JP2003005797A - Audio signal encoding method and apparatus, and encoding and decoding system - Google Patents

Abstract

(57) [Problem] To accurately suppress a pre-echo noise and a post-echo noise by generating a correction function more accurately and performing gain control. A subband frame including an audio sample signal is divided into a plurality of sections, and an absolute value of a peak in each section is obtained. Identify the segment (attack signal) that includes the sudden onset of sound, amplify the segment between the identified segment of the attack signal and the previously modified segment, process the attack signal, and The release signal is identified by attenuating the section between the identified release signal section and the previously modified section, and processing the release signal by attenuating the section between the identified release signal section and the previously modified section. And prevent the occurrence of post-echo.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to, for example, ATRAC.
The present invention relates to an audio signal encoding method and apparatus, and an encoding and decoding system, which are applied to an encoding method such as 3), and in particular, pre-echo noise generated before an attack signal which is a sharp rising signal, and The present invention relates to an audio signal encoding method and apparatus, and an encoding and decoding system, for suppressing post-echo noise generated after a release signal which is a sharp falling signal.

[0002]

BACKGROUND OF THE INVENTION The classical problem of pre-echo noise and post-echo noise is well known in transform coding of audio signals (e.g. JP 2000-25919 A).
See Publication No. 7. ). FIG. 11A is a block diagram showing the configuration of a conventional audio signal encoding / decoding device when pre-echo and post-echo occur without gain control, and FIG. 11B shows gain control. FIG. 3 is a block diagram showing a configuration of a conventional audio signal encoding / decoding device when performing pre-echo and post-echo.

An encoder having a modified discrete cosine transform (hereinafter referred to as MDCT) processing unit 101 and a quantizer 102 shown in FIG. 11A, and an inverse quantizer 103.
And a decoder including the inverse MDCT processing unit 104, quantization when a typical transform block including an attack signal (that is, a signal including a rapid rising of a sound) is input to the encoder. The unit 102 introduces noise, and the decoder then distributes the noise evenly in the time domain in the reconstructed block. That is, the phenomenon in which the pre-echo noise and the post-echo noise occur in the restored transform block is that the quantization noise generated by quantizing the attack signal is evenly distributed on the transform block including the attack signal. Is triggered. The attack signal is characterized by the presence of periods of small amplitude, all within the transform block, before the sudden increase in signal strength.

Pre-echo is audibly more annoying than post-echo for two reasons. First, the pre-echo completely covers the small amplitude signal in which it appears, unlike the post-echo, which tends to hide under the portion of the attack signal of high intensity. Second, the human ear has the masked sound in front of the attack signal portion in time compared to the forward masking duration, in which the masked sound lies behind the attack signal portion in time. The reverse masking duration to be characterized is much shorter. Specifically, the forward masking duration is about 200 msec and the backward masking duration is about 5 msec.

There are three well-known methods for dealing with this problem: (A) Block size switching method; (b) Temporal noise shaping method
ping; TNS); and (c) gain control method.

Here, the block size switching method divides a standard conversion block into shorter partial blocks and applies a shorter conversion to each partial block. This is
The pre-echo noise spread will be limited to within short sub-blocks. Also, the noise waveform shaping method in the time direction utilizes the time-frequency duality by applying predictive coding to the spectral data, since the attack signal is characterized by its non-uniformity in the time domain. . Finally, the gain control method uses a modification function to modify the signal in the time domain by amplifying the signal strength in the small amplitude part and / or suppressing the signal strength in the large amplitude part.

One example of a gain control method according to the prior art is shown in FIG. 11 (b). Here, FIG. 11 (a)
Gain control unit 1 for the encoder and the decoder of
05 and an inverse gain control unit 106 are provided. Figure 11
When a conversion block including an attack signal similar to that in (a) is input, the gain control unit 105 amplifies a weak signal and / or suppresses a strong signal. This amplification and / or suppression information is sent to the decoder together with the encoded signal, and based on the above information, the inverse gain control unit 106 suppresses the amplified part in the gain control unit 105, and /
Alternatively, the suppressed part is amplified. The quantizer 102 introduces noise and evenly distributes the noise in the restored block in the time domain, but the inverse gain controller 106
In the latter stage, noise is suppressed in a region where the signal has a small amplitude in advance, as compared with the example without gain control in FIG. 11A.

[0008]

A good way to address the problems of pre-echo noise and post-echo noise must consider several factors. It is important to capture the onset of the attack signal in the signal so that the small amplitude signal preceding the attack signal is amplified not too early and not too late. If the signal is amplified too quickly, the pre-echo will not be effectively removed. Amplifying the signal too slowly risks amplifying the attack signal itself, thus exacerbating the pre-echo problem. Similarly, the increase in signal strength as a result of amplifying a signal of small amplitude causes the increase of signal of large amplitude so that the increase in signal energy (surge) does not overwhelm the subsequent processing of spectral quantization. It must be balanced by a suitable amount of damping. The attenuation should not be excessive so as not to reduce the accuracy of the attenuated signal when it is reconstructed at the decoder. In the worst case, excessive attenuation can introduce additional artifacts that are less annoying than the post-echo we are trying to eliminate.
Therefore, the method for setting the range to which the damping should be applied is also important. If all of the above factors are taken into account, then a correction function can be generated and the generated correction function can be multiplied to the frame in order to achieve gain control. The correction function should not cause overflow of the sampled signal between the previous frame and the current frame.

Since the release signal, which is a signal including a sudden fall of sound, is involved in the generation of post-echo noise, just as the attack signal is involved in the generation of pre-echo noise, the influence of such release signal is also exerted. You must also suppress it.

An object of the present invention is to solve the above problems and to more accurately generate a correction function to execute gain control, thereby making it possible to reliably suppress pre-echo noise and post-echo noise. An object of the present invention is to provide an encoding method and apparatus, and an encoding and decoding system.

[0011]

Gain control using a correction function is a popular method for mitigating pre-echo and post-echo artifacts in many audio encoders. In the present invention, in generating the correction function, a point (position) for gain control,
We present an efficient way to realize determining gain control level (amplification factor and attenuation factor) and safeguard against overflow.

The present invention includes the steps of storing and delaying a plurality of sections of a frame in a buffer memory to process an attack signal that crosses frame boundaries; detecting the beginning of the attack signal; Determining the amplification factor applied to the signal preceding the attack signal, determining the position where the attenuation should be applied, calculating a value for said attenuation,
Preventing overflow in the current frame and the previous frame.

The audio signal encoding method according to the present invention comprises the steps of calculating a correction function based on the input audio signal and controlling the gain of the audio signal according to the calculated correction function; A step of performing an orthogonal transform process on the controlled audio signal on a frame-by-frame basis and performing an encoding process to obtain an encoded bitstream signal. Divides the input audio signal into time sections that are shorter than the frame time, delays the audio signal by at least one section time, and processes the delayed audio signal and the frames before and after that frame. Calculating a correction function based on at least three frames including That.

In the audio signal encoding method, the step of controlling the gain is preferably (a).
Using the step of obtaining the absolute value MaxPeak of the peak of each section, and (b) using a predetermined attack signal reference, identify an attack signal that is a part of the signal that includes a rapid sound rise. Step, (c)
By calculating a correction function based on at least three frames including a frame to be processed and frames before and after the frame, and performing gain control based on the calculated correction function,
Processing the attack signal to amplify a plurality of partitions between the identified partition of the attack signal and the partition that is the previous modified position, the criteria for the attack signal being: Max Peak related to the above categories
And an attack signal based on the highest value MaxPeak of a predetermined number of sections preceding the next section.

In the audio signal encoding method, the step of controlling the gain is preferably based on the at least three frames, and the correction function is related to the next section from the section to be currently processed. MaxPea
The calculation is based on k and the maximum value MaxPeak in a predetermined number of sections preceding the section to be currently processed.

Further, in the above audio signal encoding method, preferably, (a) a segment to be currently processed,
Using the step of obtaining the absolute value InterModMaxPeak of the peak with respect to the section which is the correction position before that, and (b) using a predetermined release signal standard, a sudden fall of the sound in the audio signal is caused. Identifying a release signal that is a portion of the signal that includes, and (c) calculating a correction function based on at least three frames including the frame to be processed and the frames before and after it, and based on the calculated correction function. Gain controlling to further amplify a plurality of sections between the identified section of the release signal and a section that is a correction position behind the section to process the release signal. The signal is based on the latest InterModMaxPeak, the maximum value MaxPeak of the following predetermined number of divisions, the division to be currently processed and the attack signal at the end. It is characterized in that the release signal is identified based on the time to and from the identified section.

Still further, in the audio signal encoding method, the modification function for processing the release signal is preferably the InterModMaxPeak and the maximum value MaxPeak of the subsequent predetermined number of partitions. It is characterized in that it is calculated based on.

The audio signal encoding apparatus according to the present invention calculates a correction function based on the input audio signal, and gain control means and gain control means for the audio signal according to the calculated correction function. An audio signal encoding device, which comprises means for performing an orthogonal transformation process on the encoded audio signal in frame units and performing an encoding process to obtain an encoded bit stream signal, wherein the gain control means comprises: , The input audio signal is divided into time sections which are shorter than the frame time, the audio signal is delayed by at least one section time, and the frame to be processed of the delayed audio signal and the frames before and after it are processed. Characterized in that the correction function is calculated based on at least three frames including.

In the audio signal encoding apparatus, the gain control means preferably uses (a) means for obtaining the absolute value MaxPeak of the peak of each section and (b) a predetermined attack signal reference. A means for identifying an attack signal which is a part of a signal including a sudden rise of sound in the audio signal, and (c) a modification based on at least three frames including a frame to be processed and frames before and after the frame. Calculating a function and controlling gain based on the calculated correction function so as to amplify a plurality of sections between the section of the identified attack signal and the section which is the previous correction position. And a means for processing the attack signal, the reference of the attack signal,
It is characterized in that the attack signal is identified based on the MaxPeak relating to the next division and the highest value MaxPeak of a predetermined number of divisions preceding the next division.

Further, in the audio signal encoding apparatus, preferably, the gain control means, based on the at least three frames, the correction function, the MaxPeak relating to a next section from a section to be currently processed. And a maximum value MaxPeak in a predetermined number of sections preceding the section to be currently processed.

Further, in the above audio signal encoding apparatus, preferably, (a) a segment to be currently processed,
A means for obtaining an absolute value InterModMaxPeak of a peak between a section which is a correction position before that and (b) a predetermined release signal reference, and in the audio signal,
Calculating means for calculating a correction function based on at least three frames including a frame to be processed and frames before and after the means for identifying a release signal which is a part of a signal including a sharp fall of sound; Means for amplifying a plurality of sections between the identified section of the release signal and the section which is the correction position behind the section by controlling the gain based on the corrected correction function, and processing the release signal. The release signal criterion further comprises the latest InterModMaxPeak, the maximum MaxPeak of a predetermined number of subsequent segments, the segment to be currently processed and the last attack signal identified. It is characterized in that the release signal is identified based on the time to and from the partition.

Furthermore, in the audio signal encoding device, the modification function for processing the release signal is based on the InterModMaxPeak and the maximum value MaxPeak of the following predetermined number of sections. It is calculated.

An audio signal encoding / decoding system according to the present invention is an audio signal encoding / decoding system including the audio signal encoding device and the audio signal decoding device. The audio signal decoding device includes means for obtaining an audio signal composed of a plurality of frames by decoding the bit stream signal encoded by the encoding device and performing orthogonal transform processing; and the obtained audio signal. Means for performing inverse gain control on a signal using a correction function opposite to the above-mentioned correction function to obtain and output an audio signal subjected to inverse gain control.

[0024]

DETAILED DESCRIPTION OF THE INVENTION An embodiment of the present invention will be described below with reference to the drawings. In the present specification, ATRA
A mini disk recording / reproducing system for C3 will be described with reference to an embodiment, but the encoding method of the present invention is
It can also be applied to other encoders that use gain control.

FIG. 1 is a block diagram showing the configuration of a mini disk recording / reproducing system according to an embodiment of the present invention. This embodiment is particularly characterized by the detailed configuration inside the audio encoder 2.

In FIG. 1, an A / D converter 1 A / D converts an audio input signal to convert it into a digitized audio sample signal, and then an audio encoder 2 uses a gain control method as will be described later in detail. It is used to compress and encode the converted audio sample signal to generate an ATRAC3 bitstream signal. Next, the mini disk recording device 3 is set to ATRAC3.
After modulating the bit stream signal of 1 to a predetermined modulation recording signal, the modulation recording signal is recorded on the mini disk 4.
On the other hand, the mini-disc reproducing device 5 reproduces the recording signal from the mini-disc 4 and demodulates it so that the ATRAC3
Play the bitstream signal of. Further, the audio decoder 6 decodes the compressed bit stream signal of the ATRAC3 into a digital audio signal while controlling its gain by using an inverse gain control method which is the reverse of the above gain control method. The D / A converter 7 D / A converts the digital audio signal and outputs an audio output signal.

By the above gain control and inverse gain control,
It is possible to suppress pre-echo noise and post-echo noise that occur when encoding and decoding a digital audio signal. In this embodiment, the recording / reproducing system is configured to record the bit stream signal of the ATRAC3 on the mini disc 4, but the present invention is not limited to this, and recording is performed using another recording medium or transmission medium. A reproduction system and a transmission system may be configured.

FIG. 2 is a block diagram showing the detailed arrangement of the audio encoder 2 shown in FIG. In FIG. 2, the audio encoder 2 is a typical conversion encoder using the audio signal compression technique of the ATRAC 3, but among the detailed configuration of the audio encoder 2, in particular, the gain detection unit 14
-1 to 14-4 and gain control units 15-1 to 15-4
It has a feature in the detailed configuration of. ATRAC3
Is basically a hybrid coding method of sub-band coding and transform coding technology, which is configured for a stream of audio sample signals sampled at 44.1 kHz.

As shown in FIG. 2, the audio encoder 2 uses the 2-stage QMF filter 10 to divide the frequency band into four partial band (subband) signals, and then, in the time direction for each partial band. , A signal divided into frames each having 256 sample signals (hereinafter referred to as a subband frame. A total of four partial bands is 10
It has 24 sample signals. ) Is output. Here, in detail, the QMF division filter 10 further includes a low-pass filter 11 and a high-pass filter 12 that equally divide the audio sample signal into a low-pass signal and a high-pass signal, and further divides the divided low-pass and high-pass signals with respect to frequency. It is configured by including filter banks 13-1 to 13-4 that equally divide. Therefore, the output of QMF filter 10 is four downsampled subband frames, each having 256 sample signals. Each of them is half the length of the transform block for the MDCT transform, 0 to 5.5 kHz, 5.5 kHz to 11 kHz.
z, 11 kHz to 16.5 kHz, and 16.5 kHz
It is an audio signal obtained from each of the four frequency bands from z to 22 kHz.

The filter banks 13-1 to 13-
Each of the subsequent stages of 4 may further include a circuit unit that executes joint stereo processing. The circuit part of the joint stereo processing converts, for example, an audio signal input as a stereo signal into an average signal of the right channel signal and the left channel signal, gain data of the right channel signal, and gain data of the left channel signal. However, further compression of the audio signal can be achieved.

Then, a modification function is generated and gain control is performed for each subband frame. The signals of the sub-band frames output from the filter banks 13-1 to 13-4 are input to the gain detection units 14-1 to 14-4 and the gain control units 15-1 to 15-4, respectively.
Each of the gain detection units 14-1 to 14-4 includes a buffer memory 14a-1 to 14a-4 and a correction function calculation unit 14b-1.
To 14b-4, calculates and outputs a correction function for each subband frame, and then each gain control unit 15-1 to 15-4 uses each correction function generated above to generate a correction function for each subband. Perform gain control on band frames. Hereinafter, with reference to FIG. 4, a subband frame in the frequency band of 0 to 5.5 kHz output from the filter bank 13-1 will be described in particular.

FIG. 4 is a block diagram showing the generation of conversion blocks in the gain detection unit 14-1 and the gain control unit 15-1 preceding the MDCT processing unit 15-1.

In FIG. 4, the n-th sub-band frame (hereinafter referred to as sub-band frame n) to be gain-controlled among the input audio signals composed of streams of audio sample signals is modified. Previous subband frame n-1 to compute the function
And n + 1 are stored in the buffer memory 14a-1. The correction function calculator 14b-1b delays the sub-band frames n-1 to n + 1 by one segment as described later in detail, and then delays the three sub-band frames n.
The correction function MFn for the sub-band frame n is calculated based on -1 to n + 1, and then the multiplier MP1b multiplies the correction function MFn by the sub-band frame n to cascade the sub-band frame n of the multiplication result. Operator CO
1 and the multiplier MP2b. In addition, the correction function calculation unit 14b-1a determines the subband frames n-2 and n-1.
And n are delayed by one partition as described in detail below, and then a modification function M for subband frame n-1 is based on the delayed three subband frames n-2 to n.
Fn-1 is calculated, and then the multiplier MP1a multiplies the modified function MFn-1 by the subband frame n-1 and the resulting subband frame n-1 is multiplied by the multiplier MP2a.
Output to. Further, similarly, the correction function calculation unit 14b-
1c delays the subband frames n, n + 1 and n + 2 by one section as described in detail later, and then modifies the correction function MFn + 1 for the subband frame n + 1 based on the delayed three subband frames n to n + 2. And then the multiplier MP1c calculates the correction function MFn
The subband frame n + 1 is multiplied by +1 and the resulting subband frame n + 1 is output to the cascade connection operator CO2 and the multiplier MP2c.

Prior to the modified discrete cosine transform (MDCT) processing in the MDCT processing unit 16-1, two consecutive subband frames are cascaded with each other while guaranteeing the continuity of the correction function at the boundary of the subband frames. Connected. The gain of the multiplier MP2a is controlled by the correction function MFn-1, and the sub-band frame n-1 output from the multiplier MP1a is divided into the first section (the present embodiment) of the correction function MFn for the next sub-band frame n. In the above, the term "division" refers to the division when one subband frame is divided into 32 divisions.)
Normalization is performed by multiplying by n [0], and the subband frame n-1 as the multiplication result is output to the cascade connection operator CO1. Then the cascade operator CO1
The subband frame n-1 output from 2a and the subband frame n output from the multiplier MP1b are cascade-connected and output to the MDCT processing unit 16-1. Similarly, the multiplier MP2b converts the subband frame n output from the multiplier MP1a into the next subband frame n +.
The modified function MFn + 1 for 1 is normalized by being multiplied by the gain coefficient MFn + 1 [0] of the first section and is output to the cascade operator CO2. Next, the cascade connection operator CO2 cascade-connects the subband frame n output from the multiplier MP2b and the subband frame n + 1 output from the multiplier MP1c, and the MDCT processing unit 1
Output to 6-1. That is, the ATRAC3 standard is
It requires that the value of the modification function MFn be equal to 1 (ie, the gain is 1) at the endpoints of subband frame n, so that when combined with the next modification function MFn + 1, it is The multiplication with the gain factor MFn + 1 [0] of the first partition can be normalized to the same level at the boundaries of subband frames n and n + 1. Therefore, two sub-band frames that are combined have five adjacent blocks each.
After forming the entire transform block for MDCT having 0% overlap, it is output to the MDCT processing unit 16-1.

The multiplier MP described above with reference to FIG.
1a, MP1b, MP1c, MP2a, MP2b and M
The P2c and the cascade connection operators CO1 and CO2 are configured to be included in the gain control unit 15-1 in the present embodiment. In FIG. 4, only the processing regarding the first band is illustrated, but the processing regarding the second to fourth bands is similarly executed.

Referring again to FIG. 2, the MDCT processing unit 1
6-1 to 16-4 are gain controlled in each frequency band,
Further, a modified discrete cosine transform (MDCT) process is performed on at least two sub-band frame signals that are cascade-connected to a transform block for MDCT processing, and the resulting spectrum information signal is converted into tone components and The spectrum is output to the spectrum detector 17. The tone component / spectrum detector 17 separates the tone component signal and the non-tone spectrum signal from the MDCT-processed information signal, and separates them into the tone component encoder 1
8 and the spectrum coder 19 and the tone component coder 18 and the spectrum coder 19 in the subsequent stage are divided into a tone component signal and a non-tone spectrum signal, and individually perform coding processing such as quantization. do. Here, the tone component encoder 18 and the spectrum encoder 19 are each configured to include, for example, a scale factor processing unit, a bit allocation unit, and a quantizer. Next, the bit stream multiplexer 20 is configured to include a Huffman encoder and a multiplexer plexer, and a plurality of signals (for example, 14) are encoded by the tone component encoder 18 and the spectrum encoder 19. Of the Huffman table, and then the encoded tone component, the encoded non-tone component, and the data of the correction function output from the gain detection units 14-1 to 14-4 (details will be described later). like,
It includes position data in the subband frame, level data, and the number of change points. , And side information according to the ATRAC3 standard, which includes a), is multiplexed to obtain and output a bitstream signal of ATRAC3.

FIG. 3 is a block diagram showing the detailed structure of the audio decoder 6 shown in FIG. First, the bit stream demultiplexer 21 is read from a recording medium such as a mini disk 4 or a transmission medium and then demodulated, A
The bit stream signal of TRAC3 is a code string of tone components and a code string of spectra (code string of non-tone components)
And side information including the data of the correction function.
Next, the tone component decoder 22 and the spectrum decoder 23 respectively decode each code string into spectrum coefficients, and the tone component / spectrum synthesis unit 24 separates the decoded tone component signal and the non-tone spectrum. Combine with the signal. Next, the inverse MDCT processing units 25-1 to 25-4 generate subband frames in the time domain by inverse MDCT for each frequency band, and then the inverse gain control units 26-1 to 26-4 perform bitstream decoding. Based on the data of the correction function in the side information input from the demultiplexer 21, inverse gain control is performed on each subband frame by a correction function opposite to the gain control, that is, the portion amplified by the gain control is It suppresses, amplifies the suppressed part, and outputs the sub-band frame of each band subjected to inverse gain control to the QMF synthesis filter 27. Finally, the QMF synthesis filter 27 synthesizes the sub-band frames for each band (frequency band) and outputs the synthesized digital audio signal.

When audio coding is performed on a subband frame according to the subband frame reference, the beginning of the current subband frame (begi
nning) and multiple attack signals that occur at the beginning of future subband frames must be considered. Prior to the actual processing, the subband frame, which comprises 256 audio sample signals each, is divided into 32 partitions. To remove the attack signal at the beginning of the current subband frame n, the last eight partitions from the previous subband frame n-1 are examined and the start of the current subband frame n is started. Compared to the signal at. Referring to FIG. 5, future subband frame n +
In order to eliminate the attack signal that may occur in the first partition of 1, the modification function calculator introduces a time delay of one partition, so that the last partition of the current subband frame n Are always processed with the first 31 partitions of future subband frame n + 1. The delay changes the last section of the current subband frame n into a "pseudo-future" section, thus this delay assists gain control and is safer for attack signals occurring at subband frame boundaries. It becomes a device. That is, the time delay of one section by the correction function calculation unit is the output of the QMF division filter 10 for the subband frame to be MDCT processed at least every two subband frames in the MDCT processing units 16-1 to 16-4. Is delayed by one segment as compared with the subband frame.

In FIG. 5 showing the concept of shift of the sub-band frame and division of the "pseudo future", the portion indicated by the reference symbol 4a of the physical sub-band frame is MDC.
"Pseudo-future" for subband frame n to be processed
It is a category of. It is a physical subband frame n
Are temporarily stored in the buffer memories 14a-1 to 14a-4 until the physical subband frame n + 1 is obtained. When the physical subband frame n + 1 is acquired, the subband frame n + 1 to be MDCT processed is configured as shown in FIG. 6, and the 31st section of the physical subband frame n + 1 is temporarily buffered by the buffer memory 14a. -1 to 14a-4. The “attack signal” included in the subband frame n + 1 to be MDCT-processed, which is indicated by the reference symbol 4b, is in a detectable state with respect to the subband frame n to be MDCT-processed.

In order to detect the start portion of the attack signal in the subband frame based on the section, the absolute value of the peak of a plurality of sample signals included in the section (hereinafter referred to as peak value MaxPeak) is 8. The peak value MaxPeak of the preceding sections is compared. The peak value MaxPeak is 8
It is categorized as an attack signal only when it exceeds the peak value MaxPeak of the preceding sections by a certain ratio. All sample signals preceding the attack signal up to the position where the previous gain control was performed are subject to amplification. The degree of amplification is derived from the ratio.

When the necessary conditions are met, each attack signal is accompanied by attenuation, and the position where the attenuation starts is called "release point" or "release signal" in the present invention. Between the position of the previous attack signal and the segment that could potentially be identified as the release signal (ie the current segment), the absolute value of the peak in the segment between them is sought (below this peak Absolute value, peak value Inte
Call it rModMaxPeak. ). The peak value InterModMaxPeak is
Damping is performed when the peak value MaxPeak of the current segment is exceeded by a predetermined ratio. The damping coefficient is the peak value InterModMaxPeak and the peak value MaxPeak.
And the value of. Furthermore, the execution of the attenuation is conditional on the time of the occurrence of the previous attack signal. If the temporal separation between the previous attack signal and the current potential release point is greater than the set duration, then the decay of the release signal is not started.

All amplification and attenuation factors (ie gain factors) calculated at any instant are added to the previously determined amplification factors and attenuation factors for the same subband frame. .

To prevent overflow, the present invention makes a final adjustment to all amplification and attenuation factors. Thereby, when cascaded before the processing of the MDCT, the first amplification factor of the next subband frame does not overflow the previously modified subband frame and each subsequent amplification factor is It is guaranteed not to overflow the previous peak value InterModMaxPeak.

The embodiment according to the present invention is for processing an attack signal that crosses the boundary of a subband frame,
The buffer memory 14 stores a plurality of sections of the subband frame.
a-1 to 14a-4 for storing and delaying, detecting the beginning of the attack signal, determining an amplification factor applied to the signal preceding the attack signal, and applying attenuation The steps include: determining a position to be performed, calculating an attenuation coefficient for the attenuation, and preventing overflow in the current subband frame and the previous subband frame.

7 to 10 are flow charts showing the first part of the correction function calculation processing executed by the correction function calculation units 14b-1 to 14b-4 of FIG. This gain control processing is performed by the gain detection units 14b-1 to 14b-
4 is performed below for each subband frame n of each frequency band.

In step S1 of FIG. 7, first, from the buffer memories 14a-1 to 14a-4, three sub-band frames including the sub-band frame to be processed next and the sub-band frames before and after the sub-band frame are processed. The sample signal is read and used as the data of the subband frame used to calculate the correction function for the subband frame to be processed. The buffer memory 14
It is assumed that the audio sample signals to be encoded from the filter banks 13-1 to 13-4 are sequentially input in time series and stored in a-1 to 14a-4. Then, in step S2, each subband frame including 256 sample signals is divided into 32 sections (0th to 31st), and then in step S2A,
By shifting each subband frame so that it is delayed by one section, a "pseudo future" as shown in FIG.
Is supplied to the subband frame used for processing.

Next, in step S3, the peak value MaxPeak on the logarithmic scale for each section is calculated using the following equation. Here, max (x, y) is a function that selects the larger one of x and y.

[0048]

## EQU1 ## MaxPeak [i] = log ₂ (max (1, | i
Peak values of eight sample signals in the second section |)), 0 ≦ i ≦ 31

In step S4, the maximum peak value MaxPeak [i] in the subband frame n is set to the peak value UnModFramePeak of the subband frame before modification. In step S5, various loop variables are initialized. The number of change points adjust_num is the subband frame n
Indicates the number of change points (including amplification and attenuation) for which the correction should be performed, which starts at 0. That is,
The number of change points adjust_num is 3-bit field data indicating the number of change points of the gain for each band, and its maximum value is 7. Peak value I defined as the highest peak value MaxPeak between the current segment and the previous correction point
nterModMaxPeak [0] is the previous subband frame n−
The last peak value of 1 is initialized to InterModMaxPeak.
Num_mod_pt, which represents the variable of the change point of amplification or attenuation, is initialized to zero.

In step S6, the partition index j is initialized to a value indicating the first partition, ie 0, and the partition index j is incremented in the main loop of the flowchart so that the algorithm moves from partition to partition. To be done. In step S7, the latest peak value MaxPeak [j] being processed is set to the current peak value Inter.
Compared with ModMaxPeak [num_mod_pt], select the larger one to obtain the peak value InterModMaxPeak [num_mod
_pt] is updated.

The branching conditions of steps S8 and S10 indicate the reference of the attack signal. To ensure that the next division, the j + 1th division, is an attack signal, the peak value MaxP
eak [j + 1] is 8 (2 ^{8 on the} linear scale),
(1 + MaxPeak [j]) must be exceeded.
In steps S9 and S10, the maximum value MaxValue of the peak value MaxPeak of the j-1 to j-7th segment is calculated, and the peak value MaxPeak [j + 1] is larger than all of the eight preceding segments by one. Further impose conditions that require For j = 0 to j = 7, we need to compare using the last few partitions of the previous subband frame n-1. For j = 31, the (j + 1) th section is the buffer memory 14a for temporarily storing the above.
-1 to 14a-4. FIG. 6 illustrates the special case described above. If none of the conditions of the above-mentioned attack signal (steps S8 and S10) are satisfied, the algorithm of the gain control process proceeds to the release reference of step S13 in FIG.

When the attack signal criteria of steps S8 and S10 are met, the j + 1 th partition is determined to be the beginning of the attack signal and the various variables are updated using the equation of step S11 of FIG. . expo is defined as the power of correction (indicating the level of the correction function) by

[0053]

[Equation 2] expo ← (int) (MaxPeak [j + 1] −MaxValue)

Here, (int) is a function indicating the integer value of the argument. The modification level alevcode is defined as the power of modification applied to all partitions before the j + 1th partition and after the previous modification point. That is,
The modification level alevcode is 4-bit field data indicating a gain value for one gain change. Correction position al
The occode is the index (i.e. j) of the last partition where the modification is performed, i.e. the partition where the modification is performed up to that position. That is, the corrected position aloccode is
It is 5-bit field data indicating the position for one gain change. The index last_attack_pt tracks the index of the partition immediately preceding the attack signal (it does not track the release signal). Number of change points adjust_num
And the variable num_mod_pt at the change point are incremented by one. Next, the peak value InterModMaxPeak [num_m
od_pt] is updated (initialized) to the peak value MaxPeak [j + 1]
And the highest peak value Ma from the j + 1th section forward
Start tracking xPeak. In step S12, the value of the power expo of the above correction calculated for the attack signal is set to all the previously calculated correction levels al.
Add to evcode.

Steps S13 and S15 are the reference of the release signal. In order to start the attenuation, the latest peak value InterModMaxPeak [num_mod_pt] is (1 + MaxPeak
[J]) and its temporal separation from the point of the last attack signal, as tracked by the index last_attack_pt of the partition immediately after the attack signal, is greater than the length of 32 partitions. Must be small. If the conditions are met, in steps S14 and S15, the maximum value MaxValue of the peak values MaxPeak of the j + 1 to j + 7th sections is calculated, and the peak value InterMo is calculated.
dMaxPeak [num_mod_pt] is the next 7 peak values MaxPeak
Compare further with. Peak value InterModMaxPeak [num_mod
_pt] exceeds all of them, step S1
At 6, the updating of variables similar to step S11 follows. First, the exponentiation of the correction for the detected release signal is calculated using the following equation:

[0056]

[Equation 3] expo ← (int) (InterModMaxPeak [num_mod_p
t] −MaxValue)

Here, there are two significant differences with respect to step S11. First, a lower limit value (floor value) of -4 is imposed on the value of exponentiation of the correction. Also, once a release signal is found, its index is reset immediately after the attack signal so that its value is reset to prevent further attenuation until the next attack signal arrives.
32 is subtracted from st_attack_pt. This step
It is intended to prevent the detection of release signals, i.e. the detection of multiple release signals in succession after the attack signal, which would impair the accuracy of the signal being processed. Further, the correction level alevcode is assigned the exponentiation of the correction calculated above, and the correction position aloccode is assigned the index j of the current section. Peak value Inte
rModMaxPeak [num_mod_pt] is updated by substituting the peak value MaxPeak [j]. Similarly to step S12, in step S17 of FIG. 9, the value of the exponentiation of the correction calculated for the release signal is added to all the correction levels alevcode calculated previously.

In step S18, it is checked whether the current section j is the last section. Otherwise, the algorithm proceeds to step S19, before the loop returns to step S7 to process the next partition,
The partition index j is incremented by 1. If not, the algorithm proceeds to step S20, where the index last_att of the partition immediately after the attack signal.
Subtract 32 from ack_pt to invalidate.

In step S21, the signal amplitude is 3276.
7 (ie 2 ¹⁵ −1) above or −32
768 (i.e., ^{-2 15)} in order to prevent exceeding the below, any modification level alevcode [i]
This is a multi-step process for adjusting (0 ≦ i ≦ adjust_num). It limits the value of the modification level alevcode to between -4 and 11 (inclusive) in order to comply with the standard. It also limits the combined peak value InterModMaxPeak when it is combined with adjacent subband frames for MDCT processing, by the following equation, to prevent overflow.

[0060]

[Formula 4] alevcode [i] ← min (alevcode [i], 1
5-InterModMaxPeak [i])

Finally, the first modification level alevcode [0]
, Its value must not overflow a previously modified subband frame. This is conveniently done by keeping track of PrevFramePeak, the peak value of the previously modified subband frame, as follows:

[0062]

[Formula 5] alevcode [0] ← min (alevcode [0], 1
5-PrevFramePeak)

In step S22, Mod defined as the peak value MaxPeak of the modified subband frame
FramePeak is calculated. The above value is the peak value InterModMaxPeak generated through the process and the modification level alevcod.
It can be calculated from the pair with e as follows.

[0064]

[Equation 6] ModFramePeak ← max (InterModMaxPeak
[I] + alevcode [i]) (0 ≦ i ≦ adjust_num)

In step S23, an offset of +4 is added to all modification levels alevcode that are spread in the range of -4 to 11 according to the standard request.

In step S24, it is checked whether or not the last correction level alevcode is 4, that is, whether or not it is a change in gain 1. Last modification level alev
When the code is not 4, the process proceeds to step S26, and when the modification level alevcode is 4, the standard of ATRAC3 interpolates the modification function to 1 after the last change point (correction point). The change points (correction points) of are considered for redundancy and removed. Next, in step S26, the number of change points adjust_num
Value is checked to determine if gain control should be performed for the current subband frame. If gain control is not necessary, in step S27,
Precorrected subband frame peak value PrevFr
The amePeak and peak value InterModMaxPeak [0] are set to the peak value UnModFramePeak (see step S4) of the subband frame before modification, and the variable num_mo of the change point is set.
Set d_pt to 0. When gain control is required, the peak value PrevFramePeak of the previously modified subband frame is set to ModFramePeak which is the peak value MaxPeak of the modified subband frame in step S28.

When all partitions of subband frame n have been examined, the number of change points adjust_num, modification level alevco
Gain control is executed based on de and the corrected position aloccode. The above three data are represented as a modification function MFn in FIG. 4, and are also sent to the bitstream multiplexer 20 as side information.

Next, in step S29 of FIG. 10, it is determined whether or not the subband frame n to be processed is the last subband frame in the input audio signal. When the subband frame n is not the last subband frame of the input signal (NO in step S29), the process returns to step S1 to set the next subband frame n + 1 as the next subband frame to be processed. . On the other hand, subband frame n
Is the last subband frame of the input signal (YES in step S29), the correction function calculation process ends.

An audio signal encoding method and apparatus according to an embodiment of the present invention detects an attack signal and a release signal in a sample signal, and subsequently detects a pre-echo and a post-echo for an audio signal conversion encoder. It is effective in reducing the artifacts of. This relatively low computational complexity and memory requirement is due to the low cost of L
Provides excellent candidates for SI implementation.

Although MDCT processing is performed in the above embodiments, the present invention is not limited to this, and various orthogonal transformation processing may be performed.

In the above embodiment, the correction function is calculated based on three subband frames, but the present invention is not limited to this, and at least three subband frames or subbands are calculated. It may be calculated based on at least three non-divided frames.

In the above embodiment, the three subband frames are delayed by one segment before the correction function is calculated, but the present invention is not limited to this, and the time is shorter than the subband frame. It may be configured to delay by at least one segment.

[0073]

As described in detail above, according to the audio signal encoding method or apparatus of the present invention, a correction function is calculated based on an input audio signal, and the audio signal is calculated according to the calculated correction function. A method of encoding an audio signal, in which gain control is performed on a signal, orthogonal transform processing is performed on the gain-controlled audio signal in frame units, and encoding processing is performed to obtain an encoded bit stream signal, or In the device, the gain control is performed by dividing the input audio signal into time segments which are shorter than a frame time, delaying the audio signal by at least one segment time, and processing the delayed audio signal. A correction function is calculated based on at least three frames including the frame to be performed and the frames before and after it.

Therefore, since the input audio signal is delayed by the time of at least one section, the correction function is calculated based on at least three frames including the so-called "pseudo-future", so that the adjacent two are adjacent to each other.
Even if an attack signal is present at the boundary between two frames, the attack signal and the release signal can be accurately detected, and a more accurate correction function can be obtained. As a result, the correction function can be generated more accurately, the gain control can be executed, and the pre-echo noise and the post-echo noise can be surely suppressed.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a mini disk recording / reproducing system according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a detailed configuration of the audio encoder 2 of FIG.

3 is a block diagram showing a detailed configuration of the audio decoder 6 of FIG.

FIG. 4 is a block diagram showing generation of a conversion block in a gain detection unit 14-1 and a gain control unit 15-1 preceding the MDCT processing unit 15-1.

FIG. 5 is a diagram showing the concept of subband frame shift and “pseudo future” division.

FIG. 6 shows a plurality of partitions that contribute to determining attack signals for different values of partition number j.

FIG. 7 is a diagram showing correction function calculation units 14b-1 to 14b in FIG.
4 is a flowchart showing a first part of a correction function calculation process executed by -4.

FIG. 8 is a diagram showing correction function calculators 14b-1 to 14b in FIG.
4 is a flowchart showing a second part of the correction function calculation process executed by -4.

FIG. 9 is a diagram showing correction function calculators 14b-1 to 14b in FIG.
14 is a flowchart showing a third part of the correction function calculation process executed by -4.

FIG. 10 is a schematic diagram of correction function calculation units 14b-1 to 14b of FIG.
It is a flowchart which shows the 4th part of the correction function calculation process performed by b-4.

FIG. 11 (a) is a block diagram showing the configuration of a conventional audio signal encoding / decoding device when pre-echo and post-echo occur without gain control;
(B) is a block diagram showing a configuration of a conventional audio signal encoding / decoding apparatus for performing gain control and suppressing pre-echo and post-echo.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... A / D converter, 2 ... Audio encoder, 3 ... Mini disc recording device, 4 ... Mini disc, 5 ... Mini disc reproducing device, 6 ... Audio decoder, 7 ... D / A converter, 10 ... QMF division filter, 11 ... low-pass filter, 12 ... high-pass filter, 13-1 to 13-4 ... filter bank, 14-1 to 14-4 ... gain detection section, 14a-1 to 14a-4 ... buffer memory, 14b-1 to 14b-4 , 14b-1a, 14b-1
b, 14b-1c, 14b-1d ... Correction function calculation unit, 15-1 to 15-4 ... Gain control unit, 16-1 to 16-4 ... MDCT processing unit, 17 ... Tone component and spectrum detection unit, 18 ... Tone component encoder, 19 ... Spectral encoder, 20 ... Bitstream multiplexer, 21 ... Bitstream demultiplexer, 22 ... Tone component decoder, 23 ... Spectral decoder, 24 ... Tone component and spectrum synthesis unit, 25-1 to 25-4 ... Inverse MDCT processing unit, 26-1 to 26-4 ... Inverse gain control unit, 27 ... QMF synthesis filter, CO1, CO2 ... Cascade operator, MP1a, MP1b, MP1c, MP2a, MP2b ，
MP2c ... Multiplier.

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Suahone Neo             Singapore 534415 Singapore, Thailand             Sen Avenue, Block 1022, 04-             No. 3530, Thai Sen Industrial             Estate, Panasonic Singapore             Research Institute Co., Ltd. (72) Inventor Kim Han Kua             Singapore 534415 Singapore, Thailand             Sen Avenue, Block 1022, 04-             No. 3530, Thai Sen Industrial             Estate, Panasonic Singapore             Research Institute Co., Ltd. F-term (reference) 5D045 DA08 DA20                 5J064 AA01 BA16 BB07 BC01 BC08                       BC09 BC11 BC25 BD03

Claims (11)

[Claims] 1. A step of calculating a correction function based on an input audio signal, and gain-controlling the audio signal according to the calculated correction function; and a frame unit for the gain-controlled audio signal. In the method of encoding an audio signal, the method comprises the step of performing an orthogonal transform process and an encoding process to obtain an encoded bitstream signal. Based on at least three frames including a frame to be processed and a frame to be processed of the delayed audio signal, the audio signal being delayed by at least one segment time A method for encoding an audio signal, characterized by calculating a correction function. 2. The step of controlling the gain comprises the steps of: (a) obtaining an absolute value MaxPeak of a peak of each section;
(B) a step of identifying an attack signal, which is a part of a signal including a sudden sound rise, in the audio signal by using a predetermined attack signal reference; and (c) a frame to be processed and a frame before and after the frame. Calculating a correction function based on at least three frames including a frame and gain control based on the calculated correction function, thereby identifying the identified section of the attack signal and the section that is the previous modified position. A step of processing the attack signal to amplify a plurality of sections between and, the reference of the attack signal is the MaxPea for the next section.
2. The audio signal coding method according to claim 1, wherein the attack signal is identified based on k and MaxPeak, which is the highest value among a predetermined number of partitions preceding the next partition. 3. The gain controlling step comprises: based on the at least three frames, applying the correction function to the MaxPeak relating to a next partition from a partition to be currently processed,
3. The audio signal encoding method according to claim 2, wherein the calculation is performed based on the maximum value MaxPeak in a predetermined number of sections preceding the section to be currently processed. 4. (a) Absolute value In of the peak between the segment to be currently processed and the segment which is the correction position before that.
terModMaxPeak, (b) using a predetermined release signal criterion, identifying a release signal that is a portion of the signal in the audio signal that includes a sharp fall of sound, and (c) processing By calculating the correction function based on at least three frames including the frame to be performed and the frames before and after it, and performing gain control based on the calculated correction function, the identified release signal segment and Processing the release signal by amplifying a plurality of sections between a section that is a modification position behind and the reference of the release signal is the latest InterModMaxPea.
Identify the release signal based on k, the highest value MaxPeak of the following predetermined number of partitions, and the time between the partition currently to be processed and the partition at which the last attack signal was identified. The audio signal encoding method according to claim 2 or 3, wherein 5. The modification function for processing the release signal is calculated based on the InterModMaxPeak and the highest value MaxPeak of the subsequent predetermined number of partitions. 4. The audio signal encoding method as described in 4. 6. A correction function is calculated based on the input audio signal, and gain control means is provided for the audio signal according to the calculated correction function, and frame units are provided for the gain-controlled audio signal. In the audio signal coding apparatus, which comprises means for performing an orthogonal transform process and for obtaining a coded bitstream signal by performing a coding process, the gain control means controls the input audio signal by a time of a frame. The audio signal is delayed by at least one segment time, and is corrected based on at least three frames including the frame to be processed and the frames before and after the delayed audio signal. An audio signal encoding device characterized by calculating a function. 7. The gain control means uses (a) means for obtaining the absolute value MaxPeak of the peak of each section, and (b) a predetermined attack signal reference, whereby a sudden sound in the audio signal is generated. Means for identifying an attack signal, which is a part of the signal including the rising edge of, and (c) a correction function is calculated based on at least three frames including the frame to be processed and the frames before and after it, and the calculated correction function is calculated. Controlling the gain of the attack signal to amplify a plurality of partitions between the identified partition of the attack signal and the partition that is the previous correction position. , The standard of the attack signal is the MaxPea according to the following categories.
7. The audio signal coding apparatus according to claim 6, wherein the attack signal is identified based on k and MaxPeak, which is the highest value among a predetermined number of partitions preceding the next partition. 8. The gain control means comprises at least the at least three gain control means.
Based on one frame, the correction function is based on the MaxPeak relating to the next division from the division to be currently processed and the maximum value MaxPeak in a predetermined number of divisions preceding the division to be currently processed. 8. The audio signal encoding apparatus according to claim 7, wherein the encoding is performed. 9. (a) Absolute value In of a peak between a segment to be currently processed and a segment which is a correction position earlier than that.
means for obtaining terModMaxPeak, (b) means for identifying a release signal which is a portion of a signal including a sudden fall of sound in the audio signal by using a predetermined release signal reference, and (c) processing Calculating a correction function based on at least three frames including a frame to be performed and a frame before and after the frame, and performing gain control based on the calculated correction function, thereby identifying the identified release signal;
Means for amplifying a plurality of sections between the section which is a modification position after that and processing the release signal, and the reference of the release signal is the latest InterModMaxPea.
Identify the release signal based on k, the highest value MaxPeak of the following predetermined number of partitions, and the time between the partition currently to be processed and the partition at which the last attack signal was identified. 9. The audio signal encoding device according to claim 7, wherein the audio signal encoding device is an audio signal encoding device. 10. A modification function for processing the release signal is calculated based on the InterModMaxPeak and the highest value MaxPeak of the subsequent predetermined number of partitions. 9. The audio signal encoding device according to item 9. 11. Any one of claims 6 to 10
An audio signal encoding and decoding system comprising: an audio signal encoding device and an audio signal decoding device, wherein the audio signal decoding device is encoded by the encoding device. Means for obtaining an audio signal composed of a plurality of frames by decoding the coded bit stream signal and performing orthogonal transformation processing; and a correction function opposite to the correction function for the obtained audio signal. And a means for performing inverse gain control by using the same to obtain and output an inversely gain-controlled audio signal, the audio signal encoding and decoding system.