Background technology
Under the limited condition of encoding condition, for example in the limited communication environment or safety precaution monitoring environment of code check, only adopt common voice or audio coding method often to be difficult to obtain gratifying coding effect.At this moment, the auditory properties that need consider people's ear more is with further reduction code check: for example, in the limited coding method of early stage code check more bits is distributed to the low frequency part of audio frequency or voice signal, and the HFS of signal is distributed not even allocation bit less.This simply abandon or coding method that part is abandoned HFS makes encoder bit rate reach the requirement of target bit rate, but the decoded audio signal quality may obviously descend.
Bandwidth expansion BWE (Band Width Extension) technology is a kind of improvement technology of abandoning high frequency thought to aforementioned: it is encoded by choose suitable parameter at coding side, utilize these parameters in decoding end then, the signal " expansion " that the decoded frequency band of low frequency is narrower is to the broad more scope of frequency band range.At present, the BWE technology is widely used in low code check audio frequency or the application of voice (for example AMR-WB+) coding.
As BWE The Application of Technology example, described a kind of digital audio encoding apparatus 10 that typically comprises the BWE technology at Fig. 1, it has comprised the pretreatment module 12 that receives input PCM data, the low frequency coding nuclear 14 that is connected respectively to pretreatment module 12, bandwidth extension process module 16 and with the packing data module 18 of the output packing of the output of low frequency coding nuclear 14 and bandwidth extension process module 16.Also not shown in this article with audio coding system 10 corresponding audio decoding apparatus, but those skilled in the art can understand by reading specification, this audio decoding apparatus is such device: according to the side information in the coded data that is received, each step during reverse execution audio coding system 10 deal with data, finally the PCM data before decoding end recovers coding.Below in conjunction with Fig. 1, describe the operation principle of each module in the audio coding system 10 in detail:
● the effect of pretreatment module 12 has two broad aspect: at first, this module judges that it still is the voice class signal that the pcm audio data of sending into belong to the environmental classes signal; Secondly, the processing of this module by time domain one by one PCM data superframe be divided into high frequency time-domain signal and low frequency time-domain signal (will be described in more detail below).
● the low frequency time-domain signal that low frequency coding nuclear 14 receives from pretreatment module, time domain or transform domain (as, frequency domain) carries out compressed encoding (according to the result that pretreatment module 12 is judged, the compaction coding method of selecting for use may be different, typical ACELP, TCX etc. as mentioning among the AMR-WB+); The low frequency coded data is sent to packing data module 18.
The method of judging environmental classes signal and voice class signal in the pretreatment module 12 preferably is used to distinguish the method for music signal (the environmental classes signal has the characteristics of music signal) and voice signal in the prior art in the prior art, because this part content is not the content that the present invention pays close attention to, therefore no longer describe in detail.
● bandwidth extension process module 16 receives low frequency time-domain signal and the high frequency time-domain signal from pretreatment module, and they are carried out the bandwidth extension process, finally is used for the high frequency BWE parameter of reconstruction signal HFS to 18 outputs of packing data module.
● packing data module 18 receives low frequency coded data and high frequency BWE parameter and other side information (not shown), and they are packaged into the form that is suitable for transmitting, and mails to decoding end.
Before bandwidth expansion technique of the prior art exemplarily is described, at first the method that is used for the PCM signal of input is divided into high frequency time-domain signal and low frequency time-domain signal in the pretreatment module 12 is briefly described in conjunction with Fig. 2.
(the PCM signal with a superframe is an example) as shown in Figure 2, because the input sample frequency of original audio signal may be one of 16kHz, 24kHz, 32kHz and 48kHz, need resample in the preliminary treatment of input signal before coding with various different sample frequencys, be converted to the signal of inner sample frequency Fs (kHz).In like manner, the part of the reprocessing after decoding needs sample rate conversion equally.
Subsequently, be F with above-mentioned sample rate
S, altogether L PCM signal to send into by frequency respectively be F
S/ 4 low pass filter 124 and be F by frequency
S/ 4 high pass filter 122 carries out filtering; And then via down-sampled device 126,128, with the signal that obtains do respectively one 2 times critical down-sampled, obtain Fs/2 sampling low frequency signal x respectively
L(n) and Fs/2 sampling high-frequency signal x
H(n), subscript L and H represent low frequency and high frequency, and digital L represents the number (that is, length) of sample in the audio frequency superframe, are preferably 512,1024 or 2048 (numbers before down-sampled), and concrete length is determined by the rank parameter.Correspondingly, the length of an audio frame generally is fixed as 512 samples (number before down-sampled), and alleged herein " audio frame " is for carrying out the least unit of environmental classes or voice class coding.
It should be noted that, Fig. 2 only shows in the pretreatment module 12 cuts apart a relevant example with time domain, and be used to judge that the device of signal type is also not shown, added and judged that the pretreatment module 12 ' (not shown) of type of functionality is conspicuous for those skilled in the art.
Below in conjunction with several pieces of Chinese patents, BWE technology of the prior art is described.The technical scheme of a plurality of application BWE technology or improvement BWE technology has been proposed in the prior art:
1, applying on December 14th, 2006, is being disclosed in the Chinese patent application 200610165864.4 of on June 18th, 2008 (publication number CN 101202042A), the applicant discloses a kind of extendible audio coding framework, wherein adopt bandwidth expansion technique to handle noise signal in the PCM stream, efficiently rebuild the HFS of harmonic wave and noise like composition.This technical scheme discloses and how will have the BWE technology now and be applied in the audio coding system, provides improved suggestion but the BWE technology itself be there is no.
2, applying on December 12nd, 2007, be disclosed among the Chinese patent application 200710198774.X of on June 17th, 2009 (publication number CN 101458930A), the applicant discloses improving one's methods of a kind of BWE, adopts the synthetic again mode of folding low frequency to generate required high-frequency excitation signal.The characteristics of this method are: need further the high and low frequency pumping signal to be carried out synthetic filtering after generating the upper frequency pumping signal by spectrum folding.
3, applying on September 8th, 2006, be disclosed in the Chinese patent application 200610128778.6 on March 12nd, 2008 (publication number CN101140759A), the applicant discloses the bandwidth extended method of a kind of audio frequency or voice signal.The characteristics of this method are: need ask for the normalization composite filter earlier and rebuild high-frequency signal, again the time domain space more original high-frequency signal and rebuild high-frequency signal, obtain the required parameter (gain factor) of BWE.
This paper aims to provide a kind of brand-new B WE technology, and it can be implemented under the low code check and in high quality digital signal (for example signal of audio frequency or voice class) be encoded.
Summary of the invention
In order to reach purpose of the present invention, provide a kind of new bandwidth expansion technique based on non-equiband BWE gain.
According to one embodiment of present invention, provide a kind of bandwidth extended method, it can comprise: the time-domain signal that will be within the time period t 1-t2 by preliminary treatment is divided into high frequency time-domain signal and low frequency time-domain signal; By time-frequency conversion high frequency time-domain signal and low frequency time-domain signal are transformed to high frequency frequency-region signal and low frequency frequency-region signal respectively; And, based on the low frequency frequency-region signal of part and to the high frequency frequency-region signal of small part calculate corresponding to time period t 1-t2, at least one bandwidth expansion gain.
Further, this at least one bandwidth expansion gain can be directly proportional with following value: to the energy of the high frequency frequency-region signal of small part and with the energy of the low frequency frequency-region signal of part and ratio.
Further, the bandwidth expansion gain in the time period t 1-t2 can be no less than 2, and has different frequency resolutions.Preferably, when the expansion of the bandwidth in described time period t 1-t2 gain was 2, described frequency resolution can be respectively Fs/12 and Fs/6 (Fs represents sample frequency); When the expansion of the bandwidth in described time period t 1-t2 gain was 4, described frequency resolution can be respectively Fs/108, Fs/54, Fs/18, reach Fs/6.
Further, the expansion of the bandwidth in time period t 1-t2 gain can be 1.Preferably, Bu Fen low frequency frequency-region signal is the upper frequency part of low frequency frequency-region signal.
Further, bandwidth extended method according to the present invention also can comprise: judge whether the described low frequency frequency-region signal of described part and described described high frequency frequency-region signal to small part have string; When the described low frequency frequency-region signal of described part has string and described described high frequency frequency-region signal to small part when not having string, the described bandwidth expansion gain that decays is as actual bandwidth expansion gain.
In addition, bandwidth extended method according to the present invention also can comprise: vector quantization is carried out in all the bandwidth expansion gains in the frame.
According to still a further embodiment, provide a kind of bandwidth expansion means, it can comprise: pretreatment module, and its time-domain signal that is used for being within the time period t 1-t2 is divided into high frequency time-domain signal and low frequency time-domain signal; The time-frequency conversion module, it is used for high frequency time-domain signal and low frequency time-domain signal are transformed to high frequency frequency-region signal and low frequency frequency-region signal respectively; And gain calculation module, its based on the low frequency frequency-region signal of part and to the high frequency frequency-region signal of small part calculate corresponding to time period t 1-t2, at least one bandwidth expansion gain.
Further, at least one bandwidth expansion gain can be directly proportional with following value: to the energy of the high frequency frequency-region signal of small part and with the energy of the low frequency frequency-region signal of part and ratio.
Further, the bandwidth expansion gain in the time period t 1-t2 can be no less than 2, and has different frequency resolutions.Preferably, when the expansion of the bandwidth in described time period t 1-t2 gain was 2, described frequency resolution can be respectively Fs/12 and Fs/6 (Fs represents sample frequency); When the expansion of the bandwidth in described time period t 1-t2 gain was 4, described frequency resolution can be respectively Fs/108, Fs/54, Fs/18, reach Fs/6.
Further, the expansion of the bandwidth in time period t 1-t2 gain can be 1.Preferably, Bu Fen low frequency frequency-region signal can be the upper frequency part of low frequency frequency-region signal.
Further, also can comprise according to bandwidth expansion means of the present invention: judge module, it is used for the low frequency frequency-region signal of judgment part and whether has string to the high frequency frequency-region signal of small part; And attenuation module, when the low frequency frequency-region signal of part has string and when the high frequency frequency-region signal of small part does not have string, attenuation module with regard to attenuation bandwidth expansion gain as actual bandwidth expansion gain.
In addition, bandwidth extended method according to the present invention also can comprise: the vector quantization module, it is used for vector quantization is carried out in all the bandwidth expansion gains in the frame.
Based on technique scheme, arrived the brand-new BWE technology of utilizing, under low code check, realize the technique effect of high-quality coding.
Embodiment
By describing the preferred embodiments of the present invention hereinafter by accompanying drawing.Unnecessary details in the following description, function or the structure that becomes prior art will be described in detail, because will cause the ambiguous of introducing of the present invention.In addition, in the specification with accompanying drawing in the identical Reference numeral that uses represent same device or same step.
Hereinafter will specify BWE coding method 100 in conjunction with Fig. 3 A according to the embodiment of the invention:
●
step 101, to the high frequency time-domain signal x after handling through pretreatment module
H(n) and low frequency time-domain signal x
L(n) analysis filtered that quantizes respectively is to ask for the low frequency residual error e of each subframe
L(i is n) with high frequency residual error e
H(i, n), wherein i represents the sequence number of subframe, value is (I represents the number of sub-frames that comprises the superframe, also is the number of the gain that comprises in each superframe, and preferred value is L/128) from 0 to I-1.Wherein, the analysis filter that is used for the quantification of low frequency signal
(seeing Fig. 3 B) directly obtains from low frequency coding nuclear; The analysis filter A that is used for high-frequency signal
HFWhat (z) carry out is following processing: to high-frequency signal x
HF(i, each frame n) is asked for the LPC coefficient on one group of 8 rank, and the LPC coefficient is converted to the ISP coefficient, and ISP further is transformed into ISF parameter (electricity is led spectral frequency) again and uses 9 bit quantizations.The subframe of each 64 sampling point is carried out interpolation according to following formula to the ISP coefficient:
Q wherein
nISP coefficient q for present frame
(n-1)Be the ISP coefficient of former frame,
It is the ISP coefficient of i subframe interpolation of present frame.
By high frequency time-domain signal x
H(n) and low frequency time-domain signal x
L(n) ask for low frequency residual error e
L(i is n) with high frequency residual error e
H(i, n) more specifically implementation method can be with reference to following 1986 disclosed documents: P.Kabaland R.P.Ramachandran, " The computation of line spectral frequencies usingChebyshev polynomials ", IEEE Trans.on ASSP, vol.34, no.6, pp.1419-1426, Dec.1986.
● step 102, to low frequency residual error e
L(i is n) with high frequency residual error e
H(i n) carries out time-frequency conversion (being preferably 64 FFT conversion), obtains plural low frequency residual error spectrum E
LF(i is k) with plural high frequency residual error spectrum E
HF(i, k).
●
step 103, utilize plural low frequency residual error spectrum E
LF(i, part k) is carried out the frequency spectrum copy process.According to a preferred embodiment of the present invention, low frequency signal is a subframe with 64 in time, on frequency the low frequency residual error is composed E
LF(i k) is divided into low low frequency residual error spectrum E branch such as frequency
LFL(i is k) with low high frequency residual error spectrum E
LFH(i, k).When carrying out frequency spectrum when copy, use preferably that string is lower, low high frequency residual error spectrum E
LFH(k) duplicate twice, with the details component of the high frequency band that obtains estimating
, above-mentioned copy mode will partly be illustrated in conjunction with Fig. 6 A-6C after a while at this paper.It will be appreciated by persons skilled in the art that choosing of plural low frequency residual error spectrum is not limited to adopt low high frequency residual error spectrum to copy, and also can choose plural low frequency residual error spectrum E
LF(i, any other parts k) are carried out the frequency spectrum copy, and this does not exceed scope of the present invention.
● step 104, to plural high frequency residual error spectrum E
HF(i, k) carry out the time-frequency grid division: when low frequency coding nuclear uses voice class coding nuclear (not shown) to encode, the time-frequency of high-frequency signal is divided and is always adopted high time resolution (64 sampling point temporal resolutions also are low-limit frequency resolution simultaneously, shown in Fig. 5 A); When low frequency coding nuclear environment for use class coding nuclear (not shown) is encoded, the time-frequency grid division of high frequency depends on superframe length, when superframe length is 512, can only use the time-frequency grid division structure of low-limit frequency resolution, in a time grid scope, have only a BWE gain gi this moment; When superframe length is 1024 or 2048, can use more time-frequency lattice structure, in a time grid scope, may have more than a BWE gain gi.
Some typical time-frequency grid division in table 1, have been provided.
Table 1 time-frequency grid division (a, b) typical case
In the table 1 512,1024,2048 represented the length of superframe; I representative is by the time-frequency grid division, the number (it equals the number I of subframe in each superframe just) that the BWE that superframe export the most at last gains; First component a of grid division has represented the piece number that frequency domain is divided in a time domain grid scope, and the full high-frequency scope that is about to whole signal (is F in the present embodiment
S/ 4~F
S/ 2) be divided into what zones; And second component b of grid division piece number of having represented time domain to divide, be about to plural high frequency residual error spectrum E
HF(what zones i k) is divided in time.Concrete example about the time-frequency grid division will partly specifically describe in conjunction with Fig. 6 A-6C after a while at this paper.
In addition, shown in table 1 the 1st row, in the time can determining superframe length, only need the expense of 2 bits just can indicate concrete grid division mode.
●
step 105, based on plural low frequency residual error spectrum E
LF(i, part k) and plural high frequency residual error spectrum E
HF(i, at least a portion k) (the component a in choosing according to the grid in the
step 104 decide a part or all) is calculated BWE gain gi.Aforementioned calculation is based on the following fact: the part of low frequency residual signals can be used to replace high frequency residual error, and by the high frequency composite filter
Filtering obtains rebuilding high-frequency signal (hereinafter being called " high frequency analog signals of estimation " again).In fact, except differing from a gain factor, as the plural low frequency residual error spectrum E of " reconstruction high-frequency signal "
LF(i, (for example low high frequency residual error spectrum E of part k)
LFH(i, k)) can fine approximate original high-frequency signal x
HF(i, n).To rebuild gain factor between high-frequency signal and the original high-frequency signal in order calculating, to calculate earlier the energy of two signals institute corresponding grid covering time-frequency piece respectively.Last calculating energy ratio, and conversion is a unit with dB.Each superframe just can obtain I gain (g like this
0, g
1, g
2... g
I-1).
● according to a preferred embodiment of the present invention, also with execution in step 106 and step 107, the processing below in these two steps, carrying out respectively:
Step 106:, judge primary signal (that is original high-frequency signal) and copy signal (that is plural low frequency residual error spectrum E, to each time-frequency divided block
LFThe string of (i, part k)), copying signal if primary signal has string does not have string, then suitable reducing is carried out in the gain of relevant block, too raises to prevent noise level.
Step 107: continuous 4 gain coefficients to every superframe are formed a gain vector, and quantize with 7 bits.
After the processing of execution in step 101-107, finally to packing data module 18 output high frequency BWE parameters: gain gi (also may be) and ISF parameter through the adjusted gain of gain.
Hereinafter will specify BWE device 16 in conjunction with Fig. 3 B according to the embodiment of the invention.(and in conjunction with Fig. 1-2) as shown in Figure 3, BWE device 16 comprises: receiving inputted signal, through the low frequency that quantification and oscillometry filter 162A and 162B, respectively with the time-frequency conversion module 164A that is coupled through the low frequency that quantizes and oscillometry filter 162A and 162B and 164B, the frequency spectrum copy module 166A and the gain acquisition module 166B that are coupled and intercouple with time- frequency conversion module 164A and 164B respectively.
Specifically, utilize low frequency and oscillometry filter 162A and the 162B reception high frequency time-domain signal x that quantizes
H(n) and low frequency time-domain signal x
L(n) (among the figure with x
LF(i, n), x
LF(i, n) expression there is no difference in fact through low, high frequency time-domain signal after the sub-frame division), and calculate the low frequency residual error e of each subframe
L(i is n) with high frequency residual error e
H(i, n).Time- frequency conversion module 164A and 164B are used for respectively to low frequency residual error e
L(i is n) with high frequency residual error e
H(i n) carries out time-frequency conversion.Preferably, time- frequency conversion module 164A and 164B are 2 64 FFT conversion modules, and export plural low frequency residual error spectrum E
LF(i is k) with plural high frequency residual error spectrum E
HF(i, k).Frequency spectrum copy module 166A and gain acquisition module 166B receive plural low frequency residual error spectrum E respectively
LF(i is k) with plural high frequency residual error spectrum E
HF(i k), carries out the step 103-108 among Fig. 3 A then successively, obtains high frequency BWE parameter: gain gi (also may be through the adjusted gain of gain) and ISF parameter.
Hereinafter will BWE coding/decoding method 200 (not shown) according to the embodiment of the invention be described with reference to figure 4.BWE decoding end (comprising code translator 206,
FFT module 262, LPC decoder module 264) has received high frequency BWE parameter---the BWE gain of ISF parameter and vector quantization.The ISF parameter is with generating composite filter
And gain parameter is used for the low-frequency excitation signal is carried out shaping.
Particularly, the MSVQ that is based on prediction of ISF vector isf_hf_q coding use.Define 2 bit index i
1Expression first order codebook index, 7 bit index i
2Expression second level code book, then
isf_hf_q=cb1(i
1)+cb2(i
2)+mean_isf_hf+μ
isf_hf*mem_isf_hf
Cb1 (i
1) i of expression first order code book
1Individual code vector, cb2 (i
2) i of expression second level code book
2Individual code vector, mean_isf_hf represent it is the mean value of ISF vector.μ
Isf_hfThe=0.5th, predictive coefficient, mem_isf_hf are represented the memory of ISF fallout predictor, and its renewal is as follows:
Mem_isf_hf=isf_hf_q-mean_isf_hf (the mem_isf_hf initial value is 0)
Utilize the ISP parameter (being the ISF parameter) of present frame and former frame to carry out linear interpolation.Interpolation is to carry out in the ISP territory, per 64 sample subframes once, interpolation formula is as follows:
isp
subframe-i=i/nb*isp
new+(1-i/nb)*isp
old
In the following formula, nb is that the included number of sub-frames of current decoded frame is (for 256 decoded frames, nb=4).I=0 ..., nb-1 is a subframe index, isp
OldThe ISF parameter of former frame, isp
NewThe ISF parameter of present frame.At last interpolation is obtained the linear predictor coefficient that the ISP parameter is converted to each subframe.
For the BWE behind vector quantization gain, preferably by following method decoding: according to 7 bit index of the high frequency residual error gain VQ code table of the four-dimension, the residual error gain formula of decoding is as follows:
(g
0,g
1,g
2,g
3)=cb_gain_hf(idx)
Idx the code vector of cb_gain_hf (idx) expression code book cb_gain_hf.
The BWE gain conversions linear expression of final each subframe is
Obtain the high frequency residual component by gain and corresponding low frequency copy frequency spectrum product, at last by high frequency synthetic filtering output high-frequency signal.
Above-mentioned high-frequency signal and the low frequency signal that decodes (obtaining by prior art, not shown) are returned to F through rising sampling
SSample frequency, then addition just can recover the coding before the full range band signal.Further, be different from F if wish the sample frequency of output signal
S, then carry out a sample rate conversion and get final product.
For the specific embodiment of the present invention is described better, hereinafter will advance explanation in conjunction with the instantiation of Fig. 5 A-6C.
At first in Fig. 5 A-5C, divide examples for three kinds that have specifically listed under superframe 2048 points (being 1024 points after the down-sampled) situation, (1 in corresponding successively the table 1 under 2048 situations of superframe, 16), (2,8), (4,4) three kinds of dividing mode, one by one they are illustrated below:
Fig. 5 A be illustrated in the time domain grid scope (that is, 64 points) frequency domain number of partitions is 1, therefore the gain of the BWE in each time domain grid scope gi (i=0,1 ... 15) also only have 1, concrete account form is as shown in Figure 6A; In addition, the interior time domain number of partitions of superframe scope is 16.Fig. 5 B be illustrated in the time domain grid scope (that is, 128 points) frequency domain number of partitions is 2, therefore the gain of the BWE in each time domain grid scope gi (i=0,1 ... 15) have 2, concrete account form is shown in Fig. 6 B; In addition, the interior time domain number of partitions of superframe scope is 8.Fig. 5 C be illustrated in the time domain grid scope (that is, 256 points) frequency domain number of partitions is 4, therefore the gain of the BWE in each time domain grid scope gi (i=0,1 ... 15) have 4, concrete account form is shown in Fig. 6 C; In addition, the interior time domain number of partitions of superframe scope is 4.
By that analogy, those skilled in the art know the meaning of listed and other all possible time-frequency grid division mode representative in the table 1 easily.
In Fig. 6 A-6C, illustrate the method for calculating the BWE gain under above-mentioned three kinds of situations successively, below they are specifically described.
Fig. 6 A illustrates according to a preferred embodiment of the present invention, calculates the method for BWE gain under (1,16) of 2048 of superframes situation: at first will hang down high frequency residual error spectrum E
LFH(i, k) (low high frequency residual error spectrum E in the present embodiment
LFH(i k) is plural low frequency residual error spectrum E just
LF(i, k) half, i.e. the 16th to the 31st point in 32 plural low frequency spectrums) copy twice is with the details component of the high frequency full range band that obtains to estimate (promptly the 0th arrive the K-1 point in 32 plural high spectrums, K is taken as 32)
Wherein i is the numbering (seeing Fig. 5 A) of time-frequency grid, in the present embodiment, and for the high frequency full range band in each time domain grid scope distributes 1 time-frequency grid numbering; Then, allow E
HF(i, k) and
On high frequency full range band, sue for peace, and be divided by, obtain the high frequency linearity gain; At last, be decibel with the high frequency linearity gain conversions, obtain the BWE gain gi of i time-frequency grid.
Fig. 6 B illustrates according to a preferred embodiment of the present invention, calculates the method for BWE gain under (2,8) of 2048 of superframes situation: at first will hang down high frequency residual error spectrum E
LFH(i, k) copy twice is with the details component of high frequency full range band that to estimate
Wherein i is the numbering (seeing Fig. 5 B) of time-frequency grid, in the present embodiment, and for the high frequency full range band in each time domain grid scope distributes 2 time-frequency grid numberings; Then, as shown in the figure, allow E respectively
HF(2j+m, k) and
(j=0,1 ... 7) [0, K '
0-1], [K '
0, K] frequency range on summation (on the time domain need calculate two cross-talk frames, i.e. 128 points), and be divided by respectively, obtain two high frequency linearity gains, wherein K ' on each time domain grid
0Be preferably F
SThe complex spectrum numbering of/3 correspondences; At last, be decibel with the high frequency linearity gain conversions, the BWE that obtains i time domain grid gain gi (i=0,1 ... 15).
Fig. 6 C illustrates according to a preferred embodiment of the present invention, calculates the method for BWE gain under (4,4) of 2048 of superframes situation: at first will hang down high frequency residual error spectrum E
LFH(i, k) copy twice is with the details component of high frequency full range band that to estimate
Wherein i is the numbering (seeing Fig. 5 C) of time-frequency grid, in the present embodiment, and for the high frequency full range band in each time domain grid scope distributes 4 time-frequency grid numberings; Then, as shown in the figure, allow E respectively
HF(4j+m, k) and
(j=0,1 ... 3) [0, K
0-1], [K
0, K
1-1], [K
1, K
2-1], [K
2, K] frequency range on summation (on the time domain need calculate four cross-talk frames, i.e. 256 points), and be divided by respectively, obtain four high frequency linearity gains, wherein K on each time domain grid
0Be preferably F
SThe complex spectrum numbering of/4+Fs/108 correspondence, K
1Be preferably F
SThe complex spectrum numbering of/4+Fs/36 correspondence, K
2Be preferably F
SThe complex spectrum numbering of/3 correspondences; At last, be decibel with the high frequency linearity gain conversions, the BWE that obtains i time domain grid gain gi (i=0,1 ... 15).
By that analogy, those skilled in the art know the listed and pairing BWE gain calculating of other all possible time-frequency grid division mode mode in the table 1 easily.
It will be appreciated by persons skilled in the art that above-mentioned division to high frequency full range band may be equiband divide, based on the division in Bark territory or Mel territory, or other any divisions, they all within the scope of the invention.
Though described the present invention in conjunction with being considered to most realistic and optimum embodiment at present, but those skilled in the art are to be understood that and the invention is not restricted to the disclosed embodiments, on the contrary, the present invention is intended to cover various modifications and the equivalent construction that comprises within the spirit of claims and the category.Those skilled in the art can be understood that: can various deformation and/or improvement be used the present invention as being shown in specific embodiment ground, and this does not break away from the spirit or scope of the present invention of describing in broad mode.Therefore, to be considered to be descriptive but not determinate to the embodiment of this paper in all fields.