CN101030376A - Combined method for searching energy parameter gain shape quantization of 0.6kb/s voice coder - Google Patents

Abstract

A united searching method of vocoder energy parameter gain form quantization includes obtaining energy average value after energy parameter of three continuous voice frames is obtained, obtaining quantization index by making uniform scalar quantization on energy average value, using obtained index as center to confirm a candidate region, carrying out counter scalar quantization on value in said region, obtaining normalized vector by using counter scalar quantized value to make normalization on three said frames and carrying out vector quantization on all values in said region to seek out correct value and codebook index.

Description

0.6kb/s the combined method for searching that vocoder energy parameter gain shape quantizes

Technical field

The invention belongs to the speech coding technology field, particularly low rate parametric speech coding technology.

Background technology

Voice coding in communication system, voice storage-playback, have in the consumer product of phonetic function and be widely used.International Telecommunication Union, some regional organizations and some countries had formulated a series of voice compression coding standards in succession in the last few years, were that 1.2kb/s has obtained gratifying voice quality to 16kb/s in code rate.Domestic and international research mainly concentrates on the following high-quality speech compressed encoding of 1.2kb/s speed at present, is mainly used in radio communication, secret communication, high capacity voice storage playback etc.0.6kb/s the vocoder algorithm is a focus wherein, the present invention proposes a kind ofly to unite the Syndicating search algorithm of quantification at energy parameter gain shape in the 0.6kb/s vocoder algorithm, can reduce the quantization error of energy parameter, improves the quality of synthetic speech.

Summary of the invention

The object of the present invention is to provide a kind of quantization error that can reduce energy parameter, improve the combined method for searching of the energy parameter gain shape quantification of 600b/s vocoder synthetic speech quality.

The invention is characterized in that this method realizes successively according to the following steps in digital integrated circuit:

Step (1) is divided frame to the input speech signal sampling point in proper order by the time interval of setting; Wherein said voice signal sampling point is the signal sampling point after having disturbed according to the setpoint frequency sampling and through high-pass filtering removal power frequency;

Step (2) is formed superframe to the one group of ground of every adjacent three subframes of subframe that obtains through undue frame in the step (1), asks for the energy parameter g of each subframe in the current superframe respectively according to following formula _{N, i}:

$g_{n,i}=\log \sqrt{\frac{1}{L_i}\underset{L_i}{\mathrm{\Σ}}s{\left(n\right)}^2}$

Wherein, n represents current superframe sequence number,

I represents the sequence number of i subframe in the current superframe, i=1, and 2,3,

L _iBe the length of window that the i subframe is asked for energy parameter,

S (n) is described L _iInterior voice signal;

Step (3) is asked for the mean value g of each subframe energy parameter in the current superframe by following formula _n:

${\stackrel{\‾}{g}}_n=\frac{1}{3}\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{g}_{n,i};$

Step (4) is with the mean value g of the energy parameter asked in (3) _nCarry out even scalar quantization by following formula, the span of its scalar quantization is between 10dB～77dB, uses 5 bits, thereby obtains the scalar quantization index value, represents with l:

Expression rounds downwards;

The index value l that step (5) obtains with step (4) is the center, determines the interval ψ of a scalar quantization candidate index according to following formula:

ψ＝[l-2，l+2]；

Step (6) is to each index value l in the candidate index interval of obtaining of step (5) _mAsk for the value after its reactionary slogan, anti-communist poster amount quantizes respectively

$\widehat{g_{n,l_m}}=\frac{77-10}{32}{l}_m+10,$ l _mIt is the value that belongs to ψ between the index area;

Each subframe energy parameter g of current superframe that step (7) obtains step (2) _{N, i}Respectively divided by the value of asking in the step (6)

Obtain the new vector (g after the normalization _{N, 1}', g _{N, 2}', g _{N, 3}'), wherein $g_{n,1}^{\′}=g_{n,1}/\widehat{g_{n,l_m}};$

Step (8) is carried out vector quantization with the Codebook of Vector Quantization of setting to the vector of asking in the step (7), and vector quantization adopts the method for full search, finds to make error in Codebook of Vector Quantization $E_{l_m,k}=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{(g_{n,i}-\widehat{g_{n,l_m}}*g_{k,i}^{\′\′})}^2$ Minimum index value $k=\underset{k}{\arg \min }{E}_{l_m,k},$ g _{K, i}" index is an i component of the three-dimensional code word vector of k in the quantification code book of representative setting, i=1,2,3;

Step (9) is to all l in the interval ψ of candidate index _mValue is carried out the calculating of (7), (8) two formulas, finds to make E _{Lm, k}Minimum l _mWith k, as the final quantized result of energy parameter.

Characteristics of the present invention are the gain shape associating quantization method of energy parameter have been adopted the method for Syndicating search.Original technology does not adopt the Syndicating search algorithm, but adopts progressively search, can not effectively reduce the quantization error of energy parameter.The present invention adopts gain shape to unite the combined method for searching of quantification, can further reduce the quantization error about 1% of energy parameter.

This method can reduce the quantization error of energy parameter, improves the naturalness of synthetic speech.The most suitable 600b/s low rate of this method parametric speech coding will be realized on signal processor chip DSP.

Description of drawings

The 0.6kb/s vocoder energy parameter gain shape that Fig. 1 proposes for the present invention is united the combined method for searching FB(flow block) of quantification.

Embodiment

The combined method for searching that the 0.6kb/s vocoder energy parameter gain shape that the present invention proposes is united quantification reaches embodiment in conjunction with the accompanying drawings and further specifies as follows:

Method flow of the present invention may further comprise the steps as shown in Figure 1:

Step (1) is divided frame to the input speech signal sampling point in proper order by the time interval of setting; Wherein said voice signal sampling point is the signal sampling point after having disturbed according to the setpoint frequency sampling and through high-pass filtering removal power frequency;

Step (2) is formed superframe to the one group of ground of every adjacent three subframes of subframe that obtains through undue frame in the step (1), asks for the energy parameter g of each subframe in the current superframe respectively according to following formula _{N, i}:

$g_{n,i}=\log \sqrt{\frac{1}{L_i}\underset{L_i}{\mathrm{\Σ}}s{\left(n\right)}^2}$

Wherein, n represents current superframe sequence number,

I represents the sequence number of i subframe in the current superframe, i=1, and 2,3,

L _iBe the length of window that the i subframe is asked for energy parameter,

S (n) is described L _iInterior voice signal;

Step (3) is asked for the mean value g of each subframe energy parameter in the current superframe by following formula _n:

${\stackrel{\‾}{g}}_n=\frac{1}{3}\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{g}_{n,i};$

Step (4) is with the mean value g of the energy parameter asked in (3) _nCarry out even scalar quantization by following formula, the span of its scalar quantization is between 10dB～77dB, uses 5 bits, thereby obtains the scalar quantization index value, represents with l:

Expression rounds downwards;

The index value l that step (5) obtains with step (4) is the center, determines the interval ψ of a scalar quantization candidate index according to following formula:

ψ＝[l-2，l+2]；

Step (6) is to each index value l in the candidate index interval of obtaining of step (5) _mAsk for the value after its reactionary slogan, anti-communist poster amount quantizes respectively

$\widehat{g_{n,l_m}}=\frac{77-10}{32}{l}_m+10,$ l _mIt is the value that belongs to ψ between the index area;

Each subframe energy parameter g of current superframe that step (7) obtains step (2) _{N, i}Respectively divided by the value of asking in the step (6)

Obtain the new vector (g after the normalization _{N, 1}', g _{N, 2}', g _{N, 3}'), wherein $g_{n,1}^{\′}=g_{n,1}/\widehat{g_{n,l_m}};$

Step (8) is carried out vector quantization with the Codebook of Vector Quantization of setting to the vector of asking in the step (7), and vector quantization adopts the method for full search, finds to make error in Codebook of Vector Quantization $E_{l_m,k}=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{(g_{n,i}-\widehat{g_{n,l_m}}*g_{k,i}^{\′\′})}^2$ Minimum index value $k=\underset{k}{\arg \min }{E}_{l_m,k},$ g _{K, i}" index is an i component of the three-dimensional code word vector of k in the quantification code book of representative setting, i=1,2,3;

Step (9) is to all l in the interval ψ of candidate index _mValue is carried out the calculating of (7), (8) two formulas, finds to make E _{Lm, k}Minimum l _mWith k, as the final quantized result of energy parameter.

The specific embodiment of each step of said method of the present invention is described in detail as follows respectively:

Said method step (1) divides the embodiment of frame to be by the 8kHz frequency sampling, to remove the voice sampling point that power frequency is disturbed through high-pass filtering to the input speech signal sampling point in chronological order.Every 25ms, just 200 voice sampling points constitute a subframe;

The embodiment of said method step (2) is: each frame of the superframe that three speech frames are formed is by following formulas Extraction gain parameter g _{N, i}:

$g_{n,i}=\log \sqrt{\frac{1}{L_i}\underset{L_i}{\mathrm{\Σ}}s{\left(n\right)}^2}$

Wherein, n is the sequence number of current superframe, and i is the sequence number of i subframe of current superframe, and L is the length of window, is made as long 200 sampling points of current subframe, and s (n) promptly is aforesaid 8kHz frequency sampling, removes the voice signal that power frequency is disturbed through high-pass filtering.

The embodiment of said method step (3) is: the average energy g that asks for current superframe according to following formula _n:

${\stackrel{\‾}{g}}_n=\left(\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{g}_{n,i}\right)/3$

The embodiment of said method step (4) is: with g _nCarry out even scalar quantization, in order to improve the quantification performance, it is quantized span be limited between 10dB～77dB, overflow and underflow adopt amplitude limiting processing, and this parameter represents that with 5 bits the quantization index value is obtained by following formula:

Wherein

For rounding symbol.

Embodiment in the said method step (5) is: with the value l that asks in the step (4) is the interval ψ of the selected index candidate in center, wherein

ψ＝[l-2，l+2]

The embodiment of said method step (6) is: to the index value l in the interval ψ _mCarry out the reactionary slogan, anti-communist poster amount and quantize, the value of asking for behind the inverse quantization is

Obtain by following formula:

$\widehat{g_{n,l_m}}=\frac{77-10}{32}{l}_m+10$

The embodiment of said method step (7) is: the energy parameter of each frame that current superframe is tried to achieve in step (2) is divided by in the step (6)

Obtain the new vector (g after the normalization _{N, 1}', g _{N, 2}', g _{N, 3}'), wherein

$g_{n,i}^{\′}=g_{n,i}/{\hat{g}}_{n,l_m}$

The embodiment of said method step (8) is: the normalized vector of asking in the step (7) is carried out vector quantization, and the code book of vector quantization is as shown in the table:

Quantization index value k	Code word vector g k″
		0	0.770791，0.856724，1.373817
1	0.726911，1.055533，1.217957
		2	1.216932，0.911708，0.873279
3	0.881003，1.042607，1.076917
		4	1.159534，1.037779，0.803366
5	1.061736，1.003369，0.935548
		6	0.939219，0.911485，1.149099
7	0.998837，0.999830，1.002373

The search criteria of vector quantization is to make error E _{Lm, k}Minimum, wherein E _{Lm, k}Be calculated as follows:

$E_{l_m,k}=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{(g_{n,i}-\widehat{g_{n,l_m}}*g_{k,i}^{\′\′})}^2$

Wherein, g _{K, i}" index is an i component of the vector of k in the representative quantification code book, k=1...8, i=1...3;

The embodiment of said method step (9) is: all values in the interval ψ of candidate index is carried out the calculating of step (7), (8), find to make E _{Lm, k}Minimum l _mWith k, as the final quantized result of energy parameter.

Claims (3)

1,0.6kb/s vocoder energy parameter gain shape is united the combined method for searching of quantification, it is characterized in that, this method realizes in digital integrated circuit successively according to the following steps:

Step (1) is divided frame to the input speech signal sampling point in proper order by the time interval of setting; Wherein said voice signal sampling point is the signal sampling point after having disturbed according to the setpoint frequency sampling and through high-pass filtering removal power frequency;

Step (2) is formed superframe to the one group of ground of every adjacent three subframes of subframe that obtains through undue frame in the step (1), asks for the energy parameter g of each subframe in the current superframe respectively according to following formula _N.i:

$g_{n.i}=\log \sqrt{\frac{1}{L_i}\underset{L_i}{\mathrm{\Σ}}s{\left(n\right)}^2}$

Wherein, n represents current superframe sequence number,

I represents the sequence number of i subframe in the current superframe, i=1, and 2,3,

L _iBe the length of window that the i subframe is asked for energy parameter,

S (n) is described L _iInterior voice signal;

Step (3) is asked for the mean value g of each subframe energy parameter in the current superframe by following formula _n:

${\stackrel{\‾}{g}}_n=\frac{1}{3}\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{g}_{n,i};$

Step (4) is with the mean value g of the energy parameter asked in (3) _nCarry out even scalar quantization by following formula, the span of its scalar quantization is between 10dB～77dB, uses 5 bits, thereby obtains the scalar quantization index value, represents with l:

Expression rounds downwards;

The index value l that step (5) obtains with step (4) is the center, determines the interval ψ of a scalar quantization candidate index according to following formula:

ψ＝[l-2，l+2]；

Step (6) is to each index value l in the candidate index interval of obtaining of step (5) _mAsk for the value after its reactionary slogan, anti-communist poster amount quantizes respectively

$g_{n,l_m}^{^}=\frac{77-10}{32}{l}_m+10,$ l _mIt is the value that belongs to ψ between the index area;

Each subframe energy parameter g of current superframe that step (7) obtains step (2) _N.iRespectively divided by the value of asking in the step (6)

Obtain the new vector (g after the normalization _{N, 1}', g _{N, 2}', g _{N, 3}'), wherein $g_{n,1}^{\′}=g_{n,1}/g_{n,l_m}^{^};$

Step (8) is carried out vector quantization with the Codebook of Vector Quantization of setting to the vector of asking in the step (7), and vector quantization adopts the method for full search, finds to make error in Codebook of Vector Quantization $E_{l_m,k}=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{(g_{n,i}-g_{n,l_m}^{^}*g_{k,i}^{\′\′})}^2$ Minimum index value

$k=\underset{k}{\arg \min E_{l_m,k},}{g}_{k,i}^{\′\′}$ Index is an i component of the three-dimensional code word vector of k in the quantification code book that representative is set, i=1,2,3;

Step (9) is to all l in the interval ψ of candidate index _mValue is carried out the calculating of (7), (8) two formulas, finds to make E _{Lm, k}Minimum l _mWith k, as the final quantized result of energy parameter.

2, unite the combined method for searching of quantification by the described 0.6kb/s vocoder of claim 1 energy parameter gain shape, it is characterized in that, the number of sub frames in the described superframe is more than or equal to 2.

3, uniting the combined method for searching of quantification by the described 0.6kb/s vocoder of claim 2 energy parameter gain shape, it is characterized in that, is that ψ between a candidate regions is chosen at the center with the index value after the uniform quantization of initially asking in the step (5).

Images