JP2001077698A - Method for deciding block size with respect to audio encoding application - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an efficient and simple block size deciding method, in the encoding technique of a digital audio signal. SOLUTION: This block size deciding methods divides the PCM sample of a conversion inputting buffer into K sub-blocks of an equal length, obtains the difference from a PCM sample value by each sub-block, detects the peak of the difference and the peak of the PCM sample value by each sub-block, modulates the peak of the difference by the peak of an audio PCM sample value by each sub-block to obtain a modulated value, detects the peak (peak sub-block) of the modulated value from all the sub-blocks, calculates the ratio of the modulated value between the peak sub-block and all of the other sub- blocks appearing before the peak sub-block, compares the ratio with a prescribed threshold to detect an attack signal, applies a short block size when the attack signal is detected, and applies a long block size, when it is not detected.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technology for efficiently encoding digital audio signals for data transmission or digital storage media.

[0002]

2. Description of the Related Art Audio encoding algorithms generally fall into three categories. That is, it is classified into subband coding, transform coding, and coding in which subband codes and transform codes are mixed. The subband encoder includes, for example, an ISO / IEC11172-3 MPEG-1 audio layer 2 algorithm. The transform encoder, for example,
Includes ISO / IEC13818-7 MPEG-2 Advanced Audio Coding (AAC) algorithm. For example, a hybrid encoder for subband and transform codes includes the ISO / IEC 11172-3 MPEG-1 audio layer 3 algorithm used in minidisc systems.

[0003] A typical transform coder first converts input audio PCM samples to input PCM as shown in FIG.
The buffer 21 buffers the encoded frame.
The encoding process is performed for each frame. Each frame of the PCM sample is converted to a block size
2 and the window processing and conversion module 23 at the same time. In the block size determination module 22, an appropriate predetermined block size is selected for window processing (window function processing) and conversion processing. That is, usually, two types of block sizes, a long block and a short block, are used. The long block size is used for stationary signals, and the short block size is used for attack signals. The block size information is input to the windowing and transform module 23 and is used to window the buffered PCM samples and transform them into spectral coefficients. Further processing is performed on these spectral coefficients before they are transmitted to the decoder together with side information.

[0004] The determination of the block size plays an important role in removing the pre-echo noise generated in the transform encoder. Pre-echo noise is caused by performing a transform in long blocks on an attack signal characterized by a quiet period before a sudden increase in signal strength. The quantization noise spreads throughout the transform block and has a significant effect on the quiet part of the attack signal. As such, quantization noise becomes more pronounced in quiet parts before sudden loud noises appear, and is thus described as pre-echo noise. By dividing the long block into multiple short blocks and using a short block transform for each short block, the quantization noise is confined within each short block. This allows pre-
Echo noise can be reduced. Generally, a long block transform is performed on a static signal so as to realize better coding efficiency with a high frequency resolution configuration. on the other hand,
Short block transforms are used on attack signals to reduce pre-echo noise. Therefore, it is important that the block sizing method be able to accurately classify signal types.

[0005] The present invention relates to the determination of block size in audio transform coding. In particular, it is MPEG
-2 It can be used for the AAC algorithm. MP
In the EG-2AAC algorithm, the transformation process is a modified discrete cosine transform (MDCT) originally proposed by Princen and Bradley.
Transform): Prinsen and Bradley, "Subband / Transfor
m Coding Using Filter Bank Designs Based on Time D
omain Aliasing Cancellation ", ICASSP Proceedings 1987, pp
2161-2164).

[0006] Two types of block sizes are predetermined, with long blocks using 2048 PCM samples and short blocks using 256 PCM samples. When short blocks are used, the processing frame is divided into eight sub-blocks of 256 PCM samples, each sub-block being the MDC of the short block.
Processed by T.

FIG. 6 shows an ISO / IEC13818-7.
The following describes a block size determination method described in Supplement B (informational part) of the MPEG-2 AAC standard.
It is based on the value of perceptual entropy calculated by the proposed psychoacoustic model. In this method,
The input audio PCM sample is first buffered into a frame of 2048 samples (S61). These sample frames are converted into spectral coefficients by FFT (S62). The degree of unpredictability is calculated based on the obtained spectral coefficients (S63). The threshold is calculated (S64), and the perceptual entropy is calculated (S6).
5). The perceived entropy is determined by a predetermined decision threshold swit
Compare with ch_pe (S66). If the perceived entropy is greater than a predetermined decision threshold switch_pe, a short block size is selected for MDCT (S6).
8). Otherwise, the long block size is MDCT
(S67). Predetermined decision threshold sw
itch_pe is implementation dependent.

[0008]

There is a need for an efficient block size determination method for removing pre-echo noise generated in a transform encoder. While the actual block size used during the conversion is an important factor in itself, accurate detection of signal attacks, especially critical ones, is very important. In general, the use of long blocks is desirable for transform coding of audio signals that provides better frequency resolution to achieve high redundancy and improper rejection. This is especially true for segments of the audio signal where the characteristics of the audio signal change slowly. Short blocks can only be used for segments consisting of critical attack signals.

[0009] The prior art is not very efficient in determining the correct block size. Its accuracy depends heavily on psychoacoustic models, which are very complex processes.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to provide an efficient and simple block size determination method.

[0011]

SUMMARY OF THE INVENTION The present invention provides pre-masking and post-masking.
g) The temporal masking (temporal masking) condition is taken into consideration. Pre-masking is a condition caused by a fast rise of loud sounds, such that sounds occurring before an attack are masked. On the other hand, post-masking is a loud masker (loud maske).
This is the lingering effect of the sound caused by the loud masker, in which the sound generated after r) is masked. Psychoacoustic experiments on pre-masking and post-masking effects show that pre-masking lasts between 5 and 20 msec, and that post-masking lasts between 20 and 200 msec, depending on the duration of the masker. In the present invention, based on the above experimental results, since the pre-masking period is relatively short, only the signal attack is detected. Since the post-masking period is relatively long, the release of a signal is ignored.

The present invention comprises the following means. That is, means for dividing the PCM sample of the conversion input buffer into K sub-blocks having an equal length of about 6 msec in consideration of an empirical pre-masking period, and for each of the sub-blocks, Means for calculating a difference; means for detecting the peak of the difference and the peak of the PCM sample value for each of the sub-blocks; and means for detecting the peak of the difference for each of the sub-blocks by the peak of the audio PCM sample value. Means for modulating and generating a modulation value; means for detecting a peak (peak sub-block) of the modulation value from all sub-blocks; the above-mentioned peak sub-block; and other appearing before the peak sub-block. Means for calculating the modulation value ratio between all the sub-blocks and comparing the modulation value ratio with a predetermined threshold value; Means for detecting an attack signal, and means for applying a short block size when an attack signal is detected, and applying a long block size when no attack signal is detected.

Alternatively, the present invention may be constituted by the following means. That is, the audio P of the conversion input buffer
Means for dividing one frame of a CM sample into K sub-blocks at equal time intervals, means for obtaining the difference from audio PCM samples for each sub-block, means for detecting the peak of the difference for each sub-block Means for detecting the maximum value of the difference peak among all the sub-blocks; means for labeling the sub-block containing the maximum value of the difference peak as the maximum sub-block; and a sub-block appearing before the maximum sub-block. Means for calculating the ratio of the peak of the difference between all of the
Means for comparing each ratio with a predetermined threshold value, and means for determining a block size based on the comparison result and the position of the largest sub-block in the frame.

By utilizing the pre-masking effect, the audio PCM samples in the conversion input buffer are divided into equal length sub-blocks of approximately 6 msec. this is,
Since the occurrence of pre-echo noise is masked for an attack occurring in a sub-block, it means that the noise is easily ignored. This also has the advantage of reducing the complexity of the block size determination method. Modulated value (modulation value) to detect the presence of an attack signal in the current frame
Is compared to a predetermined threshold. If any of those ratios exceed the threshold, it indicates that an attack signal is present, at which time a short block size is selected for the transform. The modulation value for each sub-block is obtained by modulating the peak of the difference with the peak of the PCM sample value detected in the sub-block. The difference is derived by taking the difference between their values for every two adjacent PCM samples. To determine their ratio, first the peak of the modulation value and its sub-blocks are identified, and the ratio between the modulation value of all the sub-blocks that appear before the peak of the sub-block and the peak of the modulation value is determined. Can be By doing so, only the attack signal is considered and not the release signal. Further, it is possible to detect both an attack that rises slowly and an attack that rises quickly.

As described above, the block size determination method of the present invention can be easily realized without using a complicated psychoacoustic model or FFT processing. Further, while the prior art determines the block size to be used depending on simultaneous masking in the frequency domain, the present invention determines the block size to be used by using the temporal masking effect of pre-masking and post-masking. .

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a block size determining method according to the present invention will be described below with reference to the accompanying drawings.

In the following, the use of an efficient block size determination method will be described as an example using the MPEG-2 AAC algorithm. This can be applied to other audio coding algorithms. MPEG-2 AAC
The transform used in the algorithm is the Modified Discrete Cosine Transform (MDCT). In addition to the MDCT, other transforms based on a predetermined block such as a discrete Fourier transform (DFT), a discrete cosine transform (DCT), and a discrete sine transform (DST) may be used.

The transform encoder for executing the block size determining method according to the present embodiment has the configuration shown in FIG. 2, and a block size determining module 22, which is one of the components, is described below. Realizing the processing.

FIG. 1 shows a flowchart of a method for determining a block size according to the present invention. In step S11, M
Divide the PCM samples of the DCT input buffer into K sub-blocks of equal length. These sub-blocks take approximately 6 msec to account for the empirical pre-masking period.
Have equal periods of time. FIG. 3 is a diagram showing how the sub-blocks are divided at this time. In step S12, the value of the PCM sample (pc
m _i ) is converted to a difference value (diff _i ) for every two adjacent PCM samples. This is described as follows: diff _i = pcm _{i + 1} -pcm _i , 1 ≦ i ≦ N−1 (1) where N is equal to the number of PCM samples in each sub-block. FIG. 4 is a diagram showing an example of a process of converting such a PCM sample into a difference value (d _{i in the} figure).

At step S13, the difference value peak (D ₁ ... _DK ) (hereinafter referred to as “peak difference value”) and the PCM sample value peak (S ₁ .
S _K ) (hereinafter referred to as “peak sample value”). Here, each peak _{(D 1 ... D K),} (S 1 ... S K)
Can be obtained by comparing the absolute value of the difference value or the PCM sample value. In step S14, the peak difference value (D ₁ ... _DK ) is converted to a peak sample value (S
₁ ... S _K ) to obtain values (P ₁ ... P _K ) (hereinafter referred to as “modulation values”) modulated for each sub-block.
This is represented by the following equation. P _i = S _i · D _i , 1 ≦ i ≦ K (2)

In step S15, the modulation value having the maximum absolute value among the modulation values P _i of all the sub-blocks is detected, and the sub-block including the maximum modulation value (hereinafter referred to as “peak sub-block”). ) Index number of pe
Label as akIndex. This is shown as follows. P _peakIndex = max (P ₁ … P _K ) (3)

In step S16, the index number peakIndex is compared with the index number of the first (first) sub-block. If they are equal,
A long block size is selected for MDCT processing (S19). Thus, the first sub-block has a peak
The reason why a long block size is selected when a sub-block is used is that the psycho-aural phenomenon of the post-masking effect is used. That is, if a signal attack occurs in the first sub-block, a post-masking effect occurs, and the temporal noise of the signal appearing after the attack is masked. Thus, in determining the block size, the position of the peak sub-block in the frame is also considered.

In step S16, if the index number peakIndex is not equal to the index number of the first sub-block, the modulation value between the peak sub-block and all other sub-blocks appearing before the peak sub-block. Is calculated (S17). The ratio R _j is defined as follows. R _j = P _peakIndex / P _j , 1 ≦ j ≦ (peakIndex-1) (4)

FIG. 5 illustrates a specific example of the calculation of this ratio. That is, in FIG. 5, the sub-block K-2 is the peak sub-block, and the ratio R _j is calculated for the sub-blocks from the sub-block 1 to the sub-block K-3 that appear before it.

In the above description, the ratio R _i is obtained based on the modulation value P _i obtained by modulating the peak difference value with the peak sample value. Based on this, the ratio R _i may be determined. That is, a peak difference which is the largest among the peak differences (D ₁ ... _DK ) of all the sub-blocks is detected, and a sub-block (the largest sub-block) including the maximum peak difference and each sub-block appearing before the sub-block. the ratio R _i of the difference may be determined in the same manner as above with the.

In step S18, the ratio R of each modulation value
_j is compared to a predetermined decision threshold d_threshold. This decision threshold depends on the device. If at least one ratio exceeds the decision threshold, the short block size is MD
Selected for CT processing (S20). Otherwise, a long block size is selected for MDCT processing (S19).

In the MDCT processing, when a short block size is selected, a short window for encoding with a short block size is used, and when a long block size is selected, a long block is selected. Long window for encoding by size (Long
Window) is used. By the way, MPEG2 AA
In C and MP3, actually, in addition to the above-mentioned short window and long window, when switching these windows, a window such as a start window (Start Window) or a stop window (Stop Window) is used. The present invention can be applied to other encoding systems using windows used during such transitions.

[0028]

According to the present invention, there is provided a method effective for detecting an attack of a signal and reducing the occurrence of pre-echo for an audio transcoder. It can be more efficient in terms of memory utilization and required processing time, with less complexity than the prior art. Therefore, the present invention
This is effective for reducing the cost in realizing an audio conversion encoder by an LSI.

[Brief description of the drawings]

FIG. 1 is a flowchart of a block size determination method according to the present invention.

FIG. 2 is a block diagram showing a configuration for a conversion process in an audio conversion encoder.

FIG. 3 is a view for explaining how one frame of an input PCM buffer is divided into sub-blocks.

FIG. 4 is a view for explaining conversion from PCM samples to difference values.

FIG. 5 is a diagram illustrating calculation of a modulation value ratio.

FIG. 6 is a flowchart of a block size determination method according to the related art.

[Explanation of symbols]

 21 Input PCM Buffer 22 Block Size Determination Module 23 Window and Conversion Module

Claims (14)

[Claims] 1. A method for adaptively determining the choice of transform block size for an audio transform encoder, comprising: a) dividing one frame of audio PCM samples in a transform input buffer into K sub-blocks at equal time intervals; B) for each of the sub-blocks, the difference is obtained from the audio PCM sample; c) for each of the sub-blocks, the difference peak and the audio PCM sample peak are detected; For each sub-block, the peak of the difference is modulated by the peak of the audio PCM sample to generate a modulation value. E) The peak of the modulation value is detected, and the sub-block containing the peak of the modulation value is peak-sub- Labeling as a block, f) all sub-blocks appearing before the peak sub-block and the peak Calculating the ratio of the modulation values to and from the sub-block; g) comparing each of the ratios with a predetermined threshold; h) comparing the comparison result with the position of the peak sub-block in the frame. Determining a block size based on the block size. 2. A method for adaptively determining the selection of a transform block size for an audio transform encoder using differential PCM samples, comprising: a) one frame of audio PCM samples in a transform input buffer at equal time intervals; Dividing into K sub-blocks, b) for each of the sub-blocks, finding the difference from the audio PCM sample, c) detecting the peak of the difference for each of the sub-blocks, d) detecting all the sub-blocks Detecting the maximum value of the difference peak in the block; and e) identifying the sub-block containing the maximum value of the difference peak,
Labeling as the largest sub-block; f) calculating the ratio of the difference peaks between all of the sub-blocks appearing before the largest sub-block and the largest sub-block; H) determining a block size based on the comparison result and the position of the largest sub-block in the frame. 3. The transform is a discrete Fourier transform (DF)
3. The transform according to claim 1, wherein the transform is based on a predetermined block such as T), discrete cosine transform (DCT), discrete sine transform (DST) or modified discrete cosine transform (MDCT). Method. 4. The time interval between the sub-blocks is 5 ms.
3. A method according to claim 1 or claim 2, wherein the method is in the range between ec and 20 msec. 5. The method according to claim 1, wherein the difference is obtained by calculating a difference between two consecutive audio PCM sample values in the sub-block. 6. The method according to claim 1, wherein the peak of the difference is obtained by comparing the absolute value of the difference in the sub-block. 7. The method according to claim 1, wherein the peak of the audio PCM sample is obtained by comparing absolute values of audio PCM samples in the sub-block. 8. The method according to claim 1, wherein the modulation process is a mathematical operation. 9. The method according to claim 1, wherein the peak of the modulation value is obtained by comparing modulation values in the frame. 10. The method according to claim 1, wherein the predetermined threshold value is device-dependent. 11. The method according to claim 1, wherein when any one of the ratios is higher than the predetermined threshold, the block size is determined to be a short block size. 12. The apparatus according to claim 1, wherein the block size is determined to be a long block size when all of the ratios are equal to or less than the predetermined threshold value.
The described method. 13. The method of claim 1, wherein the block size is determined to be a long block size when the peak sub-block is the first sub-block in the frame. 14. The method of claim 2, wherein the block size is determined to be a long block size when the largest sub-block is the first sub-block in the frame.