WO2011087333A2

WO2011087333A2 - Method and apparatus for processing an audio signal

Info

Publication number: WO2011087333A2
Application number: PCT/KR2011/000327
Authority: WO
Inventors: 정규혁; 김대환; 강인규; 김락용; 홍기봉; 박지강; 이인성; 임종하; 문상현; 이병석; 전혜정
Original assignee: 엘지전자 주식회사; 충북대학교 산학협력단
Priority date: 2010-01-15
Filing date: 2011-01-17
Publication date: 2011-07-21
Also published as: WO2011087333A3

Abstract

The present invention relates to a method and apparatus for processing an audio signal, wherein the method comprises the following steps: performing a linear predictive analysis on the audio signal of the current frame to generate a target vector corresponding to a plurality of linear predictive transform coefficients; generating, from the target vector, a plurality of sub-vectors including a first sub-vector and a second sub-vector; vector-quantizing the first sub-vector to obtain a first code vector; removing one or more redundant code vectors from an initial code book for the second sub-vector, using the last component of the first code vector; adding one or more code vectors to the initial code book using the last component of the first code vector to update the code book; and vector-quantizing the second sub-vector using the updated code book to obtain a second code vector, wherein said added code vectors are predicted on the basis of the last component of the first code vector.

Description

[DESCRIPTION]

[Invention Title]

Audio signal processing method and device [Technical Field]

The present invention relates to an audio signal processing method and apparatus capable of encoding or decoding an audio signal.

[Background Art]

In general, linear predictive coding (LPC) is performed on an audio signal, particularly when the audio signal has a strong characteristic. The linear-predictive coefficients generated by the linear predictive coding are sent to a decoder, which reconstructs the audio signal through linear predictive synthesis on the coefficients.

[Disclosure]

Technical Problem

Vector-quantization is performed to transmit the linear-prediction coefficient or the linear-prediction transform coefficient to the decoder, and since there is a quantization error, sound quality is distorted.

[Technical Solution]

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and provides an audio signal processing method and apparatus for adaptively configuring a codebook for each subvector in vector-quantizing linear-prediction transform coefficients. Another object of the present invention is to use an order property of the linear-predictive transform coefficients to remove the redundancy of the codebook and to add a code vector similar to the target vector to the codebook. A method and apparatus are provided.

It is still another object of the present invention to provide an audio signal processing method and apparatus for generating a codebook based on the ordering nature of linear-predictive transform coefficients, using the previously quantized coefficients. have.

[Advantageous Effects]

The present invention provides the following effects and advantages.

First, a code vector similar to a target vector (target subvector) is added to the codebook, but since unnecessary code vectors are excluded from the codebook, the total codebook size remains the same or almost similar, without increasing the number of bits. Quantization error can be minimized.

Second, since the minimum and maximum values are estimated using the ordering property, the quantization error can be minimized without requiring extra bits.

Description of Drawings

1 is a block diagram of an encoder in an audio signal processing apparatus according to a first embodiment of the present invention.

2 is a diagram for explaining the concept of a linear-prediction transform coefficient, a target vector, and a sub vector. 3 is a diagram illustrating a concept of a codebook for each subvector according to the present invention;

4 is a diagram for explaining a codebook update concept according to the present invention;

5 is a block diagram of a decoder in an audio signal processing apparatus according to a first embodiment of the present invention.

6 is a block diagram of an encoder in an audio signal processing apparatus according to a second embodiment of the present invention.

7 illustrates the concept of an arrangement, reference subvector and non-reference subvector according to the present invention;

8 is a first example of an array, reference subvector, non-reference subvector.

9 shows a second example of an array, reference subvector, and non-reference subvector.

10 shows a third example of an array, reference subvector, non-reference subvector.

11 shows a fourth example of an array, reference subvector, non-reference subvector.

12 shows a fifth example of an array, reference subvector, non-reference subvector.

13 is a block diagram of a decoder in an audio signal processing apparatus according to a second embodiment of the present invention.

14 is an example of a normalized codebook, and one or more L of a red-eye codebook.

15 is a schematic structural diagram of a product implemented with an audio signal processing apparatus according to an embodiment of the present invention;

16 is a relationship diagram of products in which an audio signal processing apparatus according to an embodiment of the present invention is implemented. [Best Mode]

In order to achieve the above object, an audio signal processing method according to the present invention comprises: generating a target vector for a plurality of linear-predictive transform coefficients by performing a linear-prediction analysis on an audio signal of a current frame. ; Generating the target vector with a plurality of subvectors including a first subvector and a second subvector; Obtaining a first code vector by vector quantizing the first subvector; Using the last component of the first code vector, removing one or more redundant code vectors from an initial codebook for the second sub vector; Generating an update codebook by adding one or more additional code vectors to the initial codebook using the last component of the first code vector; And, using the update codebook, vector quantizing the second subvector to obtain a second code vector, wherein the additional code vector is predicted based on a last component of the first code vector. .

According to the present invention, the number of redundant code vectors may be the same as the number of additional code vectors.

According to the present invention, the generating of the update codebook includes: obtaining an initial component prediction value by using a last component of the first code vector; Selecting two or more candidate code vectors from the initial codebook using the original component prediction value; Temporary code by interpolating the candidate code vectors Creating a vector; Selecting code vectors that satisfy the temporary code vector increment ordering property as the additional code vector; And generating the update codebook by adding the additional code vector to the initial codebook.

According to the present invention, the initial component prediction value is obtained by adding a gradient mean to the last component of the first code vector, and the increment average is a difference between the current component and the previous component of the first code vector. It may be based on.

According to the present invention, the two or more candidate code vectors may be selected based on a difference between an initial component of code vectors included in a codebook for the second sub vector and the first component prediction value.

According to the present invention, in the ascending sorting, the update codebook may be assigned a first index to a code vector located after the redundant code vector in the initial codebook.

According to another aspect of the present invention, by performing a linear-prediction analysis on the audio signal of the current frame, a linear predictor for generating a target vector for a plurality of linear-predictive transform coefficients; A vector divider configured to generate the target vector from a plurality of subvectors including a first subvector and a second subvector; A first code vector obtained by vector quantizing the first sub vector Quantization unit; Using the last component of the first code vector, one or more redundant code vectors are removed from the initial codebook for the second subvector, and the last component of the first code vector is used to remove one or more redundant code vectors from the initial codebook. A codebook updater for generating an update codebook by adding an additional code vector; And a first quantizer for obtaining a second code vector by vector quantizing the second subvector using the update codebook, wherein the additional code vector is predicted based on the last component of the first code vector. An audio signal processing apparatus is provided.

According to another aspect of the invention, there is provided a method comprising _: receiving a first codebook index of a first code vector _: and a second codebook index of a second code vector; Obtaining a last component of the first code vector using a first codebook index; Using the last component of the first code vector, removing one or more redundant code vectors from an initial codebook for a second subvector; Generating an update codebook by adding one or more additional code vectors to the initial codebook using the last component of the first code vector; Obtaining a second subvector using the update codebook and the second codebook index, wherein the additional code vector is predicted using the last component of the first code vector. Is provided. According to another aspect of the present invention, there is provided a method of generating a plurality of linear-predictive transform coefficients by performing a linear-prediction analysis on an audio signal of a current frame; Generating a reference sub-vectors the ^'plurality of linearly-arranged by the predictive transform coefficients, the one or more reference sub-vectors, and at least one ratio; Obtaining a reference code vector by performing vector quantization on the one or more reference subvectors; Estimating the minimum and maximum values of the non-referenced subvectors using two components of each component of the reference code vectors; Generating a red-eye codebook for the non-referenced subvector using the minimum and the maximum values; And acquiring a non-reference code vector by performing vector quantization on the non-reference subvector using the decoded codebook.

According to the present invention, the step of generating the red-eye codebook; Obtaining normalized code vectors; And generating the red-eye codebook by applying the minimum value and the maximum value to the normalized codevectors.

According to the present invention, the reference code vector includes a first reference code vector, and two components of the reference code vectors may be the last component and the last previous component of the first reference code vector. According to the present invention, the reference code vector includes a first reference code vector and a second reference code vector, wherein two components of the reference code vectors are the last component of the first reference code vector and the second reference. It may be the first component of the code vector.

According to the present invention, the minimum value and the maximum value may be estimated based on the ordering property of the linear-prediction transform coefficients.

According to the invention, the arrangement of the plurality of linear-prediction transform coefficients is [Wl, W2, W3, W8, W13, W14, W15, W16, W4, W5, W6, W7, W9, WIO, Wl, W12], [Wl, W2, W3, W8, W4, W5, W6, W7], [Wl, W2, W6, W3, W4, W5, W7, W8, W9], [WL W2, W6, W3, W4 , W5, WIO, Wl, W12, W7, W8, W9], and [Wl, W2, W6, W3, W4, W5, W7, W8, W12, W9, W10, Wl, W13, W14, W15] (W _X is one of the X-th linear-predictive transform coefficients), and the reference subvector includes {Wl W2 W3 W8}, {W13 W14 W15 W16}, {Wl W2 W6}, {WIO Wl W1}, and {W7 W8 W12} (where WX is the X-th linear-predictive transform coefficient), and the non-referenced subvectors are {W4 W5 W6 W7}, {W9 WIO Wl W12}, {W3 W4 W5} , {Wl W2 W6}, {WIO Wl W12}, {W7 W8 W12}, where W _x is an X-th linear-predictive transform coefficient. According to another aspect of the present invention, by performing a linear-prediction analysis on the audio signal of the current frame, a linear prediction analyzer for generating a plurality of linear-predictive transform coefficients; By arranging the plurality of linear-predictive transform coefficients, one An array for generating the at least one reference subvector and at least one non-reference subvector; An independent quantizer that obtains a reference code vector by performing vector quantization on the one or more reference subvectors; An adaptive codebook generation unit for estimating a minimum value and a maximum value of the non-reference subvector using two components of each of the reference code vectors, and generating the degenerate codebook using the minimum value and the maximum value; And a dependent quantizer for acquiring a non-reference code vector by performing vector quantization on the non-reference subvector using the adaptive codebook.

According to another aspect of the invention, there is provided a method comprising: receiving a reference codebook index of one or more reference code vectors, and a non-reference codebook index of one or more non-reference code vectors; Acquiring one or more reference code vectors using the one or more reference codebook indexes; Estimating a minimum and maximum value of a non-reference subvector using two components of each component of the reference code vectors; Generating a red-eye codebook for the non-referenced subvector using the minimum and the maximum values; And acquiring the non-reference subvector using the decod codebook and the non-reference codebook index.

[Mode for Invention] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Prior to this, terms or words used in the present specification and claims should not be construed as being limited to conventional or dictionary meanings. The inventor should be interpreted as meanings and concepts corresponding to the technical idea of the present invention based on the principle that the concept of terms can be properly defined in order to explain his invention in the best way. Therefore, the embodiments described in the specification and the drawings shown in the drawings are only the most preferred embodiment of the present invention and do not represent all of the technical idea of the present invention, these are replaced in the present application It should be understood that there may be various equivalents and variations that can be made.

In the present invention, the following terms may be interpreted based on the following criteria, and terms not described may be interpreted according to the following meanings. Coding can be interpreted as encoding or decoding in some cases, and information is a term that encompasses values, parameters, coefficients, elements, and so on. It may be interpreted otherwise, but the present invention is not limited thereto.

Here, the audio signal is broadly defined as a concept that is distinguished from a video signal, and refers to a signal that can be identified by hearing during reproduction. In narrow terms, an audio signal is a concept that is distinguished from a speech signal. Signal with little or no characteristics it means. The audio signal in the present invention should be interpreted broadly and can be understood as a narrow audio signal when used separately from a voice signal. Coding may also refer to encoding only, but may be used as a concept including both encoding and decoding.

1 is a block diagram of an audio signal processing device enhancement encoder according to a first embodiment of the present invention. 2 to 4 are diagrams for explaining the configuration of the encoder of FIG.

Referring to FIG. 1, the encoder 100 according to the first exemplary embodiment may include a vector divider 120, a first quantizer 130-1, a second quantizer 130-2, and a codebook updater 140. And a linear prediction analyzer 110, and an N-th quantizer 130 -N.

The linear prediction analyzer _{1 10} generates linear-prediction coefficients by performing linear prediction analysis on the input audio signal according to linear prediction coding (LPC). The basic idea of the model of linear predictive coding is that it can be approximated by a linear combination of p speech signals past at a given point in time _n .

[Equation 1]

S (n)-qiS (nl) + q ₂ S (n-2) +-+ q _p S (np)

Where is a linear-prediction coefficient, ^η is a frame index and ρ is a linear prediction order. Since the linear-predicted coefficients obtained in this way have a large dynamic range, they need to be quantized with fewer bits, and the linear-prediction coefficients are quantized. Because it is weak to errors, it needs to be converted into coefficients that are robust to quantization errors. Accordingly, the linear prediction analyzer 110 converts the linear-prediction coefficient into a linear-prediction transform coefficient Wi. Here, the linear-predictive transform coefficient may be one of Line Spectral Pairs (LSP), Implementance Spectral Pairs (ISP), or Line Spectrum Frequency (LSF) or Immunity Spectral Frequency (ISF), but the present invention is not limited thereto. Where ISF can be expressed as

Equation 2]

Where Φ is the linear-prediction coefficient, fi is the frequency range of [0,6400Hz] of the ISF, and f _s = 12800 is the sampling frequency.

FIG. 2 is a diagram for describing the concept of a linear-prediction transform coefficient, a target vector, and a sub vector. Referring to FIG. 2A, it can be seen that a total of 16 linear-prediction transform coefficients Wi, i: 1 to 16 are arranged in order. The linear-prediction transform coefficients may be a total of 16 (ie, order 16) as shown in FIG. 2, but the present invention is not limited thereto. When the linear-predictive transform coefficients are arranged in order, this may be referred to as a target vector of 16 dimensions. These linear-predictive transform coefficients have ordering properties, as shown in Fig. 2B and the following equation.

[Equation 3]

0 = <W ₂ <.. <W _p- i <W _p = π,

-1 <Κ ₀ <1

Wi is the i-th linear-predictive transform coefficient, ρ is the order, Κ _Ρ is the gain value of the ISF coefficient The vector divider 120 generates two or more subvectors by dividing the target vector. Herein, the subvector may include a first subvector and a second subvector, and may include up to an Nth subvector. As shown in FIG. 2 _: the first subvector is from W ₄ , the second subvector is from W ₅ to W ₈ , the third subvector is from W ₉ to W ₁₂ , and the fourth subvector is from W ₁₃ to W ₁₆ . Although great, the present invention is not limited thereto.

The first subvector generated by the vector divider 120 is transferred to the first quantizer 130-1, the second subvector is the second quantizer 130-2, and the Nth subvector is the Nth quantization. Delivered to portion 130-N.

The first quantization unit 130-1 obtains the first code vector by vector-quantizing the first sub vector. In this case, a codebook for the first sub vector may be used.

3 is a diagram for explaining the concept of a codebook for each subvector according to the present invention. Referring to FIG. 3, a codebook for each subvector (codebook of a first subvector and a codebook of a fourth subvector) exists for each of the first subvector to the fourth subvector, and the quantization of the first subvector is performed. An initial codebook is used, and an update codebook is used for quantization of the second subvector. The initial codebook means a codebook that has not been updated, and the update codebook means a codebook in which some code vectors are added except for some code vectors using an ordering property. For the codebook updating process, the codebook updating unit 140 Is performed by.

The first quantization unit 130-1 generates a first code vector quantized vector by an initial codebook for the first sub vector. Here is the initial codebook, training It may be a codebook made through the LBG algorithm using the statistical characteristics of the vector, but the present invention is not limited thereto. A first codevector is the first component, as shown in the lower part of FIG. 4 (W,) to the end component (W _4), there may be present, the last component of the first code vector (W ₄₎ is in the codebook updating unit 140 Delivered. On the other hand, Figure 4 is only one example, the last component of the first code vector may be not only the fourth component (w ₄ ), but also the eighth component ( ₈ ), may be the twelfth component (w ₁₂ ) Bar, the present invention is not limited thereto.

The codebook updater 140 removes one or more surplus code vectors from the codebook for the second subvector, using the last component W _{4 of} the first codevector. In this case, the codebook may be a codebook generated through the LBG algorithm like the codebook for the first subvector.

4 is a diagram for explaining a codebook update concept according to the present invention. Referring to FIG. 4, there is a codebook for a second subvector on the right side, which, as described above, has a fifth linear-predictive transform coefficient (W ₅ ) to an eighth linear-predictive transform coefficient ( W ₈ ) up to and including vectors. According to the ordering property described above, the fifth linear-predictive transform coefficient must be larger than the fourth linear-predictive transform coefficient (W ₄ ). Since this ordering property is the same as the quantized result, the first component (w ₅₎ of the codebook of the second subvector corresponding to the fifth linear-predictive transform coefficient is the last component (W ₄ ₎ of the codebook of the first subvector. It must be greater than). Code vectors that do not meet this ordering quality (ie, code whose first component is not greater than the last component (W ₄ ) of the codebook of the first subvector) Vectors may correspond to redundant code vectors. Such surplus code vectors are excluded from the codebook of the first subvector. The code vectors included in the codebook can be sorted in ascending order. If there are a total of M surplus code vectors and the codebook index of the initial codebook is 1, the surplus code vectors are consecutive code vectors from 1 = 1 to 1 = M. Can be. If redundant code vectors are excluded and additional code vectors are added, they can be rearranged in ascending order and given a new index (e.g. m), where m = l is assigned to the code vector corresponding to 1 = M + 1. May be added. Reordering and indexing will be described later after additional code vectors have been added.

The codebook updater 140 then adds one or more additional codevectors to the codebook for the first subvector, using the last component of the first codevector. The additional code vector here is a code vector whose quantization error is not greater than other code vectors. The process of acquiring an additional code vector is as follows.

An initial component prediction value is obtained using the last component of the first code vector.

[Equation 5]

W _i P ^red = W _i- i + Δ, (i = 5, 9, 13)

Here, Wi ^pred is the first component predicted value, _Λ is the last component of the first code vector, Δ, is a gradient mean where Wi ^pred and the liver have a Gaussian distribution. The increase average may be calculated by the following equation, but the present invention is not limited thereto.

[Equation 6]

Δ ,. = avg (W _i-2 -W _i-3, W _i -W _i-2 , Wi-W _{; -1} ) (i = 4, 8, 12) Δ, is the gradient mean, avg () Is the average value

The codebook updater 140 selects two or more candidate codevectors from a codebook for a second ^subvector using the first component prediction value Wi ^pred . That is, among the code vectors included in the codebook, the code vectors having the smallest difference between the initial component of the ^stone and the first component prediction value Wi ^pred may be the candidate code vectors.

The codebook updater 140 generates a temporary code vector by interpolating the candidate code vectors. As the interpolation method, various methods known in the art may be applied. For example, when there are two candidate vectors, interpolation may be performed by the method shown in the following equation.

[Equation 7]

CB _m ^EXT = (am * CB _k + b _m * CB _{k + M +} i) / 2

CB _m ^EXT is an additional code vector (k <m <k + M + l), CB _k and CB _{k + M + 1} are candidate code vectors,

m is the codebook index newly assigned to the update codebook,

M is the total number of additional code vectors,

a _m and b _m are weights As described above, if a codebook index (m = l) is assigned to a codevector following an excluded redundant codevector, and when generating M additional codevectors, the candidate codevectors are m = k and m = k + M An index of + l is added, and m = k + l to m = k + M may be given to newly added M additional code vectors CB _{k + 1} to CB _{k + M.}

If both the number of redundant code vectors excluded and the number of additional code vectors are equal to M, the size of the conventional initial codebook and the size of the update codebook may be kept equal to L. Therefore, no additional bit is required to transmit the codebook index. In addition, by adding additional code vectors, which are more similar to the target code vector to be subjected to vector quantization, to the codebook, there is an effect of minimizing the quantization error and improving the sound quality.

Referring back to FIG. 1, the codebook updater 140 adds an update codebook to which the additional code vectors are added to the second quantizer 130-2, except for the surplus code vectors from the initial codebook. to provide.

The second quantization unit 130-2 obtains the second code vector by vector quantizing the second sub vector using the update codebook for the second sub vector. The last component of this second code vector is transmitted to the codebook updating unit 140 as in the first quantization unit 130-1, and the codebook updating unit is based on the last component of the second code vector, and the Nth code vector is used. The codebook for the subvector is updated and transmitted to the N-th quantizer 130-N. Then, the N-th quantizer 130-N quantizes the N-th subvector using the update codebook provided by the codebook updater 140 to obtain an N-th code vector.

The first codebook index corresponding to the first code vector generated by the first quantization unit 130-1 to the Nth quantization unit 130 -N, and the N th codebook index corresponding to the N th code vector are multiplexers ( Multiplexed by 150) and included in the bitstream, which is transmitted to the decoder.

Or, instead of the first codebook index or the Nth codebook index, the candidate values of the first codebook index for the first code vector, and the candidate values of the Nth codebook index for the second code vector are assigned to the first stage quantization unit. As an output, it may be an input of a two-step quantization unit 160 (not shown). The second stage quantization unit 160 (not shown) may perform multi-stage vector quantization (MSVQ) based on the candidate values. That is, the vector of the first stage (stage) and the vector of the second stage are summed to select indices with small distortion, respectively.

At this time, the second quantization unit 160 (not shown) divides the sub vector again, for example, {WiWsWsW Ws}, {W ₆ W ₇ W ₈ W ₉ W ₁₀ }, {WHW JWUWMWBW ^},

And {W ^ OWHWUWUWMW W ^ may be performed to perform the second step quantization, but the present invention is not limited thereto. The prediction selector 170 (not shown) may further include a prediction selector 170 (not shown) to calculate a correlation between frames with respect to the linear-prediction transform coefficient. In the case of an audio signal, since it is generally stationary in a short section, the correlation between frames is high. In this case, an auto-regressive prediction technique may be used to use the inter-frame correlation, but the present invention is not limited thereto.

[Equation 8]

Where ^fi , and are respectively predicted linear-predictive transform coefficients, original linear-predictive transform coefficients of the previous frame, quantized linear-predictive transform coefficients of the previous frame, and "are prediction coefficients.

If the speech signal changes rapidly in the short term, the linear-predictive transform coefficients are outside the prediction range. Therefore, the prediction selector 150 designs a quantizer so that all the cases with and without the prediction range can be quantized. Both quantization techniques are used at the same time and quantizers with small MSE distortion are selected.

[Equation 9]

£ = ± [^-/

, = ₀ (5) where P represents 16 linear-predictive transform coefficients, and and represent original linear-predictive transform coefficients and quantized linear-predictive transform coefficients. That is, when the interframe correlation is high, the prediction selector ₁₇₀ outputs a quantized result except for the prediction by using the transform coefficient of the previous frame. When the interframe correlation is low, the prediction selector ₁₇₀ quantizes the original transform coefficient. Output the result.

5 is a block diagram illustrating a decoder of an audio signal processing apparatus according to a first embodiment of the present invention. Referring to FIG. 5, the decoder 200 includes a first inverse quantizer 210-1, a second inverse quantizer 210-2, and a codebook updater 220, and includes an Nth inverse quantizer ( 210 -N), the vector merger 230, and the linear prediction synthesizer 240 may be further included.

First, the first inverse quantization unit 210-1 receives a first codebook index based on a first code vector and searches an initial codebook to obtain a first code vector based on a first codebook index. For example, in the case where the first codebook index is 1 = 2 in FIG. 4, a code vector of 1 = 2 is selected as the first code vector from among a plurality of code vectors included in the initial codebook. The first code vector is passed to the vector merger 230, in particular the last component of the first code vector (eg,

4) is transmitted to the codebook updating unit 220.

The codebook updater 220 removes one or more surplus code vectors from the initial codebook for the second subvector using the last component of the first code vector. In addition, using the last component of the first code vector, one or more additional code vectors are added to the initial codebook to generate an update codebook for the second sub vector. The detailed operation of the codebook updating unit 220 is described above with reference to FIG. 1. Since it is almost similar to the codebook updater 140 described, a detailed description thereof will be omitted.

The codebook updating unit 220 generates an update codebook for the third sub vector using the last component of the second code vector, and updates the N sub vector using the last component of the N-1 code vector. Create a codebook.

The second inverse quantization unit 210-2 receives the second codebook index and determines, as a second code vector, a code vector for the second codebook index from among a plurality of code vectors included in the update codebook. The second code vector is passed to the vector merger 230, of which the last component of the second code vector is passed to the codebook updater 220.

The N-th inverse quantizer 210-N searches for an update codebook for the N-th subvector generated by the codebook updater 220, thereby extracting an N-th code vector for the N-th codebook index and concatenating the vector merger. Forward to 230.

The vector merging unit 230 merges the first code vector, the second code vector, and the N-th code vector in a predetermined arrangement to obtain a de-quantized linear-prediction transform coefficient and transfer it to the linear prediction synthesis unit 240. do.

The linear prediction synthesis unit 240 generates an audio signal by performing linear-prediction (LPC) synthesis, which is a reverse process of the linear-prediction analysis, on the linear-prediction transform coefficients.

6 is a view showing the configuration of an encoder in an audio signal processing apparatus according to a second embodiment of the present invention, and FIG. 7 is a view for explaining the concept of an arrangement, a reference subvector and a non-reference subvector according to the present invention. 8 to FIG. 12 shows first to fifth examples of an arrangement, a reference subvector, and a non-reference subvector. FIG. 13 is a diagram illustrating a decoder of an audio signal processing apparatus according to a second embodiment of the present invention.

First, referring to FIG. 6, the encoder 300 according to the second embodiment includes a first independent quantization unit 330-1, a first dependent quantization unit 340-1, and a red-eye codebook generation unit 350. The linear prediction analyzer 310, the array 320, the N-th independent quantizer 330 -N, and the M-th independent quantizer 340 -M may be further included.

Since the linear prediction analyzer 310 performs the same function as the linear prediction analyzer 110 described above with reference to FIG. 1, a detailed description thereof will be omitted.

The arranging unit 320 (re) arranges the plurality of linear-prediction transform coefficients received from the linear prediction analyzing unit 310 according to a predetermined arrangement rather than an ascending order. Referring to (A) of 7, line - of a prediction transform coefficients shown are arranged in ascending order of a total of up to 16 days time, the first factor (last factor from W0 (W ₁₆₎ On the other hand, FIG. 7 (B) Note that a given array is shown, rather than an array in right order, where each element of the reference subvector (or reference code vector) is non-referenced subvector (or non-referenced) based on the ordering nature. Since it is an array that can take the maximum value or the minimum value of the code vector), a total of five examples will be described below with reference to FIGS. 8 to 12, but the present invention is not limited to the specific arrangement. The arranging unit 320 stores the target vector according to the above-described arrangement into at least one reference subvector (from the first reference subvector to the Nth reference subvector), and at least one non-reference subvector (first non-reference sub). Vector to Mth reference subvectors). Here, the reference subvector refers to a subvector referred to in generating an adaptive codebook of another subvector (non-reference subvector), and a non-reference subvector is referred to to generate a decoded codebook of another subvector. Does not mean subvectors. Each array can have a different number and location of reference and non-reference subvectors.

First, referring to FIG. 8A, when the first example has 16 linear-predictive transform coefficients, an example of the array is shown. Referring to FIGS. 9A and 9B, the second example is a total When there are eight coefficients, look at an example of the arrangement, (A) and (B) of FIG. 10, the third example is an example of the arrangement, when there are a total of nine coefficients, (A) and ( An example of the arrangement is shown when B) is twelve and FIG. 12 (A) and (B) are fifteen.

One or more reference subvectors are passed to an independent quantizer 330 (not shown) (first independent to Nth independent quantizer 330 -N), and one or more non-reference subvectors are each dependent quantizer 340: (Not shown) (the first dependent quantizer 340-1 to the M-th dependent quantizer 340-M) The independent quantizer is quantized regardless of the quantized result of the other subvectors, and the dependent quantizer is the other sub Quantize using the quantized result of the vector (ie, reference code vectors). The first independent quantization unit 330-1 to the Nth independent quantization unit 330 -N quantize the first reference subvector to the Nth reference subvector using an initial codebook, thereby providing the first reference code vector to Nth. Create a reference code vector. Here, the initial codebook may be a codebook made through the LBG algorithm using statistical characteristics of the training vector, but the present invention is not limited thereto. Then, two or more components of the first reference code vector to the Nth reference code vector are transmitted to the red horn codebook generation unit 350 as a minimum value and a maximum value. Referring to FIG. 8A, first, it can be seen that the first non-reference subvector is configured from W ₄ to W ₇ . Thus, according to the ordering property described in Equation 3 above, the first non-reference subvector is larger than the last previous component (w ₃ ) of the first reference code vector, which is the result of the quantization of the first reference subvector, and the last component (w). Has a smaller value than ₈ ). That is, the last previous component (w ₃ ) becomes the minimum value of the first non-reference subvector, and the last component (W ₈ ) is the maximum value of the first non-reference subvector.

FIG. 8B shows the relationship between the minimum value and the maximum value of the second non-referenced subvector. Since the second non-reference subvector includes W ₉ through W ₁₂ , according to the ordering nature, it is larger than the last component (W ₈ ) of the first reference code vector, and the first component (W ₁₃ ) of the second reference code vector (W ₁₃ ). Since they must be smaller, each can be determined as the minimum and maximum values. In the first non-reference subvector of the second example shown in FIG. 9, as in the case of the first non-reference subvector of the first example, the minimum value is the last previous component (W ₃ ) of the first reference code vector and the maximum value. Is the last component (W ₈ ).

Also in the case of the third example shown in FIG. 10, as in the case of the first non-reference subvector of the first example and the second example, the minimum value is the last previous component (w ₂ ) of the first reference code vector, and the maximum value is The last component (w ₆ ).

Also in the case of the fourth example shown in FIG. 11, referring to FIG. 11C, the first non-reference subvectors W ₃ to W ₅ are the first ratios of the first, second and third examples. As with the reference subvector, the minimum value is the last previous component (W ₂ ) of the first reference code vector and the maximum value is the last component (w ₆ ) of the first reference code vector.

Referring to FIG. 11D, the second non-reference subvectors W ₇ to W ₉ are, like the second non-reference subvectors in the first example, the last component W of the first reference code vector. ₆ ) is the minimum value, and the first component W ₁₀ of the second reference code vector is the maximum value.

In the case of the fifth example shown in FIG. 12, the first non-reference subvectors W ₃ to W ₅ are the same as those of the first example, the second example, the third example, and the fourth non-reference subvector. Similarly, the minimum value is the last previous component (W ₂ ) of the first reference code vector, and the maximum value is the last component (W ₆ ) of the first reference code vector.

The second non-reference subvector W ₉ -W _{U of} the fifth example is the last previous component W ₈ of the System 2 reference code vector, and the maximum value is the last component Wl 2 of the second reference code vector. Referring back to FIG. 6, the independent quantization unit 330 thus determines two or more components of one or more reference code vectors that are the result of performing quantization on one or more reference subvectors. The minimum value and the maximum value are transmitted to the red horn codebook generation unit 350. The method of determining the minimum value and the maximum value for each non-referenced subvector is as described with reference to FIGS. 8 to 12. For example, in the case of the first example shown in FIG. 8, the last previous component and the last component of the first reference codevector are passed as minimum and maximum values of the first non-reference subvector, respectively, The last component and the first component of the second reference codevector are passed as the minimum and maximum values of the second non-reference subvector.

The adaptive codebook generation unit 350 generates a red-eye codebook for the corresponding non-reference subvector using the minimum and maximum values for each non-reference subvector. In the step of generating a homogen codebook, first, an initial codebook for the corresponding non-referenced subvector is generated by using the LBG algorithm, and a normalized codebook is generated by normalizing the initial codebook.

14A is an example of a normalized codebook, and FIG. 14B is an example of a red-eye codebook. 14 shows an example of a two-dimensional bag Χι Χ ₂ ] due to the limitation of the ground. By normalizing the codebook, as shown in Fig. 14A, the range of values of each component is made constant (0-1 or 1-10). Then, the adaptive codebook generation unit 350 generates a red-eye codebook by applying the minimum value and the maximum value to the normalized codebook. At this time, the following equation may be used.

[Equation 10]

Codebooka _dap = (Max-Mix) (Codebook _n + Min)

Codebooka _d ap is the adaptive codebook, Codebook _n is the normalized codebook, Max is the maximum, Mix is the minimum

In order to quantize non-reference subvectors in this manner, code vectors with components smaller than the minimum value or larger than the maximum value correspond to redundancy. Therefore, by removing such redundancy and including code vectors close to the target sub-vector in the codebook, the sound quality can be improved by minimizing the quantization error without increasing the number of bits.

The adaptive codebook is delivered to the subordinate quantization unit 340 by the red-eye codebook generation unit 350 generated in the above manner.

Referring back to FIG. 6, the dependent quantizer 340 vectorizes the first non-reference subvector using the adaptive codebook to generate one or more non-reference codevectors. Specifically, the first dependent quantization unit 340-1 generates a first non-reference codevector by quantizing the first non-reference subvector using a red-eye codebook for the first non-reference subvector. The M-th dependent quantization unit 340 -M generates a first non-reference codevector by quantizing the M non-reference subvector using an adaptive codebook for the M non-reference subvector. In this way, the bitstream includes a reference codebook index corresponding to the reference codevector, which _is the output of the independent and dependent quantization units ₃₃₀ and _340, and a non-reference codebook index corresponding to the non-reference codevector.

Like the first embodiment described above with reference to FIG. 1, the encoder 300 according to the second embodiment may further include a multiplexer, a second stage quantization unit, and a prediction selector, wherein the function of each component is described above. Since it is almost similar to the description, a detailed description thereof will be omitted.

The audio signal processing apparatus according to the present invention can be included and used in various products. These products can be broadly divided into stand alone and portable groups, which can include TVs, monitors, and set-top boxes, and portable groups can include PMPs, mobile phones, and navigation. Can be.

FIG. 15 is a diagram illustrating a relationship between products in which an audio signal processing device according to an embodiment of the present invention is implemented. FIG. First, referring to FIG. 15, the wired / wireless communication unit 510 receives a bitstream through a wired / wireless communication method. Specifically, the wired / wireless communication unit 510 may include at least one of a wired communication unit 510A, an infrared communication unit 510B, a Bluetooth unit 510C, and a wireless LAN communication unit 510D.

The user authentication unit 520 receives user information and performs user authentication. The user authentication unit 520 may include one or more of the fingerprint recognition unit 520A, the interest recognition unit 520B, the face recognition unit 520C, and the voice recognition unit 520D. Can include, respectively, fingerprint, interest information, facial contour information, voice information received, converted into user information, the user User authentication may be performed by determining whether the information matches the existing registered user data.

The input unit 530 is an input device for the user to input various types of commands, and may include one or more of a keypad unit 530A, a touch pad unit 530B, and a remote controller unit 530C. The present invention is not limited thereto.

The signal coding unit 540 encodes or decodes an audio signal and / or a video signal received through the wired / wireless communication unit 510 and outputs an audio signal of a time domain. Audio signal processing apparatus 545, which is an embodiment of the invention described above (i.e., encoder 100 and / or decoder 200 according to the first embodiment, encoder 300 according to the second embodiment). And / or decoder 400), the audio processing apparatus 545 and the signal coding unit including the same may be implemented by one or more processors.

The controller 550 receives input signals from the input devices and controls all processes of the signal decoding unit 540 and the output unit 560. The output unit 560 is a component in which an output signal generated by the signal decoding unit 540 is output, and may include a speaker unit 560A and a display unit 560B. When the output signal is an audio signal, the output signal is output to the speaker, and when the output signal is a video signal, the output signal is output through the display.

16 is a relationship diagram of products in which an audio signal processing apparatus according to an embodiment of the present invention is implemented. FIG. 16 illustrates a relationship between a terminal and a server corresponding to the product illustrated in FIG. 15. Referring to FIG. 16A, the first terminal 500. And the second terminal (500.2) it can be seen that each terminal can communicate data to the bitstream in both directions through the wired or wireless communication unit. Referring to FIG. 16B, it can be seen that the server 600 and the first terminal 500.1 may also perform wired or wireless communication with each other.

The audio signal processing method according to the present invention can be stored in a computer-readable recording medium which is produced as a program for execution in a computer, and multimedia data having a data structure according to the present invention can also be stored in a computer-readable recording medium. Can be stored. The computer readable recording medium includes all kinds of storage devices for storing data that can be read by a computer system. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, and may also be implemented in the form of carrier wave (for example, transmission over the Internet). Include. In addition, the bitstream generated by the encoding method may be stored in a computer-readable recording medium or transmitted through a wired / wireless communication network. As described above, although the present invention has been described by way of limited embodiments and drawings, the present invention is not limited to this and the technical spirit of the present invention and the following by those skilled in the art to which the present invention pertains. Of course, various modifications and variations are possible within the scope of the appended claims. [Industrial Applicability] The present invention can be applied to encoding and decoding an audio signal.

Claims

[Range of request]

[Claim 1]

Generating a target vector for performing a plurality of linear-prediction transform coefficients by performing a linear-prediction analysis on the audio signal of the current frame;

Generating the target vector with a plurality of subvectors including a first subvector and a second subvector;

Obtaining a first code vector by vector quantizing the first subvector;

Removing one or more redundant code vectors from the initial codebook for the second subvector using the last component of the first code vector;

Generating an update codebook by adding one or more additional code vectors to the initial codebook using the last component of the first code vector; And obtaining a second code vector by vector quantizing the second sub vector using the update codebook.

Wherein the additional code vector is predicted based on a last component of the first code vector.

[Claim 2]

The method of claim 1,

And the number of redundant code vectors is equal to the number of additional code vectors.

[Claim 3]

The method of claim 1,

Generating the update codebook,

Obtaining an initial component prediction value using the last component of the first code vector;

Selecting two or more candidate code vectors from the initial codebook using the original component prediction value;

Generating a temporary code vector by interpolating the candidate code vectors;

Selecting code vectors that satisfy an ordering property among the temporary code vectors as the additional code vector; And,

Generating the update codebook by adding the additional code vector to the initial codebook.

[Claim 4]

The method of claim 3, wherein

The first component prediction value is obtained by adding a gradient mean to the last component of the first code vector,

And the increasing average is based on the difference between the current component and the previous component of the first code vector.

[Claim 5]

The method of claim 3,

The two or more candidate code vectors are

And selecting the first component of the code vectors included in the codebook for the second sub vector based on a difference between the first component predicted value and the first component predicted value.

[Claim 6]

The method of claim 1,

The update codebook,

The audio signal processing method according to the ascending order, characterized in that a first index is given to a code vector located after the redundant code vector in the initial codebook.

[Claim 7]

A linear predictor for performing a linear-prediction analysis on the audio signal of the current frame, thereby generating a target vector for the plurality of linear-prediction transform coefficients; A vector divider configured to generate the target vector from a plurality of subvectors including a first subvector and a second subvector;

A first quantizer for obtaining a first code vector by vector quantizing the first sub vector;

Using the last component of the first code vector, removing one or more redundant code vectors from the initial codebook for the second subvector, and using the last component of the first code vector, one or more in the initial codebook A codebook updater for generating an update codebook by adding an additional code vector; And,

A first quantizer for obtaining a second code vector by vector quantizing the second sub vector using the update codebook,

[Claim 8]

Receiving a first codebook index of a first code vector and a second codebook index of a second code vector;

Obtaining a last component of the first code vector using a first codebook index;

Using the last component of the first code vector to remove one or more redundant code vectors from an initial codebook for a second subvector;

Creating an update codebook by adding one or more additional code vectors to the initial codebook using the last component of the first code vector; Using the update codebook and the second codebook index, obtaining a second subvector;

And the additional code vector is predicted using the last component of the first code vector.

[Claim 9]

Generating a plurality of linear-predictive transform coefficients by performing a linear-prediction analysis on the audio signal of the current frame;

Generating one or more reference subvectors and one or more non-reference subvectors by arranging the plurality of linear-predictive transform coefficients;

Obtaining a reference code vector by performing vector quantization on the one or more reference subvectors;

Estimating the minimum and maximum values of the non-referenced subvectors using two components of each component of the reference code vectors;

Generating a red-eye codebook for the non-referenced subvector using the minimum and the maximum values; And,

And obtaining a non-reference code vector by performing vector quantization on the non-reference subvector using the decoded codebook.

[Claim 10]

The method of claim 9,

Generating the ungwoo codebook;

Obtaining normalized code vectors; And ，

Generating the red-eye codebook by applying the minimum value and the maximum value to the normalized codevectors.

[Claim 11]

The method of claim 9.

The reference code vector comprises a first reference code vector,

Two components of the reference code vectors are the last component and the last previous component of the first reference code vector.

[Claim 12]

The method of claim 9,

The reference code vector comprises a first reference code vector and a second reference code vector,

Two components of the reference code vectors are the last component of the first reference code vector and the first component of the second reference code vector.

[Claim 13]

The method of claim 9,

Wherein the minimum value and the maximum value are estimated based on the ordering quality of the linear-predictive transform coefficients.

[Claim 14]

The method of claim 9,

The arrangement of the plurality of linear-predictive transform coefficients is

[Wj, W ₂ , W ₃ , W ₈ , W ₁₃ , W ₁₄ , Wis, Wie, W ₄ , W ₅ , W ₆ , W ₇ , W ₉ , Wio, W _n , W ₁₂ ]

[W _l5 W ₂ , W ₃ , W ₈ , W ₄ , W ₅ , W ₆ , W ₇ ]

[W _l5 W ₂ , W ₆ , W ₃ , W ₄ , W ₅ , W ₇ , W ₈ , W ₉ ]

[WW ₂ , W ₆ , W ₃ , W ₄ , W ₅ , W ₁₀ , W _n , W ₁₂ , W ₇ , W ₈ , W ₉ ], and

[W _b W ₂ , W ₆ , W ₃ , W ₄ , W ₅ , W ₇ , W ₈ , W ₁₂ , W ₉ , Wio, w _n , W ₁₃ , W ₁₄ , W ₁₅ ]

(W _x is one of the X-th linear-predictive transformation coefficients),

The reference subvector includes {Wi W ₂ W ₃ W ₈ } ₅ {Wn W ₁₄ W ₁₅ W ₁₆ }, {Wi W ₂ W ₆ }, {Wio W _H W ₁₂ }, and {W ₇ W ₈ W ₁₂ } (Where W _x is one or more of the X-th linear-predictive transformation coefficients),

The non-referenced subvectors include: {W ₄ W ₅ W ₆ W ₇ }, {W ₉ Wio Wn W ₁₂ }, {W ₃ W ₄ W ₅ }, {Wi W ₂ W ₆ }, {W ₁₀ W _n Wi ₂ }, {W ₇ W ₈ W ₁₂ }, where W _x is at least one of the X-th linear-predictive transform coefficients. [Claim 15]

A linear prediction analyzer for generating a plurality of linear-prediction transform coefficients by performing linear-prediction analysis on the audio signal of the current frame;

An array unit for generating one or more reference subvectors and one or more non-reference subvectors by arranging the plurality of linear-prediction transform coefficients;

An independent quantizer that obtains a reference code vector by performing vector quantization on the one or more reference subvectors;

A red-eye codebook generator for estimating the minimum and maximum values of the non-referenced sub-vectors using two components of each of the components of the reference code vectors, and generating the red-hero codebook using the minimum and the maximum values; And,

And a dependent quantizer for acquiring a non-reference code vector by performing vector quantization on the non-reference subvector using the decoded codebook. [Claim 16]

Receiving a reference codebook index of one or more reference code vectors, and a non-reference codebook index of one or more non-reference code vectors;

Obtaining one or more reference code vectors using the one or more reference codebook indices;

Estimating a minimum and maximum value of a non-reference subvector using two components of each component of the reference code vectors;

Generating an adaptive codebook for the non-referenced subvector using the minimum and maximum values; And,

And obtaining the non-reference subvector using the decoded codebook and the non-reference codebook index.