noise Criterria
− By JAMES D. JOHNSTON ”. Therefore,
In the band of the signal-to-mask ratio SMR ≦ 0, human beings cannot hear the signal sound, so that it is not necessary to allocate the quantization bit. Assuming linear quantization, for example, S
In the band of MR> 0, a larger band of SMR requires more quantized bits, and a band of smaller SMR requires a smaller number of quantized bits.

[0006]

As described above, in the prior art,
Quantized bits were adaptively allocated independently for each channel. However, if the quantization bit is independently assigned to each channel, the number of data bits for all channels secured on the composite frame may not be effectively utilized. That is, since a certain channel is mainly silence or noise, the actual number of quantization bits is small, and as a result, the number of bits reserved for this channel is surplus. On the other hand, in other channels, the sound is the main, so S
Although it is desired to perform finer quantization in order to improve the / N ratio, the number of bits secured for this channel is constant, so the S / N ratio cannot be improved further.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a quantized bit allocation system which can effectively utilize the total number of bits on a composite frame and improve the quality of a reproduced signal.

[0008]

The above-mentioned problems can be solved by the structure shown in FIG. That is, the quantized bit allocation method of the present invention is a method of quantizing and encoding signals of a plurality of channels respectively, and then frame synthesizing and transmitting these signals, and a predetermined bit for each channel signal In addition to collectively evaluating the characteristics of the above, based on the evaluation result, the quantization error of each channel is adaptively adjusted so that the quantization error of each channel signal is minimized within a range not exceeding the number of transmission frame bits of the system. Bit allocation is performed.

[0009]

In the figure, in the transmission system of this example, input signals S _{0 to} S of a plurality of channels CH _{0 to} CH _{m- 1} are input.
_{After m-1} is quantized and encoded respectively, these are frame-synthesized and transmitted. In this case, the bit allocation unit collectively evaluates the predetermined property of each channel signal, and based on the evaluation result, the quantization error of each channel signal is within a range not exceeding the number of transmission frame bits of the system. Bit allocation is adaptively performed for quantization in each channel so as to minimize. Therefore, the total number of data bits of the combined frame FS can be effectively utilized and the quality of the reproduced signal can be improved.

Preferably, the predetermined property of each channel signal is the amplitude of the input signals S _{0 to} S _{m- 1} . According to the so-called adaptive coding principle,
If the input signal S ₀ can always be quantized with a constant M bits regardless of its amplitude, the maximum S / N in M-bit quantization is obtained.
It is known that ratios can be achieved. Other input signal S ₁
The same applies to ~ S _m-1 .

The bit allocation unit in this case is basically
By using the redundancy included in the amplitude of the input signal S ₀ , for example, the quantization width Δ _n at the present time can be calculated based on the input signal S _{0 (n−1)} at the immediately previous time, Although it is adaptively changed (so-called forward adaptive encoding) by Δ _n = f (S _{0 (n-1)} ) so that M is always constant, this is also a kind of bit allocation control. In addition, even in this case, it is not necessary to assign the quantization bit to the channel where the silence or noise is the main, and the excess quantization bit for that channel can be passed to another channel, for example, the input signal S ₁
Is always quantized with a constant (M + 1) bits regardless of its amplitude, the S / N ratio of channel 1 can be further improved.

Therefore, the bit allocator in this case collectively evaluates the amplitudes of the respective input signals S _{0 to} S _m-1 and, based on the evaluation result, within a range not exceeding the number of transmission frame bits of the system, Bit allocation is adaptively performed for quantization in each channel so that the quantization error of each channel signal is minimized. Further, preferably, the predetermined property of each channel signal is the amplitude of the quantized signals C _{0 to} C _m-1 . This is so-called backward adaptive coding, but the present invention can be applied to this case as well.

Further, preferably, the predetermined property of each channel signal is the masking levels M _{0 to} M _{m-1 in} consideration of the power spectra PS _{0 to} PS _m-1 of the input signal and the noise signal and human hearing characteristics. Is. Details of the present invention will be apparent from the following description of the embodiments. Further, preferably, the presence / absence of a sound is determined for each frame unit of each of the channel signals S _{0 to} S _m-1 , the bit is not allocated to the silent channel, and the bit is adaptively allocated between the channels of the sound. I do.

Further, preferably, the power spectrum of the input signals S _{0 to} S _m-1 and the masking levels M _{0 to} M _m-1 considering the auditory characteristics are obtained for each frame unit of each channel signal, and the masking is performed in the entire band. If the level is above the power spectrum, the channel is silenced; otherwise, it is voiced.

[0015]

Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals denote the same or corresponding parts throughout the drawings. FIG. 2 is a diagram showing the configuration of the quantized bit allocation system of the embodiment, which shows an example of the configuration of a band division coding system in which two-channel stereo audio signals S ₀ and S ₁ are input. .

In the figure, 1 ₀ and 1 ₁ are band division filters, 2 ₀ and 2 ₁ are fast Fourier transform units (FFT),
3 ₀ and 3 ₁ are auditory calculation units, 4 ₀ and 4 ₁ are quantization units,
5 ₀ and 5 ₁ are a code part, 7 is a frame combining part, and 1
0 is a bit allocation unit of the embodiment. The operation of the parts other than the bit allocation unit 10 is the same as that described with reference to FIGS. Hereinafter, the bit allocation unit 1
The operation of 0 will be described.

FIG. 3 is a flow chart of the voiced / non-voiced determination process of the embodiment. In the bit allocation unit 10, first, the presence / absence of sound is determined based on the outputs of the auditory calculators 3 ₀ and 3 ₁ . That is, a channel signal having a negative signal-to-mask ratio SMR in all bands is determined to be silent, and if SMR is positive in one band, it is recognized as sound. Specifically, steps S1 and S2
Then, "0" is set to each of the channel counter m and the frequency band counter n. SMR in step S3
It is determined whether or not (m, n)> 0.

If SMR (m, n)> 0, this band is silent, so the flow advances to step S4, and the band counter n
To +1. In step S5, it is determined whether or not n = K (for example, 32). If not n = K, the process returns to step S3,
Inspect the next band. In this way, the same processing as described above is repeated, and when n = K eventually, the input signal of this channel has no sound component over the entire band, so that the flag AF (m) is set to 0 (silence) in step S6. Set. Moreover, SMR (m, n)>
If 0 is detected, the process proceeds to step S9 and the flag AF
Set 1 (voiced) to (m).

In step S7, the channel counter m is set to +
Do 1 In step S8, it is determined whether or not m = L (for example, 2). If not m = L, the process returns to step S2 to inspect the next channel. If m = L, the process is exited. Figure 4
6 is a flowchart of a quantized bit allocation process according to the embodiment. When the voiced / non-voiced determination process is completed, the process is continuously input to this process.

In step S21, 0 is set in the channel counter m. In step S22, the register MIN holding the minimum value of the mask-to-noise ratio MNR is previously stored in the MNR
The predetermined maximum value MAX of is set. Step S2
In 3, it is checked whether or not flag AF (m) = 1 (sound), and A
If F (m) = 1 is not satisfied, it is not necessary to allocate a quantized bit because this channel is silent, so step S32 is performed.
Then, the channel counter m is incremented by one. Step S3
In 3, it is determined whether or not m = L. If m = L, the process returns to step S23 to inspect the next channel. AF
If (m) = 1, this channel is voiced, so the flow advances to step S24 to set 0 in the band counter n.

In step S25, MNR (m, n) = M
Whether or not it is AX is determined. If MNR (m, n) = MAX, it is a band to which a quantized bit has already been allocated by the process described later, and therefore the process proceeds to step S30, and the band counter n is incremented by one. In step S31, it is determined whether or not n = K. If not n = K, the process returns to step S25, and the MNR of the next band is inspected. If MNR (m, n) = MAX is not satisfied, the process proceeds to step S26 to compare MNR (m, n) with the contents of the register MIN.

When MNR (m, n) <MIN, a new low MNR is detected, and therefore the process proceeds to step S27, in which the current MNR (m, m, is stored in the register MIN holding the minimum value.
Set the contents of n). In step S28, the recording of one or more (m, n), which has been recorded as being the same as the content of the register MIN by the processing described later, is canceled. Then, in step S29, (m, n) of the newly minimum MNR (m, n) is recorded.

When MNR (m, n) = MIN,
Since the MNR (m, n) is the same as the smallest MIN content at the present time, the process proceeds to step S29, where (m, n)
Record n). Furthermore, when MNR (m, n)> MIN, the MNR (m, n) is the smallest MI at the present time.
Since it is higher than the content of N, the process proceeds to step S30. Thus, the minimum MNR (m,
n) is detected and its (m, n) is recorded.

When the detection and recording of the minimum value described above are completed, the flow proceeds to step S34, where it is first determined whether or not the content of the register MIN is equal to MAX. If MIN = MAX, it means that the content MAX set in the register MIN in step S22 has not been rewritten by this time MNR (m, n) lower than that. This is because, for example, when all channels are previously determined to have AF (m) = 0 (silence), or before the quantized bit allocation process described later is completed before the total number of bits on the composite frame is used up. This can happen if it happens. In any case, in this case, the processing is immediately terminated. If MIN = MAX is not satisfied, the process proceeds to step S35 to adaptively allocate quantized bits.

FIG. 5 is a diagram showing a specific example of quantized bit allocation. Here, for simplicity of explanation, the number of channels L
= 2, the number of band divisions K = 16 is shown. In this example, it is the band-divided signal B ₀₈ of m = 0, n = 8 that has the smallest MNR that is first assigned a quantized bit. The bit allocation unit 10 considers the value of MNR (0,8) and, for example, the quantization bit number D ₀₈ (for example, 6 so that the value of the quantization noise Δ / 2 becomes smaller than the value of MNR (0,8). bit)
Assign In this way, since the quantization noise in this band is smaller than the masking level in this band, this quantization noise is not recognized in the reproduced sound. In other words, S
NR increases by, for example, α [dB], and MNR = SN
Due to the R-SMR relationship, the MNR value also increases by α [dB].

Returning to FIG. 4, in step S36, a predetermined maximum value MAX is set for each MNR (m, n) to which the quantized bit has been assigned so that duplicate assignment is not performed. In step S37, the cumulative number of bits to which the quantized bits have been assigned so far is subtracted from the total number of bits on the composite frame, and it is determined whether or not the result is> 0. Result is> 0
If so, the process returns to step S21 to detect and record the next minimum value of MNR.

Returning to FIG. 5, in this example, the second quantized bit is assigned m = 0, and n = 6, 7, 9
Band-division signals B ₀₆ , B ₀₇ , and B ₀₉ . The bit allocation unit 10 performs the quantization bit numbers D ₀₆ , D ₀₇ , D _{09 in the same} manner as above.
For example, 5 bits are allocated to each. The allocation of the number of bits is performed so that the MNR value is even in each band. At this point, the cumulative number of bits to which the quantized bits have been assigned is 21 bits in total, which is smaller than the total number of bits (for example, 64 bits) on the composite frame.

Returning to FIG. 4, the above-described minimum value detection and recording processing is repeated, and eventually step S3.
When the determination of 7 is NO, the process is exited. Returning to FIG.
In this example, finally, the quantized bits are allocated up to the sixth MNR from the bottom, and as a result, the cumulative number of bits is 56 bits in total for channels 0 and 1. In this example, when the value of MNR (m, n) is larger than the predetermined value, it is not necessary to allocate the quantized bit, so the quantized bit allocation is stopped midway.

FIG. 6 is a diagram showing the structure of a quantized bit allocation system of another embodiment, which shows the structure when the present invention is applied to a backward adaptive coding system. 20 in the figure
Is a bit allocation unit of another embodiment. Bit allocation unit 20
Basically uses, for example, the redundancy included in the amplitude of the code data C ₀ , so that, for example, the current quantization width D _0n is based on the code data C _{0 (n-1)} at the _immediately preceding time point. D _0n = f such that the number of quantization bits is always constant M.
It is adaptively changed by (C _{0 (n-1)} ). Similarly for the current quantization width D _1n of channel 1, D _1n = f (C
_{1 (n-1)} ) to adaptively change.

Even in this case, for example, it is almost unnecessary to allocate the quantized bits to the channel 0 where the silence or noise is the main, and if the extra quantized bits are passed to the channel 1, for example, the input signal S For example, ₁ can always be quantized with constant (M + 1) bits regardless of its amplitude. Therefore, the S / N ratio of channel 1 can be further improved. Therefore,
The bit allocation unit 20 collectively evaluates the amplitudes of the code data C ₀ and C ₁ and, based on the evaluation result, within a range that does not exceed the total number of bits of the combined frame, the input signals S ₀ and S ₁ are Bit allocation is adaptively applied to the quantization in each channel so that the quantization error is minimized. Of course,
The decoding side can correctly reproduce the signal by the reverse algorithm.

In the case of the embodiment shown in FIG. 2, when all channels are silent, for example, total bits are divided into L, and bit allocation is performed for each channel so that the MNR value of each band becomes equal. You can go. Further, although a plurality of characteristic configurations are shown in the above embodiment, the quantized bit allocation method of the present invention can be applied to various other coding methods.

[0032]

As described above, according to the present invention, predetermined characteristics of each channel signal are collectively evaluated, and
Based on the evaluation result, the bit allocation is adaptively performed for the quantization in each channel so that the quantization error of each channel signal is minimized within a range that does not exceed the number of transmission frame bits of the system. It is possible to effectively use the transmission frame bit and improve the quality of the reproduced signal.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a diagram showing a configuration of a quantized bit allocation system according to an embodiment.

FIG. 3 is a flowchart of a sound / silence determination process according to the embodiment.

FIG. 4 is a flowchart of a quantized bit allocation process according to the embodiment.

FIG. 5 is a diagram showing a specific example of quantized bit allocation.

FIG. 6 is a diagram showing a configuration of a quantized bit allocation method according to another embodiment.

FIG. 7 is a diagram showing a configuration of a conventional quantized bit allocation system.

FIG. 8 is a diagram for explaining the principle of quantized bit allocation in the configuration of FIG.

[Explanation of symbols]

 1 band division filter 2 high-speed Fourier transform unit 3 auditory arithmetic unit 4 quantization unit 5 coding unit 7 frame synthesis unit 10 and 20 bit allocation unit

Front page continuation (72) Inventor Koji Okazaki 1015 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited (72) Inventor Naoji Matsuo 1015, Kamedotachu, Nakahara-ku, Kawasaki, Kanagawa Prefecture (72) Invention Kiichi Matsuda 1015 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited

Claims (4)

[Claims] 1. In a quantized bit allocation system in a system in which signals of a plurality of channels are respectively quantized and encoded, and then these are frame-synthesized and transmitted, a predetermined property of each channel signal is evaluated together. , Based on the evaluation result, bit allocation is adaptively applied to the quantization in each channel so that the quantization error of each channel signal is minimized within a range not exceeding the number of transmission frame bits of the system. Quantized bit allocation method. 2. The predetermined property for each channel signal is an amplitude of an input signal, an amplitude of a quantized signal, or a masking level considering the power spectrum of the input signal and noise and human auditory characteristics. The quantized bit allocation method according to claim 1. 3. A method for determining whether sound is present or not for each frame unit of each channel signal, and bit allocation is not performed for silent channels, and bit allocation is adaptively performed between sound channels. The quantized bit allocation system according to claim 1. 4. A power spectrum of an input signal and a masking level in consideration of auditory characteristics are obtained for each frame unit of each channel signal, and if the masking level exceeds the power spectrum in the entire band, there is no sound, otherwise. Is voiced. 4. The quantized bit allocation system according to claim 3, wherein