JP2002091498A - Audio signal encoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To conduct the section discrimination of MDCT coefficients prior to conducting quantization of the coefficients. SOLUTION: The device is composed of an auditory psychology analyzer 1 which computes auditory psychology parameters from audio signals, an MDCT equipment 2 which converts the audio signals to MDCT coefficients, a scale factor computer 3 which computes the scale factors of the scale factor band of the MDCT coefficients so that quantization noise caused by the MDCT coefficients does not become larger than the allowable quantization noise electric power of the auditory psychology parameters, a section discriminator 9 which divides each scale factor band to respective plural sections, a quantizer 4 which quantizes the MDCT coefficients by the scale factors and the entire quantization step numbers, a variable length encoder 6 which conducts a variable length encoding of the quantized values, a bit number discriminator 7 which discriminates to determine whether the encoded bit number is within a usable bit number or not and a bit stream generator 8 which summarizes the encoded data and generates a bit stream.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an MPEG2 AA
In an audio signal encoding device to which the C system is applied,
The present invention relates to an audio signal encoding device configured to perform a section determination of an MDCT coefficient before quantizing the MDCT coefficient.

[0002]

2. Description of the Related Art There are various methods for encoding an audio signal. An example of a method for converting an audio signal input on a time axis into an audio signal on a frequency axis and performing high-efficiency encoding is MPEG2 AAC (Moving Picture). Image
Coding Experts Group Phase 2 Advanced Audio Codin
g) The method is applied.

FIG. 4 is a block diagram for explaining a conventional audio signal encoding device, and FIG. 5 is a diagram for explaining an example of sectioning of MDTC coefficients.

[0004] The conventional audio signal encoding apparatus shown in FIG. 4 is one to which the above-mentioned MPEG2 AAC method is applied.

As shown in FIG. 4, in the conventional audio signal encoding apparatus, the audio signal on the time axis is transmitted to the psychoacoustic analyzer 1 and the MDCT (Modified Discrete Cosine Transformer).
form) device 2.

First, the psychoacoustic analyzer 1 performs an FFT process on an input audio signal on a time axis to obtain a frequency spectrum, calculates masking based on the frequency spectrum, and sets in advance based on human auditory characteristics. Calculated permissible quantization noise power for each frequency band as psychoacoustic parameters,
The permissible quantization noise power for each frequency band is output to the scale factor calculator 3 side.

On the other hand, the MDCT unit 2 converts the input audio signal on the time axis into MDCT coefficients (spectral data) on the frequency axis, and sends the MDCT coefficients to the scale factor calculator 3 and the quantizer 4. Output.
At this time, the conversion is performed by overlapping the operation block length by 50%, for example, by converting 2048 samples to 1024
It is converted to MDCT coefficients of a book.

[0008] Next, the scale factor calculator 3 calculates 1024 lines of M for each frequency band based on human auditory characteristics.
Dividing the DCT coefficients into multiple scale factor bands,
The scale factor of each scale factor band (the number of quantization steps) so that the quantization noise calculated for each scale factor band does not become larger than the permissible quantization noise power of the corresponding frequency band calculated by the psychoacoustic analyzer 1. ) Is calculated and output to the quantizer 4 and the variable-length encoder 6.

Next, the quantizer 4 quantizes the MDCT coefficients from the MDCT unit 2 in units of each scale factor band, and calculates the scale factor of each scale factor band calculated by the scale factor calculator 3. The MDCT coefficients in each scale factor band are quantized from the total number of quantization steps, and the quantized MDCT coefficients are output to the section generator 5 side. At this time, the quantization is executed by controlling the total number of quantization steps so that the number of bits required for quantization falls within the number of usable bits.

Next, the section generator 5 divides the quantized value of the MDCT coefficient output from the quantizer 4 into a plurality of block units (hereinafter, referred to as sections) as shown in FIG. I do. When performing encoding, a plurality of Huffman codebooks are used, and each Huffman codebook changes the Huffman codebook to be used according to the maximum value of the quantization value in each scale factor band. Inside the section, the maximum value of the included scale factor band further determines the codebook to be used. Sectioning is a technique for improving coding efficiency. A continuous scale factor band is defined as one section, and the section is encoded with one Huffman codebook, thereby reducing the amount of code in the Huffman codebook. The coding rate is improved especially in the case of a low coding rate.

That is, in the example of the sectioning of the MDTC coefficient shown in FIG. 5, sfbN represents the N-th scale factor band, section0 consisting of sfb0 to sfb2 uses codebook0, and sfbN is used.
Section1 consisting of 3 to sfb6 is codebo
Using ok1, secti composed of sfb7 to sfb8
On2 uses codebook2, and similarly, s
The sectionM composed of fbM-1 to sfbM + 1 uses codebookM.

For example, when encoding a stereo signal having a sampling frequency of 48 KHz at an encoding rate of 64 Kbps, the average number of allocated bits per frame of one channel is 682 bits, and a different Huffman codebook is used for each scale factor band. When the encoding is performed, the number of bits required to represent the Huffman code book becomes 441 bits, and the allocation of the code amount of the quantized value is reduced, which leads to deterioration in sound quality. Therefore, by sectioning, the code amount of the Huffman codebook is reduced, and the allocation of the code amount to the quantization value is increased to improve the sound quality. Further, a plurality of section patterns are prepared and sent to the variable length encoder 6.

Next, in the variable length encoder 6, the quantizer 4
For each section determined by the section generator 5, the quantized value of the MDCT coefficient output from the section is subjected to variable-length encoding within the section using the same Huffman codebook corresponding to this section. Further, the scale factor is also subjected to variable length coding to reduce the redundancy. Then, a pattern of the section having the smallest generated code amount is selected from the patterns of the plurality of sections, and this is output to the bit number determination unit 7 side.

Next, the bit number determiner 7 determines whether or not the number of bits subjected to variable length encoding by the variable length encoder 6 is within the usable number of bits. If the number of bits exceeds the number of usable bits, the process returns to the quantizer 4 to perform quantization again, and is repeated until the number of generated bits falls below the number of usable bits. The variable-length encoded data output satisfying the number of usable bits is multiplexed with encoding parameters such as a sampling frequency and an encoding rate by a bit stream generator 8 and transmitted as a bit stream. I have.

[0015]

By the way, in the conventional audio signal encoding apparatus described above, the MPEG2 AA
Although the C method is applied, the best method of sectioning the MDCT coefficients has not been established.
Each time the coefficients are quantized, the MDCT coefficients must be sectioned into quantized values and selected from among several types of section division so as to have the minimum number of bits. Problem has arisen.

[0016]

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and a first invention is an audio signal encoding apparatus in which an audio signal is inputted and encoded. Psychoacoustic analysis means for calculating psychoacoustic parameters based on human auditory characteristics from an audio signal;
MDCT conversion means for converting into T coefficients (spectral data), and a quantum calculated based on the MDCT coefficients of each scale factor band by dividing the MDCT coefficients obtained by the MDCT conversion means into a plurality of scale factor bands. Scale factor calculating means for calculating a scale factor of each scale factor band so that quantization noise is not larger than the permissible quantization noise power of the psychoacoustic parameter calculated by the psychoacoustic analysis means; Section determining means for determining a maximum value of the MDCT coefficient, normalizing the value based on the scale factor calculated by the scale factor calculating means, and dividing the aggregate of the scale factor bands into a plurality of sections. And the scale factor calculating means Quantizing means for quantizing the MDCT coefficients in units of scale factor bands using the scale factor of each scale factor band calculated in the above and the total number of quantization steps, and a quantized value from the quantizing means, A variable-length encoding unit that performs variable-length encoding using a Huffman codebook corresponding to the section for each section determined by the section determination unit; Bit number determining means for determining whether the number is within the range, and an output of the scale factor calculating means and an output of the bit number determining means, and a bit stream generating means for generating a bit stream by summing the encoded data. An audio signal encoding device characterized by comprising:

According to a second aspect of the present invention, in the audio signal encoding apparatus according to the first aspect, the section determining means includes a step of setting the MD for each of the scale factor bands.
The maximum value of the CT coefficient is obtained and normalized, the section division is determined according to the coding rate, and further, the maximum value of the normalized value of the section is obtained and used from the normalized value of the scale factor band for each section. An audio signal encoding device characterized by determining a Huffman codebook.

According to a third aspect of the present invention, in the audio signal encoding apparatus according to the first aspect, the section determining means includes a step of setting the MD for each of the scale factor bands.
Determine the maximum value of the CT coefficient, normalize, select the higher order of the normalized value of the scale factor band by the number according to the coding rate, and select the scale factor band included in each section based on the selected normalized value. Is determined, a section division is determined, and a Huffman codebook to be used is determined for each section by determining the maximum value of the normalized value of the section from the normalized value of the scale factor band. It is an audio signal encoding device.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of an audio signal encoding apparatus according to the present invention will be described below in detail with reference to FIGS.

FIG. 1 is a block diagram illustrating an audio signal encoding apparatus according to the present invention, and FIG. 2 illustrates an example of section determination of MDCT coefficients by a section determiner in the audio signal encoding apparatus according to the present invention. FIG. 3 is a diagram for explaining a modification of the section determination of the MDCT coefficient by the section determiner in the audio signal encoding apparatus according to the present invention.

For convenience of explanation, the same reference numerals are given to the same constituent members as those shown in the conventional example, and description will be appropriately made, and new reference numerals will be given to constituent members different from the conventional example. A description is given below.

The audio signal encoding apparatus according to the present invention shown in FIG. 1 employs the MPEG2 AAC system as in the conventional example.

In the conventional example described above with reference to FIG.
While the sectioning of DCT coefficients (spectral data) is performed after quantization, the audio signal encoding apparatus according to the present invention employs MDC before quantization of MDCT coefficients.
By performing the section determination of the T coefficient, the processing time of the audio signal encoding can be reduced. That is, in the conventional example, the section generator 5 is provided after the quantizer 4 as shown in FIG. 4, but in the present invention, as shown in FIG.
, A section determiner 9 is newly provided before the quantizer 4.

As shown in FIG. 1, in the audio signal encoding apparatus according to the present invention, the audio signal on the time axis is inputted to the psychoacoustic analyzer 1 and the MDCT unit 2 and inputted by the psychoacoustic analyzer 1. On the basis of the audio signal on the time axis, the permissible quantization noise power for each frequency band set in advance based on the human auditory characteristics is calculated as a psychoacoustic parameter and output to the scale factor calculator 3 side.
The audio signal on the time axis input by the T unit 2 is converted into MDCT coefficients (spectral data) on the frequency axis, and the MDCT coefficients are output to the scale factor calculator 3 and the section determiner 9 side.

Next, in the scale factor calculator 3, 1024 M lines are set for each frequency band based on the human auditory characteristics.
Dividing the DCT coefficients into multiple scale factor bands,
The scale factor of each scale factor band (the number of quantization steps) so that the quantization noise calculated for each scale factor band does not become larger than the permissible quantization noise power of the corresponding frequency band calculated by the psychoacoustic analyzer 1. ) Is calculated and output to the section determiner 9, the quantizer 4 and the bit stream generator 8 side.

Next, the session decision unit 9 provided between the MDCT unit 2 and the scale factor calculator 3 and the quantizer 4 is a main part of the present invention.

The session determiner 9 calculates the maximum value of the MDCT coefficient for each scale factor band.
This value is normalized by the scale factor sent from the scale factor calculator 3 as in equation (1).

Y (sfb) = | x | ^3/4 × 2 ^{3/16 × sf (sfb)} Equation (1) where sfb represents a scale factor band number, and sf (sfb) is the sfb-th number. Represents the scale factor of the scale factor band, and x is the MDCT which is the maximum absolute value inside the sfb-th scale factor band.
It represents a coefficient.

Further, the session determiner 9 divides the aggregate of scale factor bands into a plurality of sessions based on a table in which the pattern of the session division is determined according to the coding rate. Here, as the simplest session pattern, a pattern in which the number of scale factor bands of each section is equally divided as shown in FIG. 2 will be described as an example.

The number of scale factor bands in the session is increased or decreased according to the coding rate. The lower the coding rate, the larger the number of scale factor bands in one session. To reduce the amount of code used to represent the Huffman codebook. The maximum value of the normalized value of each session is obtained from the normalized value of the scale factor band calculated from the above equation (1), and is output to the quantizer 4 along with the pattern of the session.

In the above description, the number of scale factor bands in the session is set to be equal. However, the present invention is not limited to this. , A pattern table weighted such that the number of scale factor bands increases as the section becomes larger.

Next, the quantizer 4 performs quantization in units of scale factor bands. Scale factor calculator 3
The MDCT coefficient in each scale factor band is quantized from the scale factor of each scale factor band calculated in step (1) and the total number of quantization steps. Further, the maximum value of the normalized value of each session output from the session determination unit 9 is also quantized as in the following equation (2).
It is output to the variable length encoder 6 side.

Z ′ (sec) = z (sec) × 2 ^{−3 / 16 × common scale} ... Equation (2) Here, sec is the section number, common scale is the total number of quantization steps, z (sec) is the maximum normalized value of the sec-th section, and z '(sec) is the sec-th. Are the quantized values of the section.

In the variable length encoder 6, the quantized value of the MDCT coefficient output from the quantizer 4 is used by the section decision unit 9.
For each section determined in the above, a Huffman codebook is determined from the maximum quantization value in the section calculated by the above equation (2), and the same Huffman codebook corresponding to this section is used in the section. Variable length coding is applied. Further, the scale factor is also subjected to variable length coding to reduce the redundancy. Then, a pattern of the section having the smallest generated code amount is selected from the patterns of the plurality of sections, and this is output to the bit number determination unit 7 side.

Next, the bit number determiner 7 determines whether or not the number of bits subjected to variable length encoding by the variable length encoder 6 falls within the usable number of bits. If the number of bits exceeds the number of usable bits, the process returns to the quantizer 4 to perform quantization again, and is repeated until the number of generated bits falls below the number of usable bits. The output from the scale factor calculator 3 and the output from the bit number determination 7 are supplied to the variable-length coded data output satisfying the number of usable bits in the bit stream generator 8, It is multiplexed with encoding parameters such as a sampling frequency and an encoding rate and transmitted as a bit stream.

In the above-described embodiment, when the session determination unit 9 performs the session determination of the MDCT coefficients, the scale factor band is determined based on the table in which the pattern of the session division is determined according to the coding rate. Is divided into a plurality of sessions. Instead of performing sectioning using a fixed section table in advance in this way, as shown in a modification shown in FIG.
The sectioning to the DCT coefficient may change fluidly. In this case, first, from the normalized values calculated in each scale factor band by the section determiner 9, several higher-order values are selected in accordance with the coding rate.
As the coding rate is lower, the number of normalization values to be selected is reduced, the number of sections is reduced, and the code amount used to represent the Huffman codebook is reduced.

That is, as shown in FIG. 3, a modification of the section judgment of the MDCT coefficient by the section judgment unit 9 (FIG. 1) will be described. The vertical axis represents the normalized value, and the horizontal axis represents the scale factor band. Let the columns be represented (the horizontal axis has a maximum of N scale factor bands). In FIG. 3, it is assumed that, for example, the top six normalized values are selected.

In FIG. 3, each normalized value (y in FIG. 3)
0, y1,... Y5), a straight line having a slope | a | is drawn left and right from each point (thin dotted line in FIG. 3), and a point intersecting the straight line of the slope | a | (P in FIG. 3
(0, P1, P2, P3), a vertical line is drawn from the intersection, and the scale factor band intersecting the horizontal axis is the boundary of the section. As shown in the area B of FIG. 3, a portion that does not intersect with the straight line of the gradient | a | of each normalized value is defined as an intermediate value of the intersection between the straight line of each normalized value | a | (Q0 in FIG. 3). The region A or the region C at the end in FIG. 3 is included in the section closest to them. The high-order value of the section pattern and the normalized value thus obtained is output to the quantizer 4.
Similarly, quantization and variable length encoding are performed by the variable length encoder 6.

Here, the inclination |
The section was determined using the straight line of a |
Is an arbitrary numerical value and may be varied according to the characteristics of the input audio signal. In addition, a logarithmic function or an exponential function may be used instead of the straight line.

[0040]

According to the audio signal encoding apparatus according to the present invention described in detail above, the processing is performed particularly by determining the sectioning of the quantization of the MDCT coefficients before quantizing the MDCT coefficients (spectral data). Time can be reduced. Further, when the coding rate is low, a large amount of code can be allocated to the audio signal by reducing the code amount of the Huffman codebook, and improvement in sound quality can be expected.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an audio signal encoding device according to the present invention.

FIG. 2 is a diagram for explaining an example of section determination of MDCT coefficients by a section determiner in the audio signal encoding device according to the present invention.

FIG. 3 is a diagram for explaining a modification of section determination of MDCT coefficients by a section determiner in the audio signal encoding device according to the present invention.

FIG. 4 is a block diagram for explaining a conventional audio signal encoding device.

FIG. 5 is a diagram for explaining an example of sectioning of MDTC coefficients.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Psychological analyzer, 2 ... MDCT device, 3 ... Scale factor calculator, 4 ... Quantizer, 6 ... Variable length encoder, 7
... bit number judging device, 8 ... bit stream generator, 9 ...
Section determiner.

Claims (3)

[Claims] An audio signal encoding apparatus for receiving and encoding an audio signal, comprising: a psychoacoustic analysis means for calculating a psychoacoustic parameter based on a human auditory characteristic from the input audio signal; MDCT conversion means for converting the audio signal into MDCT coefficients (spectral data); and dividing the MDCT coefficients obtained by the MDCT conversion means into a plurality of scale factor bands, based on the MDCT coefficients of each scale factor band. Scale factor calculation means for calculating the scale factor of each scale factor band so that the calculated quantization noise is not larger than the permissible quantization noise power of the psychoacoustic parameter calculated by the psychoacoustic analysis means, The MD for each scale factor band Section determining means for determining the maximum value of the T coefficient, normalizing the value based on the scale factor calculated by the scale factor calculating means, and dividing the aggregate of the scale factor bands into a plurality of sections; Quantizing means for quantizing the MDCT coefficients in units of scale factor bands based on the scale factor of each scale factor band calculated by the scale factor calculating means and the total number of quantization steps; Variable-length encoding means for performing variable-length encoding using a Huffman codebook corresponding to a section for each section determined by the section determining means, and the number of bits encoded by the variable-length encoding means. Bit number determining means for determining whether or not is within the number of usable bits; An audio signal encoding apparatus, comprising: a bit stream generation unit that receives an output of the scale factor calculation unit and an output of the bit number determination unit and collectively generates encoded data to generate a bit stream. 2. The audio signal encoding apparatus according to claim 1, wherein the section determination unit determines a maximum value of the MDCT coefficient for each of the scale factor bands, normalizes the maximum value, and determines a section division according to a coding rate. And then
An audio signal encoding apparatus, wherein a maximum value of a normalized value of a section is obtained from a normalized value of the scale factor band for each section to determine a Huffman codebook to be used. 3. The audio signal encoding device according to claim 1, wherein the section determination means obtains a maximum value of the MDCT coefficient for each of the scale factor bands, normalizes the maximum value, and calculates a normalized value of the scale factor band. Higher ranks are selected by the number according to the coding rate, the number of scale factor bands included in each section is determined based on the selected normalized value, section division determination is performed, and the scale factor is further determined for each section. An audio signal encoding apparatus, wherein a maximum value of a section normalization value is obtained from a band normalization value to determine a Huffman codebook to be used.