CN104282311A - Quantitative method and device for sub-band division in audio coding bandwidth expansion - Google Patents

Abstract

The invention relates to a quantitative method and device for sub-band division in audio coding bandwidth expansion. Training sample high-frequency signals are divided four times in sequence, sample point division points are obtained according to the minimum value of the product of sample point standard deviations of two new sub-bands every time the training sample high-frequency signals are divided, finally, the sixteen sample point division points are obtained, the input audio coding bandwidth is expanded on the basis of the sixteen sample point division points, the better high-frequency spectrum structure can be obtained and is more similar to the frequency spectrum shape, and therefore the high frequency spectrum rebuilding quality is improved. According to the sub-band division method, the statistical property in sub-band quantization is sufficiently considered; compared with a Bark band or uniform division mode, the subjective hearing MOS is slightly improved, and the signal to noise ratio is remarkably improved.

Description

The quantization method of sub-band division and device in a kind of audio coding bandwidth expansion

Technical field

The present invention relates to audio coding field, particularly relate to quantization method and the device of sub-band division in a kind of audio coding bandwidth expansion.

Background technology

Psychologic acoustics research shows that people have difference for the susceptibility under audio frequency different frequency, more responsive to low frequency, and insensitive to high frequency, therefore in audio coding, adopts less information to carry out encoding to save code check for HFS.And HFS lack the discomfort can brought in sense of hearing completely, therefore the mode of bandwidth expansion is often adopted to recover high frequency, and in bandwidth expansion, all can use sub-band division, conventional subband divides the mode often using Bark band or evenly divide, and this mode have ignored the statistical property in quantized subband.Therefore by carrying out statistical study to quantitative information, and carrying out sub-band division accordingly, favourable minimizing quantizing distortion, improving bandwidth extension encoding quality.

Summary of the invention

The object of the invention is to the defect for overcoming prior art, and quantization method and the device of sub-band division in a kind of audio coding bandwidth expansion are provided.

For achieving the above object, the present invention is by the following technical solutions: the quantization method of sub-band division in a kind of audio coding bandwidth expansion, comprises the following steps:

S1, for the high-frequency signal of training sample, every frame is N number of sampling point, gets any sampling point position m one by one, high-frequency signal is divided into two subbands between sampling point 1 to sampling point N, calculates two subband sampling point standard deviation product K _{m [i]}, obtain one group of sampling point standard difference-product vector { S _{j [i]},

$\begin{array}{c}{K}_{m\left[i\right]}=\left[\sqrt{\frac{1}{m}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{(N-m)}\underset{j=m+1}{\overset{N}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]\\ \left\{S_{j\left[i\right]}\right\}=\{K_2,K_3,K_4,K_5,....K_{N-1}\}\end{array}$

Wherein i=1, m=2,3,4,5 ... N-1, x _jrepresent sampling point signal, with represent the arithmetic mean of two subband sampling points,

If K _{m [i]}for { S _{j [i]}in least member, then get the cut-point position that m value is division two subbands, make sampling point cut-point P [i] ₁₁=m ₁₁;

S2, newly divides two subbands of formation to step S1, get any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points _{m [i]}, and obtain two groups of sampling point standard difference-product vector { S _{j [i]},

$\begin{array}{c}{K}_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]\\ \left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}\end{array}$

Wherein i=1,2, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband,

If K _{m [i]}for { S _{j [i]}in least member, then get m ₂₁, m ₂₂value is the cut-point position of division two subbands, makes sampling point cut-point P [i] ₂₁=m ₂₁, P [i] ₂₂=m ₂₂;

S3, divides four subbands formed, gets any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points to new step S2 _{m [i]}, and obtain four groups of sampling point standard difference-product vector { S _{j [i]},

$\begin{array}{c}{K}_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]\\ \left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}\end{array}$

Wherein i=1,2,3,4, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband, if K _{m [i]}for { S _{j [i]}in least member, then get m ₃₁, m ₃₂, m ₃₃, m ₃₄value is the cut-point position of division four subbands, makes sampling point cut-point P [i] ₃₁=m ₃₁, P [i] ₃₂=m ₃₂, P [i] ₃₃=m ₃₃, P [i] ₃₄=m ₃₄;

S4, newly divides eight subbands of formation to step S3, get any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points _{m [i]}, and obtain eight groups of sampling point standard difference-product vector { S _{j [i]},

$\begin{array}{c}{K}_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m_s+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]\\ \left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}\end{array}$

Wherein i=1,2,3,4,5,6,7,8, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband, if K _{m [i]}for { S _{j [i]}in least member, then get m ₄₁, m ₄₂, m ₄₃, m ₄₄, m ₄₅, m ₄₆, m ₄₇, m ₄₈value is the cut-point position of division two subbands, makes sampling point cut-point P [i] ₄₁=m ₄₁, P [i] ₄₂=m ₄₂, P [i] ₄₃=m ₄₃, P [i] ₄₄=m ₄₄, P [i] ₄₅=m ₄₅, P [i] ₄₆=m ₄₆, P [i] ₄₇=m ₄₇, P [i] ₄₈=m ₄₈, finally obtain 16 subbands newly divided.

In described step S1, S2, S3 divide formed new subband in, as wherein any one subband only comprises 1 sampling point, in the respective subsequent step of described step S1, S2, S3, only the subband comprising more than 2 or 2 sampling points is divided, the new division number of sub-bands then exported in step s 4 which is less than 16, and the quantity of sampling point cut-point is less than eight.

Every frame sampling point N of described step S1 is 256,512 or 1024.

The high-frequency signal of described training sample is 400,000 frames.

A device for sub-band division in audio coding bandwidth expansion, comprises the first sub-band division module, the second sub-band division module, the 3rd sub-band division module, the 4th sub-band division module, wherein,

Described first sub-band division module, for the high-frequency signal of input is divided into 2 subbands, split under the condition that the sampling point standard deviation product of two subbands is minimum, obtained two subband signals are outputted to the second sub-band division module, and obtain first time sampling point cut-point;

Described second sub-band division module, for the high-frequency signal being divided into 2 subbands that basis is inputted by the first sub-band division module, by each subband signal again row be divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, and output to the 3rd sub-band division module, obtain second time sampling point cut-point simultaneously;

Described 3rd sub-band division module, for the high-frequency signal being divided into 4 subbands that basis is inputted by the second sub-band division module, each subband signal is further divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, and output to the 4th sub-band division module, obtain third time sampling point cut-point simultaneously;

Described 4th sub-band division module, for the high-frequency signal being divided into 8 subbands that basis is inputted by the 3rd sub-band division module, each subband signal is further divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, obtain final 16 subbands divide, acquisition the 4th sampling point cut-point simultaneously.

When the subband sampling point that described second sub-band division module, described 3rd sub-band division module, described 4th sub-band division module receive is 1, then only the subband that sampling point is more than 2 is split further.

The present invention's beneficial effect is compared with prior art: the sub-band division method that the present invention uses takes into full account the statistical property in quantized subband, compare Bark band or even dividing mode, subjective hearing MOS of the present invention divides and slightly promotes, and signal to noise ratio (S/N ratio) has obvious lifting.

Below in conjunction with the drawings and specific embodiments, the invention will be further described.

Accompanying drawing explanation

Fig. 1 is be the quantization method FB(flow block) of sub-band division in audio coding bandwidth expansion of the present invention;

Fig. 2 is be the quantization device schematic diagram of sub-band division in audio coding bandwidth expansion of the present invention.

Embodiment

In order to more fully understand technology contents of the present invention, below in conjunction with specific embodiment technical scheme of the present invention being introduced further and illustrating.

As shown in Figure 1, the quantization method of sub-band division in a kind of audio coding bandwidth expansion, comprises the following steps:

Step S1, for the high-frequency signal of training sample 400,000 frame, every frame is N number of sampling point, gets any sampling point position m one by one, high-frequency signal is divided into two subbands between sampling point 1 to sampling point N, calculates two subband sampling point standard deviation product K _{m [i]}, obtain one group of sampling point standard difference-product vector { S _{j [i]},

$\begin{array}{c}{K}_{m\left[i\right]}=\left[\sqrt{\frac{1}{m}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{(N-m)}\underset{j=m+1}{\overset{N}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]\\ \left\{S_{j\left[i\right]}\right\}=\{K_2,K_3,K_4,K_5,....K_{N-1}\}\end{array}$

Wherein i=1, m=2,3,4,5 ... N-1, x _jrepresent sampling point signal, with represent the arithmetic mean of two subbands,

As shown in Figure 1, the quantization method of sub-band division in a kind of audio coding bandwidth expansion, comprises the following steps:

Step S2, newly divides two subbands of formation to step S1, get any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points _{m [i]}, and obtain two groups of sampling point standard difference-product vector { S _{j [i]},

$\begin{array}{c}{K}_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]\\ \left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}\end{array}$

Wherein i=1,2, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband,

If K _{m [i]}for { S _{j [i]}in least member, then get m ₂₁, m ₂₂value is the cut-point position of division two subbands, makes sampling point cut-point P [i] ₂₁=m ₂₁, P [i] ₂₂=m ₂₂then obtain four new subbands.

Step S3, divides four subbands formed, gets any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points to new step S2 _{m [i]}, and obtain four groups of sampling point standard difference-product vector { S _{j [i]},

$\begin{array}{c}{K}_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]\\ \left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}\end{array}$

Wherein i=1,2,3,4, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband, if K _{m [i]}for { S _{j [i]}in least member, then get m ₃₁, m ₃₂, m ₃₃, m ₃₄value is the cut-point position of division four subbands, makes sampling point cut-point P [i] ₃₁=m ₃₁, P [i] ₃₂=m ₃₂, P [i] ₃₃=m ₃₃, P [i] ₃₄=m ₃₄, then eight new subbands are obtained.

Step S4, newly divides eight subbands of formation to step S3, get any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points _{m [i]}, and obtain eight groups of sampling point standard difference-product vector { S _{j [i]},

$\begin{array}{c}{K}_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m_s+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]\\ \left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}\end{array}$

Wherein i=1,2,3,4,5,6,7,8, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband, if K _{m [i]}for { S _{j [i]}in least member, then get m ₄₁, m ₄₂, m ₄₃, m ₄₄, m ₄₅, m ₄₆, m ₄₇, m ₄₈value is the cut-point position of division two subbands, makes sampling point cut-point P [i] ₄₁=m ₄₁, P [i] ₄₂=m ₄₂, P [i] ₄₃=m ₄₃, P [i] ₄₄=m ₄₄, P [i] ₄₅=m ₄₅, P [i] ₄₆=m ₄₆, P [i] ₄₇=m ₄₇, P [i] ₄₈=m ₄₈, finally obtain 16 subbands newly divided.

In above-mentioned cutting procedure, there is a kind of special case, namely in step S1, S2, S3 divide in the new subband formed likely only containing a sampling point, the probability of even now is extremely low, 1 sampling point is only comprised as run into wherein any one subband, then this subband cannot continue segmentation further, in its subsequent step, can only divide the subband comprising more than 2 or 2 sampling points, the new division subband final amt then exported in step s 4 which is less than 16, and the quantity of sampling point cut-point is less than eight.

High-frequency signal obtains signal sampling point cut-point P [i] through this method ₁₁=m ₁₁, P [i] ₂₁=m ₂₁, P [i] ₂₂=m ₂₂, P [i] ₃₁=m ₃₁, P [i] ₃₂=m ₃₂, P [i] ₃₃=m ₃₃, P [i] ₃₄=m ₃₄, P [i] ₄₁=m ₄₁, P [i] ₄₂=m ₄₂, P [i] ₄₃=m ₄₃, P [i] ₄₄=m ₄₄, P [i] ₄₅=m ₄₅, P [i] ₄₆=m ₄₆, P [i] ₄₇=m ₄₇, P [i] ₄₈=m ₄₈, in the audio coding bandwidth expansion to input, use the above-mentioned cut-point determined to split, better high frequency spectrum structure can be obtained, closer to high frequency spectrum shape, thus improve high frequency spectrum reconstruction quality.

Concrete, every frame sampling point N of step S1 is 256,512 or 1024.

Further, the high-frequency signal of described training sample is 400,000 frames.

As shown in Figure 2, the device of sub-band division in a kind of audio coding bandwidth expansion, comprises the first sub-band division module 10, second sub-band division module 20, the 3rd sub-band division module 30, the 4th sub-band division module 40, wherein,

First sub-band division module 10, for the high-frequency signal of input is divided into 2 subbands, split under the condition that the sampling point standard deviation product of two subbands is minimum, obtained two subband signals are outputted to the second sub-band division module, and obtain first time sampling point cut-point.

Second sub-band division module 20, for the high-frequency signal being divided into 2 subbands that basis is inputted by the first sub-band division module, by each subband signal again row be divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, and output to the 3rd sub-band division module, obtain second time sampling point cut-point simultaneously.

Described 3rd sub-band division module 30, for the high-frequency signal being divided into 4 subbands that basis is inputted by the second sub-band division module, each subband signal is further divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, and output to the 4th sub-band division module, obtain third time sampling point cut-point simultaneously.

Described 4th sub-band division module 40, for the high-frequency signal being divided into 8 subbands that basis is inputted by the 3rd sub-band division module, each subband signal is further divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, obtain final 16 subbands divide, acquisition the 4th sampling point cut-point simultaneously.

Above-mentioned sub-band division device obtains first time sampling point cut-point, second time sampling point cut-point, third time sampling point cut-point the 4th sampling point cut-point, in the audio coding bandwidth expansion to input, the above-mentioned cut-point determined is used to split, better high frequency spectrum reconstruction signal can be obtained, reduce because frequency spectrum copies the probability of medium-high frequency distortion spectrum.

When second sub-band division module 20, described 3rd sub-band division module 30, described 4th sub-band division module 40 are when received subband sampling point is 1, then only the subband that sampling point is more than 2 is split further, obtain sampling point cut-point.

The above only further illustrates technology contents of the present invention with embodiment, so that reader is easier to understand, but does not represent embodiments of the present invention and is only limitted to this, and any technology done according to the present invention extends or recreation, all by protection of the present invention.

Claims (6)

1. the quantization method of sub-band division in audio coding bandwidth expansion, is characterized in that, comprise the following steps:

S1, for the high-frequency signal of training sample, every frame is N number of sampling point, gets any sampling point position m one by one, high-frequency signal is divided into two subbands between sampling point 1 to sampling point N, calculates two subband sampling point standard deviation product K _{m [i]}, obtain one group of sampling point standard difference-product vector { S _{j [i]},

$K_{m\left[i\right]}=\left[\sqrt{\frac{1}{m}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{(N-m)}\underset{j=m+1}{\overset{N}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]$

{S _j[i]}＝{K ₂,K ₃,K ₄,K ₅,....K _N-1}

Wherein i=1, m=2,3,4,5 ... N-1, x _jrepresent sampling point signal, with represent the arithmetic mean of two subband sampling points,

If K _{m [i]}for { S _{j [i]}in least member, then get the cut-point position that m value is division two subbands, make sampling point cut-point P [i] ₁₁=m ₁₁;

S2, newly divides two subbands of formation to step S1, get any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points _{m [i]}, and obtain two groups of sampling point standard difference-product vector { S _{j [i]},

$K_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]$

$\left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}$

Wherein i=1,2, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband,

If K _{m [i]}for { S _{j [i]}in least member, then get m ₂₁, m ₂₂value is the cut-point position of division two subbands, makes sampling point cut-point P [i] ₂₁=m ₂₁, P [i] ₂₂=m ₂₂;

S3, divides four subbands formed, gets any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points to new step S2 _{m [i]}, and obtain four groups of sampling point standard difference-product vector { S _{j [i]},

$K_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]$

$\left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}$

Wherein i=1,2,3,4, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband, if K _{m [i]}for { S _{j [i]}in least member, then get m ₃₁, m ₃₂, m ₃₃, m ₃₄value is the cut-point position of division four subbands, makes sampling point cut-point P [i] ₃₁=m ₃₁, P [i] ₃₂=m ₃₂, P [i] ₃₃=m ₃₃, P [i] ₃₄=m ₃₄;

S4, newly divides eight subbands of formation to step S3, get any sampling point position m within the scope of each sub-band division one by one, each subband is further subdivided into two new subbands, calculate the standard deviation product K of two new subband sampling points _{m [i]}, and obtain eight groups of sampling point standard difference-product vector { S _{j [i]},

$K_{m\left[i\right]}=\left[\sqrt{\frac{1}{m-m_s+1}\underset{j=m_s}{\overset{m}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2}\right]*\left[\sqrt{\frac{1}{m_e-m}\underset{j=m+1}{\overset{m_e}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x^{\′}})}^2}\right]$

$\left\{S_{j\left[i\right]}\right\}=\{K_{m_s+1},K_{m_s+2},....K_{m_e-1}\}$

Wherein i=1,2,3,4,5,6,7,8, m _sbe the starting position of i-th subband, m _ebe the end position of i-th subband, if K _{m [i]}for { S _{j [i]}in least member, then get m ₄₁, m ₄₂, m ₄₃, m ₄₄, m ₄₅, m ₄₆, m ₄₇, m ₄₈value is the cut-point position of division two subbands, makes sampling point cut-point P [i] ₄₁=m ₄₁, P [i] ₄₂=m ₄₂, P [i] ₄₃=m ₄₃, P [i] ₄₄=m ₄₄, P [i] ₄₅=m ₄₅, P [i] ₄₆=m ₄₆, P [i] ₄₇=m ₄₇, P [i] ₄₈=m ₄₈, finally obtain 16 subbands newly divided.

2. the quantization method of sub-band division in audio coding bandwidth expansion according to claim 1, it is characterized in that, in described step S1, S2, S3 divide formed new subband in, as wherein any one subband only comprises 1 sampling point, in the respective subsequent step of described step S1, S2, S3, only divide the subband comprising more than 2 or 2 sampling points, then the new division number of sub-bands exported in step s 4 which is less than 16, and the quantity of sampling point cut-point is less than eight.

3. the quantization method of sub-band division in audio coding bandwidth expansion according to claim 1, it is characterized in that, every frame sampling point N of described step S1 is 256,512 or 1024.

4. the quantization method of sub-band division in audio coding bandwidth expansion according to claim 1, it is characterized in that, the high-frequency signal of described training sample is 400,000 frames.

5. the device of sub-band division in audio coding bandwidth expansion, is characterized in that, comprise the first sub-band division module, the second sub-band division module, the 3rd sub-band division module, the 4th sub-band division module, wherein,

Described first sub-band division module, for the high-frequency signal of input is divided into 2 subbands, split under the condition that the sampling point standard deviation product of two subbands is minimum, obtained two subband signals are outputted to the second sub-band division module, and obtain first time sampling point cut-point;

Described second sub-band division module, for the high-frequency signal being divided into 2 subbands that basis is inputted by the first sub-band division module, by each subband signal again row be divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, and output to the 3rd sub-band division module, obtain second time sampling point cut-point simultaneously;

Described 3rd sub-band division module, for the high-frequency signal being divided into 4 subbands that basis is inputted by the second sub-band division module, each subband signal is further divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, and output to the 4th sub-band division module, obtain third time sampling point cut-point simultaneously;

Described 4th sub-band division module, for the high-frequency signal being divided into 8 subbands that basis is inputted by the 3rd sub-band division module, each subband signal is further divided into two subband signals, the sampling point standard deviation product of two new subbands that each subband newly marks off is minimum, obtain final 16 subbands divide, acquisition the 4th sampling point cut-point simultaneously.

6. the device of sub-band division in audio coding bandwidth expansion according to claim 5, it is characterized in that, when the subband sampling point that described second sub-band division module, described 3rd sub-band division module, described 4th sub-band division module receive is 1, then only the subband that sampling point is more than 2 is split further.