CN101471072B - High-frequency reconstruction method, encoding device and decoding module - Google Patents

Abstract

The embodiment of the invention discloses a method for rebuilding high frequency, which comprises the following steps: carrying out the filtering processing to acoustic frequency or voice signals to obtain a low-frequency sub-band and a high-frequency sub-band, confirming the frequency band copy strategy, obtaining the correlation between the low-frequency sub-band and the high-frequency sub-band according to the frequency band copy strategy, selecting the low-frequency sub-band as an optimum copy frequency band whose correlation is higher for the high-frequency sub-band, outputs high-frequency rebuilding parameter information which comprises the corresponding correlation of the selected frequency band. The invention further provides a method for rebuilding the high-frequency, which comprises the following steps: receiving the high-frequency rebuilding parameter information which comprises the corresponding correlation of the selected frequency band, the corresponding correlation of the selected frequency band is particular the corresponding correlation between the low-frequency sub-band with big correlation and the high-frequency sub-band, and at the high-frequency band, the low-frequency sub-band is copied as a high-frequency sub-band according to the high-frequency rebuilding parameter information which comprises the corresponding correlation of the selected frequency band. Correspondingly, the embodiment of the invention provides a coding module and a decoding module. The technical proposal provided by the embodiment of the invention can more accurately carry out the high-frequency rebuilding.

Description

High-frequency reconstruction method, code device and decoding device

Technical field

The present invention relates to communication technical field, be specifically related to a kind of high-frequency reconstruction method, coding module and decoder module.

Background technology

In audio frequency and voice processing technology, carrying out high-frequency reconstruction is a relatively more crucial technology.With spectral band replication (SBR) is that high frequency compression and the recovery technology of representative is the reasonable a kind of high-frequency reconstruction method of effect so far; It the waveform copy of low-frequency range to the high band part; The energy adjustment parameter harmonic adjustment parameter of having extracted when utilizing coding is again repaired the high band that duplicates, thereby reaches the purpose of high-frequency reconstruction.

The method that prior art utilizes the low frequency frequency band signals to carry out high-frequency reconstruction mainly contains two kinds, sees also following introduction:

Prior art one:

Through a digital filter bank, obtain one group of low frequency sub-band signal to the low frequency signal of audio frequency or voice; Carry out duplicating of high-frequency signal to this low frequency sub-band group as a monoblock signal again.The clone method of whole high-frequency band signal is that high-frequency band is divided into plurality of sections from low to high according to frequency, and every section bandwidth with above-mentioned monoblock low frequency signal is roughly the same; Then with monoblock low frequency sub-band group continuous compound rate each section to high-frequency band.Like this; Monoblock low frequency sub-band group can periodically be used several times in high-frequency band; Till the whole high-frequency band that need recover all is replicated completion; Concrete mode has two kinds: 1) move to corresponding high-frequency band to monoblock low frequency sub-band group, this mode can be consulted Fig. 1, is that synoptic diagram is duplicated in the integral translation of prior art one low frequency sub-band; 2) folding earlier monoblock low frequency sub-band group, promptly put upside down subband and put in order, move to corresponding high-frequency band to monoblock low frequency sub-band group again, this mode can be consulted Fig. 2, is the foldable integral translation synoptic diagram of prior art one low frequency sub-band.In the reproduction process, mode 1) and 2) may intersect use.Like this, monoblock low frequency sub-band group can periodically be used, till the whole high-frequency band that need recover all is replicated completion.

Seeing also Fig. 4, is the energy waveform figure (for comparison directly perceived, the waveform of preceding 29 subbands that only draw among the figure) of prior art original audio and each subband signal thereof.Fig. 5 is the three-dimensional plot of each sub belt energy waveform of prior art original audio.Fig. 6 is the mode 1 that adopts in the prior art one) the energy waveform figure of each subband signal of carrying out obtaining behind the high-frequency reconstruction, Fig. 7 is each corresponding sub belt energy three-dimensional plot.Fig. 8 is the mode 2 that adopts in the prior art one) energy waveform of each subband signal of carrying out obtaining behind the high-frequency reconstruction, Fig. 9 is each corresponding sub belt energy three-dimensional plot.For energy waveform figure, wherein shown in the structure of waveform all be: nethermost waveform is the original audio waveform; Label the 0th is the low frequency sub-band waveform to the 8th curve, and they will be used to duplicate high-frequency sub-band; It between the label the 8th and the 9th separatrix of high and low frequency; From all subbands that the 9th curve upwards represented, all be the scope of high-frequency reconstruction and processing.For the energy three-dimensional plot, the audio frequency parameter of describing among the figure is respectively can discharge amplitude, audio frame number (30 frame) and sub band number (29).29 subbands of the processing shown in corresponding the energy waveform figure.Wherein, the 9th is the high frequency processing section more than the subband.

Prior art two:

Through the low-pass filter group, obtain one group of low frequency sub-band to low frequency sub-band.Here, no longer as prior art one, make the low frequency sub-band group of choosing as a whole, the whole section ground continuous compound rate HFS that need recover.And utilize the subband in the low frequency sub-band group, recover the high-frequency sub-band of some Discrete Distribution respectively accordingly.

At HFS, if profuse harmonic component is arranged, the frequency of harmonic component much all is the integral multiple of its corresponding fundamental frequency so.Under the guidance of this thought; Prior art two proposes; If the sub-band serial number of some subband of HFS is natural integral multiples such as 2,3,4,5, in other words, there is the corresponding relation of multiple between some high-frequency sub-band and the low frequency sub-band; There is abundant harmonic components probably in these subbands so, need emphasis to recover.

Seeing also Fig. 3, is the discrete synoptic diagram that duplicates of prior art two low frequency sub-bands.The whole frequency of sound signal through sub-band filter can be divided 33 subbands (sub-band serial number is followed successively by 0,1,2 ..., 31,32).Wherein, the low frequency sub-band group comprise the 0th, 1,2 ..., eight subbands such as 7, the high-frequency sub-band group that needs to recover comprises the 8th, 9,10 ..., 25 subbands such as 31,32.Wherein, the low frequency sub-band group provides four continuous subbands to accomplish to duplicate at every turn as far as possible.

That at first begin is reproduction process II.Because the sequence number of the 8th, 10,12,14 4 subband in the high-frequency sub-band group all is 2 integral multiple, then from the low frequency sub-band group, choose the 4th, 5,6,7 four subband, the 8th, 10,12,14 4 high-frequency sub-band of corresponding successively recovery.

Be reproduction process III then.The sequence number of the 9th, 12,15,18,21 5 subband in the high-frequency sub-band group all is 3 integral multiple; But the 12nd subband just was replicated; Its position has also influenced the continuity of these subbands; Then from the low frequency sub-band group, choose the 3rd, 5,6,7 four subband, the 9th, 15,18,21 4 high-frequency sub-band of corresponding successively recovery.

Then be reproduction process IV.The sequence number of subbands such as the 8th, 12,16,20,24,28 in the high-frequency sub-band group all is 4 integral multiple; But 8th, 12 subbands were replicated; Then from the low frequency sub-band group, choose the 4th, 5,6,7 four subband, the 16th, 20,24,28 4 high-frequency sub-band of corresponding successively recovery.

Be reproduction process V at last.The sequence number of subbands such as the 10th, 15,20,25,30 in the high-frequency sub-band group all is 5 integral multiple; But 10th, 15,20 subbands were replicated; Then only need from the low frequency sub-band group, to choose the 6th, 7 two subband, the 25th, 30 two high-frequency sub-band of corresponding successively recovery.

Like this, the process of recovering the high-frequency sub-band of Discrete Distribution with continuous low frequency sub-band group has just been accomplished.At last,, also to choose the close with it low frequency sub-band of waveform, the high-frequency sub-band of omitting recovered, thereby accomplished duplicating of all high-frequency sub-band for the high-frequency sub-band that said method is omitted.

In research and practice process to prior art, the inventor finds that there is following problem in prior art:

No matter carry out periodically shift copy or folding duplicating to low frequency sub-band as a monoblock in the prior art by prior art one; Still carrying out frequency multiplication by prior art two duplicates; All being mechanically to recover harmonic wave, not considering the diversity and the variability of audio speech signal, is to extract successively according to sub-band serial number to duplicate when duplicating in addition; Because the waveform of low frequency sub-band and high-frequency sub-band is original just different; So the high-frequency sub-band that is replicated is compared with original high-frequency sub-band, may there be big shape differences or peak value difference, the high-frequency signal accuracy of therefore rebuilding is not too high.As observe the above-mentioned oscillogram of mentioning, contrast finds that waveform and original waveform after art methods is rebuild differ greatly; Observe the above-mentioned energy waveform figure that mentions, the result of contrast shows that art methods has lost many high-frequency harmonics after rebuilding.

Summary of the invention

The technical matters that the embodiment of the invention will solve provides a kind of high-frequency reconstruction method, coding module and decoder module, can carry out high-frequency reconstruction more accurately.

For solving the problems of the technologies described above, embodiment provided by the present invention realizes by the following technical programs:

The embodiment of the invention provides a kind of high-frequency reconstruction method, comprising: sound signal is carried out Filtering Processing obtain low frequency sub-band and high-frequency sub-band; Said sound signal is carried out specificity analysis in short-term, according to carrying out the different frequency bands of the corresponding selection as a result replication strategy of specificity analysis in short-term; Obtain the correlativity of said low frequency sub-band and high-frequency sub-band according to said definite spectral band replication strategy, select the big low frequency sub-band of correlativity for high-frequency sub-band and duplicate frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band as optimum; Said said sound signal is carried out specificity analysis in short-term, the different frequency bands of the corresponding selection as a result replication strategy of specificity analysis is specially according to carrying out in short-term: low frequency part average energy value, HFS average energy value, each sub belt energy value of low frequency part, each sub belt energy value of HFS of calculating sound signal; First weighted value of each sub belt energy value of low frequency part and low frequency part average energy value is compared; If the sub belt energy value of part subband that has low frequency part is smaller or equal to first weighted value of low frequency part average energy value; The strategy of selecting is: choose the low frequency sub-band of concentration of energy, duplicate said low frequency sub-band at the big high band of low frequency sub-band correlativity that choose and said concentration of energy; If the sub belt energy value of low frequency part subband is all greater than first weighted value of low frequency part average energy value, further second weighted value with HFS average energy value and low frequency part average energy value compares; If the HFS average energy value is smaller or equal to second weighted value of low frequency part average energy value, the strategy of selection is: choose whole low frequency sub-band, duplicate said low frequency sub-band what choose with the big high band of whole low frequency sub-band correlativity; If the HFS average energy value is greater than second weighted value of low frequency part average energy value, the strategy of selection is: high frequency is divided into a plurality of frequency bands that duplicate, duplicates frequency band for each and select the big low frequency sub-band of correlativity to duplicate.

The embodiment of the invention provides a kind of coding module, comprises the analysis filter module, specificity analysis module and frequency band are selected module in short-term; Said analysis filter module is used for that sound signal is carried out Filtering Processing and obtains low frequency sub-band and high-frequency sub-band; The said module of specificity analysis in short-term is used for said sound signal is carried out specificity analysis in short-term; Said frequency band selects module to comprise replication strategy selection module and optimum frequency band selection module; Said replication strategy is selected module, is used for the different frequency bands of the corresponding selection as a result replication strategy according to the said module analysis of specificity analysis in short-term; Said optimum frequency band is selected module; Be used for obtaining the correlativity of said low frequency sub-band and high-frequency sub-band according to said definite spectral band replication strategy; Select the big low frequency sub-band of correlativity for high-frequency sub-band and duplicate frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band as optimum; The said module of specificity analysis is in short-term carried out in short-term to said sound signal, and specificity analysis is specially: low frequency part average energy value, HFS average energy value, each sub belt energy value of low frequency part, each sub belt energy value of HFS of calculating sound signal; The said replication strategy selection module basis different frequency bands of the corresponding selection as a result replication strategy of specificity analysis module analysis in short-term is specially: first weighted value of each sub belt energy value of low frequency part and low frequency part average energy value is compared; If the sub belt energy value of part subband that has low frequency part is smaller or equal to first weighted value of low frequency part average energy value; The strategy of selecting is: choose the low frequency sub-band of concentration of energy, duplicate said low frequency sub-band at the big high band of low frequency sub-band correlativity that choose and said concentration of energy; If the sub belt energy value of low frequency part subband is all greater than first weighted value of low frequency part average energy value, further second weighted value with HFS average energy value and low frequency part average energy value compares; If the HFS average energy value is smaller or equal to second weighted value of low frequency part average energy value, the strategy of selection is: choose whole low frequency sub-band, duplicate said low frequency sub-band what choose with the big high band of whole low frequency sub-band correlativity; If the HFS average energy value is greater than second weighted value of low frequency part average energy value, the strategy of selection is: high frequency is divided into a plurality of frequency bands that duplicate, duplicates frequency band for each and select the big low frequency sub-band of correlativity to duplicate.

Technique scheme can be found out; Embodiment of the invention scheme takes into full account the correlativity of low frequency sub-band and high-frequency sub-band; Through obtain the correlativity of said low frequency sub-band and high-frequency sub-band according to the spectral band replication strategy of confirming; Select the big low frequency sub-band of correlativity for high-frequency sub-band and duplicate frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band, thereby can carry out high-frequency reconstruction according to these high-frequency reconstruction parameter informations as optimum.The energy waveform figure and the energy three-dimensional plot of each subband signal that obtains behind the high-frequency reconstruction can significantly be found; HFS and original audio signal after the embodiment of the invention is rebuild are comparatively approaching; Effect is better than prior art scheme, so embodiment of the invention technical scheme can be carried out high-frequency reconstruction more accurately.

Description of drawings

Fig. 1 is that synoptic diagram is duplicated in the integral translation of prior art one low frequency sub-band;

Fig. 2 is the foldable integral translation synoptic diagram of prior art one low frequency sub-band;

Fig. 3 is the discrete synoptic diagram that duplicates of prior art two low frequency sub-bands;

Fig. 4 is the energy waveform figure of prior art original audio and each subband signal thereof;

Fig. 5 is the energy three-dimensional plot of each sub belt energy waveform of prior art original audio;

Fig. 6 is the mode 1 that adopts in the prior art one) the energy waveform figure of each subband signal of carrying out obtaining behind the high-frequency reconstruction;

Fig. 7 is the mode 1 that adopts in the prior art one) carry out each sub belt energy three-dimensional plot of obtaining behind the high-frequency reconstruction;

Fig. 8 is the mode 2 that adopts in the prior art one) the energy waveform figure of each subband signal of carrying out obtaining behind the high-frequency reconstruction;

Fig. 9 is the mode 2 that adopts in the prior art one) carry out each sub belt energy three-dimensional plot of obtaining behind the high-frequency reconstruction;

Figure 10 is the principle and the structured flowchart of embodiment of the invention high-frequency reconstruction;

Figure 11 is an embodiment of the invention high frequency segmentation low frequency coupling replication strategy synoptic diagram;

Figure 12 is the leading frequency range high frequency coupling of an embodiment of the invention low frequency replication strategy synoptic diagram;

Figure 13 is an embodiment of the invention low frequency frequency range high frequency coupling replication strategy synoptic diagram;

Figure 14 is the replication strategy synoptic diagram of embodiment of the invention expansion;

Figure 15 (a) is the structured flowchart of embodiment of the invention coding side self-adaptation spectral band replication mode;

Figure 15 (b) is the structured flowchart of embodiment of the invention coding side fixed frequency band copy mode;

Figure 16 is the high-frequency reconstruction method process flow diagram of embodiment of the invention self-adaptation spectral band replication mode;

Figure 17 is an embodiment of the invention spectral band replication policy selection process flow diagram;

Figure 18 is that the optimum frequency band of the embodiment of the invention is selected process flow diagram;

Figure 19 is the algorithm flow chart that embodiment of the invention time-varying characteristics detect;

Figure 20 is embodiment of the invention decoding end is carried out high-frequency reconstruction according to the coding side parameter information a synoptic diagram;

Figure 21 is an embodiment of the invention decoding end high frequency maker algorithm flow chart;

Figure 22 is the energy waveform figure after embodiment of the invention method is recovered original audio signal;

Figure 23 is the energy three-dimensional plot after embodiment of the invention method is recovered original audio signal;

Figure 24 is embodiment of the invention coding module one structural representation;

Figure 25 is embodiment of the invention coding module two structural representations;

Figure 26 is an embodiment of the invention decoder module structural representation.

Embodiment

The embodiment of the invention provides a kind of high-frequency reconstruction method, can carry out high-frequency reconstruction more accurately.

Seeing also Figure 10, is the principle and the structured flowchart of embodiment of the invention high-frequency reconstruction.

Shown in figure 10, wherein the first half is to handle relevant module with high frequency in the coding side, and the latter half is to handle relevant module with high frequency in the decoding end.

At coding side; Original audio signal is the subband signal that is distributed in different frequency range through the analysis filter module converts; Comprise low frequency sub-band and high-frequency sub-band; Low frequency sub-band will and be transferred to decoding end by the core encoder coding, in addition, also will handle the parameter information that obtains instructing high-frequency reconstruction to low frequency sub-band.Specifically, at coding side, low frequency sub-band is through analyzing and detection module, and the testing result that obtains is input to frequency band and selects module, is used in reference to the analysis strategy that the pilot tone band is selected module; Frequency band selects module according to the tutorial message of analysis with detection module; Select suitable replication strategy; And to use the maximal correlation criterion be the low frequency sub-band that each or every section high-frequency sub-band are selected coupling, carries out the envelope parameters extraction, exports the parameter information of high-frequency reconstruction at last.

In decoding end, the coded message of the low frequency sub-band that the core decoder utilization receives decoding recovers the low frequency sub-band signal, duplicates high-frequency sub-band according to the parameter information from the high-frequency reconstruction of coding side again, carries out the high-frequency sub-band that the envelope adjustment obtains rebuilding then; The signal of last low frequency sub-band and high-frequency sub-band is handled audio frequency and the voice signal that recovers the full range band through synthesis filter.

Below introduce the spectral band replication strategy of the embodiment of the invention earlier:

The low-frequency range that the spectral band replication method that prior art is traditional is selected certain limit usually uses certain fixing rule to duplicate to HFS as fundamental frequency then.For example shift copy, frequency multiplication relation is duplicated and is folded and duplicates etc., and such clone method has limited the accuracy of high-frequency reconstruction.Because though high frequency and low-frequency component have certain correlativity; But translation and the folding correspondence that to guarantee this correlativity of duplicating; The low frequency sub-band that is used to duplicate possibly not be fine with the high-frequency sub-band correlativity that is replicated; Even correlativity is relatively poor, will introduce noise like this or change tonequality; The frequency multiplication clone method has utilized property harmonic period, but be not each subband all be harmonic components, correlativity is also bad, so also can introduce noise or change tonequality; The diversity of sound source in voice and the sound signal in addition causes more easily with certain fixing clone method and to duplicate misusing of frequency band.

The spectral band replication strategy that the embodiment of the invention proposes; Take into full account the correlativity of low frequency sub-band and high-frequency sub-band; The characteristic in short-term and the time-varying characteristics that also are fit to audio frequency and voice signal in addition; Have frequency band selection function flexibly, can guarantee that the frequency band that is used to duplicate and be replicated has optimum correlativity.The embodiment of the invention proposes following three kinds of spectral band replication strategies and their expanding policy thereof:

(1) high frequency segmentation low frequency coupling replication strategy (strategy 1):

This strategy is divided into a plurality of frequency bands that duplicate with the high-frequency band composition, duplicates the division of frequency band and can carry out according to diverse ways.For example duplicating frequency band can divide according to Bark (Bark scale) proportional band, and in the Bark proportional band, by different resolutions subband grouping formation is duplicated frequency band etc.Resolution is represented to duplicate the subband number that frequency band comprises, and number is few more, and resolution is high more, and number is many more, and resolution is low more.In the Bark band, divide when duplicating frequency band, resolution descends with the rising of frequency.Duplicate frequency band division and also can radio-frequency component be divided into a plurality of frequency bands that duplicate, select maximally related low-frequency range for each high band then and duplicate by identical resolution.

Shown in figure 11, be embodiment of the invention high frequency segmentation low frequency coupling replication strategy synoptic diagram.B0 is the end subband of low frequency sub-band signal, and B1 is the end subband of high frequency processing signals, and b1, b2, b3 duplicate the frequency band division border.After frequency band was duplicated in division, this strategy was respectively each and duplicates frequency band and choose maximally related low frequency sub-band and be used to duplicate, and low frequency sub-band can be reused, and has maximum correlation as long as guarantee itself and the frequency band that duplicates that is replicated.

This replication strategy has made full use of the correlativity of high frequency and low frequency; Be applicable to spectrum envelope situation more stably, low-and high-frequency all has good spectrum envelope energy, because the radio-frequency component of this moment has important audio frequency component; Especially near the high-frequency band signals of low frequency part; If high frequency duplicates the generation distortion, will introduce noise, influence tonequality, segmentation selects replication strategy to duplicate for the low-frequency range of each high band selection maximal correlation; Guaranteed that the correlativity between frequency range is corresponding, can avoid misusing the distortion that frequency band causes.

This replication strategy compared with prior art; Difference is: prior art one usefulness monoblock low frequency signal duplicates high-frequency signal continuously and repeatedly; When low-and high-frequency signal correlation difference, use the low frequency sub-band that differs greatly and duplicate high-frequency sub-band, introduce bigger distortion, and this replication strategy is a benchmark with original high-frequency sub-band signal; The low frequency sub-band of selecting to have maximum correlation with it duplicates, and can effectively keep away frequency band and misuse problem.

(2) the leading frequency range high frequency coupling of low frequency replication strategy (strategy 2):

This strategy is at first chosen the frequency range of concentration of energy in the low frequency signal as leading frequency range; Select therewith the relevant high-frequency signal section of section low frequency signal then, this high-frequency signal section with the said low frequency signal of choosing in the frequency range of concentration of energy duplicate as high-frequency sub-band; Those remaining little frequency ranges that are not replicated as yet then are respectively it and select immediate low-frequency range to duplicate for the high-frequency signal section.This method is advanced to handle to the frequency band with good harmonic characteristic earlier, selects immediate low frequency sub-band to duplicate to scattered frequency band therebetween then.Selected the best high-frequency harmonic of correlativity with low frequency before this when frequency band is selected, and then selected maximally related low frequency sub-band with non-harmonic high-frequency sub-band.

Shown in figure 12, be the leading frequency range high frequency coupling of embodiment of the invention low frequency replication strategy synoptic diagram.At first choose the low frequency signal of concentration of energy; Select the HFS of good relationship then according to selected low frequency signal; Duplicate in this part; As high frequency band signal i and j is low frequency signal select duplicate frequency band, then for removing i and the suitable low-frequency range of scattered high band selection the j is duplicated.

The leading frequency range high frequency coupling of low frequency replication strategy utilizes the signal frequency-domain harmonic characteristic, and the class fundamental frequency signal in the low frequency is chosen the high-frequency harmonic of different orders, the voice and sound signal use that are fit to have good harmonic characteristic.General aspects according to harmonic conversion; Harmonic wave can descend at the interval that HFS occurs gradually; The frequency band range that harmonic wave covers can raise gradually, can produce high frequency distortion so the harmonic wave that the traditional usefulness frequency multiplication clone method of prior art carries out duplicates, and the embodiment of the invention is used the maximal correlation decision method; Can accurately confirm the high-frequency harmonic position; And also use the maximal correlation method to choose the low frequency sub-band that is used to duplicate to the scattered subband between high-frequency harmonic, can guarantee the integrality of harmonic wave, remedy the phenomenon that high-frequency harmonic broadens.

This replication strategy compared with prior art; Difference is: when prior art adopts the frequency multiplication clone method, use the low frequency sub-band signal to duplicate with multiple to high frequency, the high-frequency harmonic of formation has comprised the harmonic wave of different low frequency signals; Destroyed the continuity of harmonic wave; And this replication strategy to the continuous displacement of class fundamental frequency signal to high-frequency harmonic, guaranteed the harmonic wave continuity, can not cause high frequency distortion.

(3) low frequency frequency range high frequency coupling replication strategy (strategy 3):

This strategy is regarded whole low-frequency range as a benchmark frequency band signals, carries out copy choice according to the Optimum Matching principle at the harmonic wave of high band then.

Shown in figure 13, be embodiment of the invention low frequency frequency range high frequency coupling replication strategy synoptic diagram.High band i is the harmonic components comparatively relevant with low frequency signal that chooses with j, in this position whole low-frequency range is duplicated, and scattered subband therebetween still duplicates band selection methods with maximal correlation, chooses low frequency sub-band and duplicates.

Such harmonic wave selects replication strategy to be applicable to that spectrum envelope steadily descends than steady and high-frequency energy; The sound signal that high-frequency energy is lower, the general high-frequency harmonic energy of such sound signal are along with the rising of order forms index decreased, because high-frequency energy is less; Can regard the mixing of harmonic wave and noise as; So can duplicate to high-frequency selectivity with whole low frequency signal, but it is extremely important when low code check, to duplicate accuracy near the high-frequency sub-band of low-frequency band, should do careful associated frequency band and select.

This replication strategy compared with prior art; Difference is: prior art use the low frequency signal monoblock continuous duplicate high-frequency signal; A whole back is selected from radio-frequency component and the high-frequency harmonic of low frequency signal maximal correlation and this replication strategy is regarded low frequency as; In this position whole low-frequency range is duplicated,, allow to have transition band between the harmonic wave like i among Figure 13 and j; Use optimum band selection methods to select suitable low-frequency band to duplicate again, can prevent the skew of harmonic wave like this as transition band.

(4) expanding policy of spectral band replication strategy (strategy 1, strategy 2 and strategy 3):

This kind expanding policy method be with before carry out high frequency and duplicate the lower high-frequency sub-band frequency range of frequency that obtains and also be used for more duplicating of high band.

Because low frequency signal possibly not cover a complete harmonic wave under the low code check situation; So enlarge the frequency band range of choice; Regard low frequency signal and the high-frequency sub-band of closing on low frequency signal on a small quantity as an integral part, and then confirm to duplicate frequency band with the method in strategy 1, strategy 2 or the strategy 3.Select the low frequency sub-band of maximal correlation to duplicate to a small amount of high-frequency sub-band that adds.When carrying out frequency band selection detection; At first concern to confirm the scope of the extending bandwidth high-frequency sub-band of copy source (when promptly the recovering as) based on coding bit rate harmonic integrality; And select the low frequency sub-band that is used to duplicate for each extending bandwidth with the maximal correlation criterion; Because extending bandwidth needs the highest reconstruction accuracy, so frequency band adopts high frequency band resolution (with single subband for duplicating frequency band) when selecting; Then extending bandwidth and low frequency sub-band altogether as copy source, adopt strategy 1, strategy 2 or strategy 3 to come to select to duplicate frequency band again for high-frequency sub-band.

Shown in figure 14, be the replication strategy synoptic diagram of embodiment of the invention expansion.After duplicating frequency band 1 and rebuilding, utilize low-frequency band and the lower band that duplicates in the frequency band 1 to be combined into continuous frequency range again, and be used to duplicate frequency band 2 and more than duplicate the signal reconstruction of frequency band.

This replication strategy is suitable for when hanging down code check, using; Because the low-frequency band of being handled by core codec when hanging down code check is shorter; Possibly not contain all fundamental frequency overtones; And the overtone of comparing Mid Frequency with the low frequency overtone more approaches high frequency overtone characteristic, so can behind the high-frequency signal that guarantees with higher resolution reconstruction lower frequency, the signal after rebuilding be used for more duplicating of high frequency band.Through this clone method can be more complete the portrayal harmonic wave, help enlarging the high-frequency reconstruction scope.

Below in detail further introduce embodiment of the invention high-frequency reconstruction method in detail.

Embodiment of the invention high-frequency reconstruction method can have two kinds, i.e. self-adaptation spectral band replication mode and fixed frequency band copy mode:

(1) self-adaptation spectral band replication mode: shown in Figure 15 (a), be the structured flowchart of embodiment of the invention coding side self-adaptation spectral band replication mode.This mode utilizes energy spectrum analysis and method of estimation to detect the characteristic of sound signal, and testing result output tutorial message instructs and selects replication strategy, and then instructs optimum frequency band to select.Because the characteristic of voice and sound signal is normally identical in certain time length, promptly accurate smooth performance there is no need to carry out again frequency band and selects, and detects so introduce time-varying characteristics, only need during greater than tolerance limit, just carry out frequency band again and select at the time-varying characteristics variable.

(2) fixed frequency band copy mode: shown in Figure 15 (b), be the structured flowchart of embodiment of the invention coding side fixed frequency band copy mode.This mode is as long as select a kind of fixing copy mode according to actual needs in advance; Promptly confirm one of spectral band replication strategy of the above-mentioned proposition of the embodiment of the invention (like strategy 1, strategy 2, strategy 3 or their expanding policy); In whole Audio Processing process, remain unchanged, and combine suitable optimum frequency band to select to realize.Because under the fixed frequency band copy mode, need not instruct the selection of spectral band replication strategy based on the result of specificity analysis module in short-term, through being set, parameter specifies replication strategy, so do not need specificity analysis module in short-term.

Need to prove that also in self-adaptation spectral band replication mode and the fixed frequency band copy mode, it not is necessary that time-varying characteristics detect.

Below the high-frequency reconstruction method of self-adaptation spectral band replication mode is introduced.

Seeing also Figure 16, is the high-frequency reconstruction method process flow diagram of embodiment of the invention self-adaptation spectral band replication mode, comprises step:

Step 1601, the subband signal that the analysis filter module is obtained carry out specificity analysis in short-term;

Step 1602, the basis result of specificity analysis are in short-term carried out the spectral band replication policy selection;

Step 1603, carry out optimum frequency band according to the spectral band replication strategy of selecting and select;

Step 1604, carry out spectral band replication according to optimum frequency band.

Below each step is specifically introduced.

Step 1601, the subband signal that the analysis filter module is obtained carry out specificity analysis in short-term;

For original audio signal, be the subband signal that is distributed in different frequency range through the analysis filter module converts, again subband signal is carried out specificity analysis in short-term.

Specificity analysis is the preliminary work in order to select suitable spectral band replication strategy to carry out in short-term.After audio frequency or voice signal carry out time-frequency conversion earlier, analyze to the energy distribution of harmonic wave situation, low frequency part and HFS then, the parameter result of analysis determines the spectral band replication strategy.

The implementation algorithm of specificity analysis is a lot of in short-term, and the embodiment of the invention is used wherein a kind of algorithm but is not limited to this.

If the low frequency sub-band sampling point is X _Low(n, l).Wherein n representes the low frequency sub-band sequence number, 1≤n＜k ₀, k ₀It is first sub-band serial number that high frequency is handled; L representes the sampling point in the subband, 0≤l＜32.If the high-frequency sub-band sampling point is X _High(k, l), wherein k represents high-frequency sub-band, k ₀≤k≤k _e, k _eIt is the end subband that high frequency is handled.

1., calculate the energy of each subband of low frequency part, be shown below:

$E\left(n\right)=\underset{l=0}{\overset{31}{\mathrm{\Σ}}}{X}_{\mathrm{Low}}^2(n,l)$

2., calculate the average energy value of whole low frequency part, be shown below:

$E_{\mathrm{Low}}=\frac{1}{k_0-1}\underset{n=1}{\overset{k_0-1}{\mathrm{\Σ}}}E\left(n\right)$

3., calculate the energy of each subband of HFS, be shown below:

$E\left(k\right)=\underset{l=0}{\overset{31}{\mathrm{\Σ}}}{X}_{\mathrm{High}}^2(k,l)$

4., calculate the average energy value of whole HFS, be shown below:

$E_{\mathrm{High}}=\frac{1}{k_e-{\mathrm{<figure-callout\; id="0"\; label="k"\; filenames="H200710305087320080415E000021.png,H200710305087320080415E000022.png"\; state="\{\{state\}\}">k</figure-callout>}}_0}\underset{k=k_0}{\overset{k_e-k_0}{\mathrm{\&Sigma;}}}E\left(k\right)$

Through aforementioned calculation, specificity analysis finishes in short-term, and analytical parameters will be applied to spectral band replication policy selection part.

Step 1602, the basis result of specificity analysis are in short-term carried out the spectral band replication policy selection;

Above-mentionedly mentioned four kinds of spectral band replication strategies,, therefrom selected a kind of strategy and duplicate according to the result of specificity analysis in short-term.After the spectral band replication strategy was confirmed, output replication strategy zone bit and replication strategy information instructed the optimum frequency band of back to select.

Seeing also Figure 17, is embodiment of the invention spectral band replication policy selection process flow diagram.

Each the low frequency sub-band energy value that obtains specificity analysis in short-term and the average energy value of low frequency part relatively, again HFS average energy value and the comparison of low frequency part average energy value.According to situation relatively, when part low frequency sub-band energy value far below average, other has the energy of partial continuous low frequency sub-band very near average or on average, then selection strategy 2; If each sub belt energy of low frequency is approaching, the low frequency part energy trace is mild continuously, and HFS average energy value and low frequency part average energy value gap are bigger, and then selection strategy 3; Otherwise, selection strategy 1.

The judgement flow process that strategy 2, strategy 3 and strategy 1 are formed is the main body of policy selection; And the spectral band replication expanding policy is as an auxiliary strategy; Mainly be in recovering high-frequency process; To the narrow situation of low frequency frequency range, expand the width of the low frequency frequency range that is used for duplicating, improve the integrality of fundamental frequency.When code rate is lower, the low frequency sub-band number is very limited, and the spectral band replication expanding policy can make the HFS can be too not limited to the selection of low frequency sub-band.It will also need the some medium-high frequency subbands and the low frequency part of spectral band replication to be bundled into together originally, formed new low frequency part, supplied most of high-frequency sub-band to select, and extracted it and duplicated parameter accordingly.Simultaneously,, can from former low frequency part, select the low frequency sub-band that matees most with them, it accomplished successively duplicate Parameter Extraction for the some medium-high frequency subbands in the new low frequency part.

The spectral band replication expanding policy is the expansion to strategy 2, strategy 3 and strategy 1.。When output extend_flag indicated in the spectral band replication policy selection flow process, then service band duplicated expanding policy.So when extend_flag exported, selected strategy 2, strategy 3 or strategy 1 can correspondingly become extended pattern strategy 2, extended pattern strategy 3 or extended pattern strategy 1.

Idiographic flow is as follows among Figure 17:

Step 1701, time-frequency conversion are accomplished, input QMF (Quadrature Mirror Filter, quadrature mirror filter) subband;

Whether the subband of step 1702, judgement input is low code rate pattern, if, get into step 1703, if not, entering step 1705;

The scope of the low frequency part that step 1703, expansion participation are duplicated is formed new low frequency part, gets into step 1704;

Step 1704, output identification extend_flag are used for the spectral band replication expanding policy, get into step 1705;

Step 1705, judge whether the low frequency sub-band that exists energy low excessively, if not, get into step 1708, if get into step 1706;

The average energy value E of energy E of each low frequency sub-band (n) and low frequency part _LowCompare, if exist the energy E (n) of certain low frequency sub-band to satisfy following formula: E (n)≤δ ₁* E _Low,

The situation that exists sub belt energy to fall suddenly in the low frequency sub-band is described, the fundamental frequency energy distribution is discontinuous, gets into step 1706, otherwise gets into step 1708.Wherein, δ ₁Span be 0＜δ ₁＜1, this span is that the corresponding waveform situation of this replication strategy is carried out observable empirical value, and value can be set as requested.

The low frequency sub-band of step 1706, the higher and continuous distribution of search energy is also confirmed selection strategy 2;

It is interval that this step is mainly sought the continuous subband of low frequency part energy distribution, as the fundamental frequency part of strategy 2.Evaluation algorithm is as follows:

If satisfy E (n)＞δ ₂* E _Low, E (n+1)＞δ ₂* E _Low..., E (n+q-1)＞δ ₂* E _Low, E (n+q)＜δ ₂* E _Low, (q>=1,1≤n＜k wherein ₀, δ ₂Span be 0＜δ ₁＜δ ₂＜1, this span is that the corresponding waveform situation of this replication strategy is carried out observable empirical value, and value can be set as requested),

Then strategy 2 is adopted in decision, record sub-band serial number n and subband interval number q;

The zone bit Flag of step 1707, output policy 2 correspondences, and output sub-band serial number n and subband interval number q; What need explanation is that if export extend_flag simultaneously, then current strategies is an extended pattern strategy 2.

Step 1708, through comparing with the low frequency energy average, judge whether the high-frequency energy average low excessively, if not, get into step 1709, if get into step 1710;

If pass through step 1705 with each low frequency sub-band energy E (n) and low frequency energy average E _LowRelatively, all satisfy following formula: E (n)＞δ ₁* E _Low, then the emphasis of analyzing is transferred to the energy relationship of low frequency and HFS.

E _HighWith λ * E _LowCompare, wherein the span of λ is 0＜λ＜1, is to the observable empirical value of waveform correlation, and value can be set as requested.

When satisfying E _High≤λ * E _Low, then strategy 3 is adopted in decision, gets into step 1710, otherwise E _High＞λ * E _LowThe time, strategy 1 is adopted in decision, gets into step 1709;

Step 1709, output policy 1 corresponding marker bit Flag; What need explanation is that if export extend_flag simultaneously, then current strategies is an extended pattern strategy 1.

Step 1710, output policy 3 corresponding marker bit Flag; What need explanation is that if export extend_flag simultaneously, then current strategies is an extended pattern strategy 3.

Step 1603, carry out optimum frequency band according to the spectral band replication strategy of selecting and select;

It is standard with the maximal correlation that optimum frequency band is selected module, searches flexibly to be used for the Optimum Matching frequency band that certain reference band is duplicated, and guarantees the correlativity of spectral band replication, makes the high-frequency signal after duplicating not need too much adjustment, just approaches original signal.

According to the replication strategy of confirming and the guidance of replication strategy information (comprising corresponding initialization frequency band table), select optimum low-and high-frequency signal corresponding relation.The optimum frequency band of spectral band replication strategy instruction is selected; It is to select low frequency signal with high-frequency signal that the decision frequency band is selected; Still select high-frequency harmonic with low frequency signal, for example at tactful 1 time, optimum frequency band is chosen as each high frequency and duplicates the low frequency signal that is used to duplicate that band signal is chosen optimum; And at tactful 2 times, optimum frequency band is selected at first to select the high-frequency harmonic that can duplicate for a type fundamental frequency signal.The estimation frequency range that initialization frequency band table instructs optimum frequency band to select, and selected frequency band range.

It is comparison low-and high-frequency signal correlation that optimum frequency band is selected, and low-and high-frequency signal envelope feature similarity degree, confirms optimum low-and high-frequency Signal Matching relation according to two parametric synthesis that obtain then.The calculated difference that the signal energy amplitude is brought during for fear of comparison correlativity and envelope characteristic; Earlier signal is carried out normalization according to initialization frequency band table scope before estimating; Focus on the similarity degree of analytic signal characteristic when selecting matched signal like this, capacity volume variance can be adjusted when signal reconstruction.

The reference band no matter optimum frequency band is selected is high-frequency signal or low frequency signal; It all is identical that optimum frequency band is selected employed algorithm; For the ease of explanation; Be example with strategy 1 below, and the single subband of replacement table maximum frequency resolution is explained the general algorithm that optimum frequency band is selected for duplicating frequency band.

Seeing also Figure 18, is that the optimum frequency band of the embodiment of the invention is selected process flow diagram, comprises step:

Step 1801, divide according to the initialization frequency band table in the spectral band replication policy information and to duplicate frequency band and alternative frequency band, by initialization frequency band table the input subband signal is carried out low-and high-frequency and divide;

Represent with nb if duplicate the frequency band number, establish the low frequency sub-band sampling point and use X _Low(n, l) expression, wherein 1≤n＜k ₀Refer to the low frequency sub-band sequence number, k ₀Be first sub-band serial number that high frequency is handled, the sampling point in 0≤l＜32 expression subbands is established the high-frequency sub-band sampling point and is used X _High(k, l) expression, wherein k ₀≤k≤k _eThe indication high-frequency sub-band, k _eIt is the end subband that high frequency is handled.

Step 1802, by the division of duplicating frequency band in the initialization frequency band table, the normalization that the low-and high-frequency signal duplicates in the frequency band length is handled, hypothesis is duplicated frequency band and is equaled the conversion subband here, normalization is handled as follows,

$X_{\mathrm{Low}}^{\&prime;}(n,l)=\frac{X_{\mathrm{Low}}(n,l)}{\max \left(X_{\mathrm{Low}}(n,l)\right)},$ $X_{\mathrm{High}}^{\&prime;}(k,l)=\frac{X_{\mathrm{High}}(k,l)}{\max \left(X_{\mathrm{High}}(k,l)\right)}$

Step 1803, calculate the related function of possible low-frequency range of each high band or high frequency band and each or low-frequency band;

Consider the sampling point drift condition, for the low-frequency band that obtains approaching most with the high frequency band waveform, calculate related function again after the low-frequency band sampling point done skew, formula is following,

$r_k^m\left[n\right]=\underset{l=1}{\overset{31}{\mathrm{\&Sigma;}}}{X}_{\mathrm{High}}(k,l)X_{\mathrm{Low}}(n,l-m),m=\mathrm{0,1,2},$ Wherein, m is a skew sampling point quantity,

The correlation function value of masterpiece sampling point skew back height frequency band.

If time-frequency conversion adopts plural variation; The subband sampling point is a complex values; Can get the real part of

and analyze, have

$r_k^m\left[n\right]=\mathrm{Re}\left[\underset{l=1}{\overset{31}{\mathrm{\&Sigma;}}}{X}_{\mathrm{High}}(k,l)X_{\mathrm{Low}}^{*}(n,l-m)\right],m=\mathrm{0,1,2}.$

With the skew related function of certain high frequency band in, for each low-frequency band selects correlation maximum

$r_k^{\mathrm{Max}}\left[n\right]=\mathrm{Max}\left(r_k^m\right[n\left]\right).$

More than calculate all high frequency band k are carried out, obtain

The maximal correlation matrix R that forms _Max[k] [n], R _Max[k] [n] write down the maximum related value of all height frequency bands.

The variation characteristic of step 1804, estimation height band envelopes calculates high frequency band and low-frequency band envelope variation property difference;

The envelope trait method of estimation is, regards the sampling point that duplicates in the frequency band length as a sample, and the second order that calculates it is with interior autocorrelation function, through the square error of the autocorrelation function of height frequency band relatively, obtains low-and high-frequency envelope trait difference then.

At first, by the second order autocorrelation function that duplicates frequency band length computation low-and high-frequency composition:

$r_k^m\left[l\right]=\mathrm{Re}\left[\underset{l=2}{\overset{32}{\mathrm{\&Sigma;}}}{X}_{\mathrm{High}}(k,l)X_{\mathrm{High}}^{*}(k,l-m)\right],m=\mathrm{0,1,2},$

$r_n^m\left[l\right]=\mathrm{Re}\left[\underset{l=2}{\overset{32}{\mathrm{\&Sigma;}}}{X}_{\mathrm{Low}}(k,l)X_{\mathrm{Low}}^{*}(k,l-m)\right],m=\mathrm{0,1,2},$ Wherein,

With

The autocorrelation function of representing high band and low-frequency range respectively, m represent auto-correlation at interval.

Then, calculate the envelope difference of low-and high-frequency intersubband:

$e(k,n)={\mathrm{\&rho;}}_1\underset{l=1}{\overset{31}{\mathrm{\&Sigma;}}}{\left(r_k^1\right[l]-r_n^1[l\left]\right)}^2+{\mathrm{\&rho;}}_2\underset{l=2}{\overset{31}{\mathrm{\&Sigma;}}}{\left(r_k^2\right[l]-r_n^2[l\left]\right)}^2,$ ρ wherein ₁And ρ ₂Be two coefficients, ρ ₁+ ρ ₂=1, (k n) represents the envelope difference of high band k and low-frequency range n to e.

Step 1805, comprehensive relatively

and e (k; N), duplicating frequency band for each selects optimum low-frequency band and duplicates;

Obviously the big more correlativity of height frequency range of having represented of the value of is good more; And e (k; N) value is more little, represents the envelope of height frequency range similar more, find optimum height frequency range matching relationship; Must form unified parameter and compare, so just like down conversion:

Wherein α and β are weight coefficients,

be that the frequency band that forms is selected coefficient.

$\mathrm{\&beta;}=\frac{\mathrm{Min}(\mathrm{Mean}\left(r_k^0\right[l\left]\right),\mathrm{Mean}\left(r_n^0\right[l\left]\right))}{\mathrm{Max}(\mathrm{Mean}\left(r_k^0\right[l\left]\right),\mathrm{Mean}\left(r_n^0\right[l\left]\right))},$ Wherein

With

It is respectively array

With

Average, α=1-β.

Step 1806, generation frequency band option table F _Table, required optimum frequency band is duplicated in indication.

To each high-frequency sub-band k select to send as an envoy to n of

value maximum; When selecting the spectral band replication expanding policy for use; Reference band is not only a low frequency sub-band, can be to detect frequency band all frequency bands before.Generate frequency band option table F _Table[k _e-k ₀+ 1], required optimum frequency band is duplicated in indication.

Step 1603 is carried out after optimum frequency band selects according to the spectral band replication strategy of selecting, can the said optimum frequency band of until data, and also can further carry out time-varying characteristics and detect, reselect frequency band according to testing result.

Below specifically introducing time-varying characteristics detects:

Audio frequency generally has identical characteristic (being accurate smooth performance) with voice signal in one section duration; So in continuous some frames, possibly can use identical high frequency replication strategy; Under the constant situation of replication strategy,, in continuous several frames, can use the identical frequency band option table that duplicates according to the time cycle property of sound signal; Needn't all do optimum frequency band selection by every frame; The frequency band option table can guarantee the interframe continuity again in case definite can continuous multiple frames the use so both can be saved calculated amount and transmission bit rate.In order to judge whether to continue to use the frequency band option table of previous frame, introduce the time-varying characteristics detection module.The effect that time-varying characteristics detect is to be used to judge whether present frame can continue to use the frequency band option table of previous frame, if detect acoustic characteristic difference between consecutive frame greater than thresholding, just refreshes the frequency band option table, reselects frequency band; Otherwise the frequency band option table remains unchanged.

The time-varying characteristics detection method is that the acoustic characteristic of estimation present frame and former frame low frequency signal changes, and specifically can adopt envelope difference comparative approach.If envelope difference is very little, then according to the correlativity of low-and high-frequency, the difference of high-frequency signal also can be very little, at this moment can continue to use the frequency band option table that former frame generates; If the envelope difference of current low frequency signal and former frame low frequency signal in marginal range, still has frequency shift (FS), if frequency shift (FS) is greater than 5% of critical band; Then need reselect optimum frequency band; Refresh the frequency band option table, because according to the tone dissonance theory, the frequency difference of two groups of overtones in the critical band that it limited is 5%～50%; Then these two groups of overtones are lacked of proper care, and can produce the appreciable difference of the sense of hearing.

Seeing also Figure 19, is the algorithm flow chart that embodiment of the invention time-varying characteristics detect, and comprises step:

The low frequency sub-band energy square error E of step 1901, calculating present frame and former frame _Error

$E_{\mathrm{Error}}=\underset{n=1}{\overset{k_0-1}{\mathrm{\&Sigma;}}}{(\frac{E\left(n\right)}{E^{\&prime;}\left(n\right)}-\frac{E(n-1)}{E^{\&prime;}(n-1)})}^2,$ Wherein E (n) represents each low frequency sub-band energy of present frame, and E ' (n) represents former frame low frequency sub-band energy,

Step 1902, judgement low frequency sub-band energy square error E _ErrorBe Thr whether less than decision threshold, if, get into step 1903, if not, get into step 1906;

The deviation frequency Δ of the low frequency frequency range that step 1903, estimated energy are concentrated _f

Choose the highest frequency band of energy or frequency range, its centre frequency is made as f _c,

F wherein _lAnd f _hRepresent the lower bound and the upper bound of highest energy frequency band respectively, the calculated rate shifted by delta _f=f _c-f ' _c

Step 1904, judgement deviation frequency Δ _fWhether less than 5% of current critical frequency bandwidth, if, get into step 1905, if not, get into step 1906;

Step 1905, continue to use the frequency band option table that former frame generates;

Step 1906, carry out optimum frequency band again and select.

Need to prove; If during the fixed frequency band copy mode; Catching up with the process of stating self-adaptation spectral band replication mode compares; Need not carry out specificity analysis in short-term, not need to carry out the spectral band replication policy selection yet according to the result of specificity analysis in short-term to the subband signal that the analysis filter module obtains, and optimum frequency band to select flow process be identical with the flow process of time-varying characteristics detection.

Step 1604, carry out spectral band replication according to optimum frequency band.

After decoding end obtains optimum frequency band, can carry out spectral band replication based on optimum frequency band.Seeing also Figure 20, is embodiment of the invention decoding end is carried out high-frequency reconstruction according to the coding side parameter information synoptic diagram.

Embodiment of the invention decoding end is compared with the prior art decoding end, and the function of most of module and cooperation relation do not change, and just " the high-frequency sub-band replication strategy " of high frequency maker module are revised.In the SBR code stream of input high frequency maker, three parameters have been added according to high-frequency reconstruction guide parameters information, i.e. " new algorithm service marking ", " the frequency band option table substitutes sign " and " frequency band option table ".

" frequency band option table " is the crucial parameter of comparison, when it writes down every frame signal and recovers high-frequency sub-band, high-frequency sub-band and low frequency sub-band duplicate corresponding relation.

" new algorithm service marking " determined when decoding, to use the new algorithm of coding side to instruct high-frequency reconstruction, also is to use standard SBR method to instruct high-frequency reconstruction.Here said new algorithm refers to the embodiment of the invention described the front and finally confirms employed algorithm in the process of high-frequency reconstruction parameter at coding side.If zone bit is " 1 ", then rebuild high frequency according to new algorithm; If zone bit is " 0 ", then rebuild high frequency according to standard SBR method.Through such setting, can be new algorithm and the compatible spare interface of standard SBR method in follow-up study.

" the frequency band option table substitutes sign ", when having determined the recovery high-frequency sub-band, how current demand signal obtained " frequency band option table ".If zone bit is " 0 ", then current demand signal directly prolongs the low-and high-frequency subband corresponding relation with the former frame signal, instructs high frequency to duplicate; If zone bit is " 1 ", then according to read in the SBR code stream refresh after " frequency band option table " parameter, accomplish high frequency and duplicate.The main effect of " the frequency band option table substitutes sign " is in order to reduce to the decoding end data quantity transmitted.Need to prove, when being 0, will not comprise " frequency band option table " parameter in the SBR code stream of transmission when " the frequency band option table substitutes sign ".

Below introduce the code stream information that decoding end receives in detail.

See also table 1, description be header file HeadFile data structure, be used for when decoding end is started working, carrying out the initialization setting.

Table 1HeaderFile data structure table

ENV_DATA is a data structure of describing every frame SBR information.Parameter " new algorithm service marking ", " frequency band option table substitute sign " and " frequency band option table " all are added in the ENV_DATA structure of description SBR information into.

Define a structure variable and be used for storing " new algorithm service marking ", " the frequency band option table substitutes sign " and " frequency band option table " three parameter informations specially, specific as follows:

The value of two parameters of " new algorithm service marking ", " frequency band option table substitute sign " is " 0 " perhaps " 1 ", so two character type variablees " flag_1 " are set, " flag_2 " comes to describe respectively " new algorithm service marking ", " the frequency band option table substitutes sign ".

What " frequency band option table " stored is to need the high-frequency sub-band of recovery and the corresponding sequence number of low frequency sub-band, is stored among the array FreTable [28].Wherein, the different coding pattern, the high-frequency sub-band number that needs to recover has nothing in common with each other.The coding mode of maximum bit rate, the high-frequency sub-band that needs to recover is 28.Along with coding bit rate reduces, need the also corresponding minimizing of high-frequency sub-band number that recovers.

The position of structure variable i ndex Vector in the ENV_DATA structure is as shown in table 2.

Table 2ENV_DATA organization definition table

Below introducing the algorithm flow of high frequency maker, see also Figure 21, is embodiment of the invention decoding end high frequency maker algorithm flow chart, comprises step:

Step 2101, reception " new algorithm service marking ", " the frequency band option table substitutes sign " and " frequency band option table ";

Step 2102, judgement " new algorithm service marking " if be 0, get into step 2103, if be 1, get into step 2104;

Step 2103, by standard SBR method decoding;

Step 2104, judgement " the frequency band option table substitutes sign " if be 0, get into step 2105, if be 1, get into step 2106;

If step 2105 zone bit is " 0 ", then current demand signal directly prolongs the low-and high-frequency subband corresponding relation with the former frame signal, instructs high frequency to duplicate;

It should be noted that every frame signal after having confirmed low-and high-frequency subband corresponding relation, can be in buffer memory the backup of the frequency band option table of present frame.If desired, the next frame signal can call the frequency band option table in the buffer memory.

If step 2106 zone bit is " 1 ", then, instruct high frequency to duplicate according to " frequency band option table " parameter that reads in the SBR code stream;

Step 2107, the preliminary high frequency of completion duplicate.

Through the high-frequency sub-band of tentatively duplicating, can get into modules such as envelope adjustment, interpolation harmonic components subsequently and handle, finally accomplish high frequency and duplicate.

Through the high-frequency reconstruction method of the embodiment of the invention, realize the reconstruction of high-frequency signal more accurately.Seeing also Figure 22, is the energy waveform figure after embodiment of the invention method is recovered original audio signal; Figure 23 is the energy three-dimensional plot after embodiment of the invention method is recovered original audio signal.Figure through these two figure and prior art compares, and can find that the high-frequency reconstruction effect of the embodiment of the invention is better than prior art.Therefore, embodiment of the invention method can utilize the information of minority low frequency sub-band just to rebuild high-frequency signal more accurately, also helps compressed audio information, can significantly improve the compression efficiency of audio frequency and speech coder, improves audio quality simultaneously; Reduce distortion and noise that bit rate audio frequency and Speech Signal Compression coding brings effectively.And to different acoustic characteristics, proposed multiple corresponding spectral band replication strategy, the high-frequency reconstruction method of adaptation can be provided, raising audio frequency and voice signal processing more flexible various audio frequency and voice signal.

Foregoing has been introduced embodiment of the invention high-frequency reconstruction method in detail, and corresponding, the embodiment of the invention provides a kind of coding module and decoder module.

Seeing also Figure 24, is embodiment of the invention coding module one structural representation.

Coding module comprises: analysis filter module 241, frequency band are selected module 242.

Analysis filter module 241 is used for that audio frequency or voice signal are carried out Filtering Processing and obtains low frequency sub-band and high-frequency sub-band.

Frequency band is selected module 242; Be used for confirming the spectral band replication strategy; Obtain the correlativity of said low frequency sub-band and high-frequency sub-band according to said definite spectral band replication strategy; Select the big low frequency sub-band of correlativity for high-frequency sub-band and duplicate frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band as optimum.

Said coding module further comprises: the specificity analysis module 243 in short-term, and said audio frequency or voice signal are carried out specificity analysis in short-term.

Said frequency band selects module 242 to comprise: replication strategy selects module 2421, optimum frequency band to select module 2422.

Replication strategy is selected module 2421, is used for the different frequency bands of the corresponding selection as a result replication strategy according to specificity analysis module 243 analyses in short-term;

Optimum frequency band is selected module 2422; Be used for obtaining the correlativity of said low frequency sub-band and high-frequency sub-band according to said definite spectral band replication strategy; Select the big low frequency sub-band of correlativity for high-frequency sub-band and duplicate frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band as optimum.

Said coding module further comprises: time-varying characteristics detection module 244 is used for audio frequency after the Filtering Processing or voice signal are carried out the time-varying characteristics detection; Accordingly, optimum frequency band selects module 2422 also to select optimum to duplicate frequency band according to the result that time-varying characteristics detection module 244 detects.

243 pairs of said audio frequency of the said module of specificity analysis in short-term or voice signal carry out in short-term, and specificity analysis is specially: low frequency part average energy value, HFS average energy value, each sub belt energy value of low frequency part, each sub belt energy value of HFS of calculating audio frequency or voice signal; Said replication strategy selects module 2421 to be specially according to the different frequency bands of the corresponding selection as a result replication strategy of specificity analysis module 243 analyses in short-term: each sub belt energy value of low frequency part and low frequency part average energy value are compared; If the sub belt energy value of part subband that has low frequency part is smaller or equal to first weighted value of low frequency part average energy value; The strategy of selecting is: choose the low frequency sub-band of concentration of energy, duplicate said low frequency sub-band at the big high band of low frequency sub-band correlativity that choose and said concentration of energy; If the sub belt energy value of low frequency part subband is all greater than first weighted value of low frequency part average energy value, further second weighted value with HFS average energy value and low frequency part average energy value compares; If the HFS average energy value is smaller or equal to second weighted value of low frequency part average energy value, the strategy of selection is: choose whole low frequency sub-band, duplicate said low frequency sub-band what choose with the big high band of whole low frequency sub-band correlativity; If the HFS average energy value is greater than second weighted value of low frequency part average energy value, the strategy of selection is: high frequency is divided into a plurality of frequency bands that duplicate, duplicates frequency band for each and select the big low frequency sub-band of correlativity to duplicate.

The strategy that said replication strategy selects module 2421 to select further comprises: the high-frequency sub-band that further will close on low frequency sub-band when duplicating with the said low frequency sub-band of choosing as copy source; Said high-frequency sub-band of closing on low frequency sub-band is chosen with the big low frequency sub-band of its correlativity and is duplicated, and so just is equivalent to above-mentioned each tactful expanding policy.

Seeing also Figure 25, is embodiment of the invention coding module two structural representations.

Coding module comprises: analysis filter module 241, frequency band are selected module 242.

Analysis filter module 241, be used to receive audio frequency or voice signal after, carry out Filtering Processing and obtain low frequency sub-band and high-frequency sub-band.

Frequency band is selected module 242; Be used for confirming the spectral band replication strategy; Obtain the correlativity of said low frequency sub-band and high-frequency sub-band according to said definite spectral band replication strategy; For high-frequency sub-band selects the big low frequency sub-band of correlativity as optimum frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band.

Said frequency band selects module 242 to comprise: replication strategy setting module 2423, optimum frequency band are selected module 2422.

Replication strategy setting module 2423 is used for confirming unique spectral band replication strategy according to the parameter that is provided with in advance.The spectral band replication strategy is the described wherein a kind of strategy of corresponding Figure 24 or its corresponding expanding policy.

Optimum frequency band is selected module 2422; Be used for obtaining the correlativity of said low frequency sub-band and high-frequency sub-band according to said definite spectral band replication strategy; For high-frequency sub-band selects the big low frequency sub-band of correlativity as optimum frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band.

Said coding module further comprises: time-varying characteristics detection module 244 is used for audio frequency after the Filtering Processing or voice signal are carried out the time-varying characteristics detection; Accordingly, optimum frequency band selects module also to select optimum to duplicate frequency band according to the result that the time-varying characteristics detection module detects.

Seeing also Figure 26, is embodiment of the invention decoder module structural representation.

Decoder module contains high frequency maker module 261, and said high frequency maker module 261 comprises: receiving element 2611, reconstruction unit 2612.

Receiving element 2611 is used to receive the high-frequency reconstruction parameter information that comprises the corresponding relation of selecting frequency band, and the corresponding relation of said selection frequency band is specially between big low frequency sub-band of correlativity and the high-frequency sub-band corresponding.

Reconstruction unit 2612 is used for low frequency sub-band being duplicated as high-frequency sub-band according to the said high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band that comprises at high band.

The said parameter information that said receiving element 2611 receives comprises that also new algorithm service marking and frequency band option table substitute sign; Said reconstruction unit 2612 is confirmed the algorithm that reproduction process is used according to said new algorithm service marking; Substitute the frequency band option table that sign confirms that reproduction process is used according to said frequency band option table, according to said definite algorithm and frequency band option table the low frequency sub-band in the said corresponding relation is duplicated as high-frequency sub-band at high band.

In sum; Embodiment of the invention scheme takes into full account the correlativity of low frequency sub-band and high-frequency sub-band; Through obtain the correlativity of said low frequency sub-band and high-frequency sub-band according to the spectral band replication strategy of confirming; Select the big low frequency sub-band of correlativity for high-frequency sub-band and duplicate frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band, thereby can carry out high-frequency reconstruction according to these high-frequency reconstruction parameter informations as optimum.The energy waveform figure and the energy three-dimensional plot of each subband signal that obtains behind the high-frequency reconstruction can significantly be found; HFS and original audio signal after the embodiment of the invention is rebuild are comparatively approaching; Effect is better than prior art scheme, so embodiment of the invention technical scheme can be carried out high-frequency reconstruction more accurately.

Further, embodiment of the invention scheme can comprise self-adaptation spectral band replication mode and fixed frequency band copy mode dual mode, has frequency band selection function flexibly.

Further, embodiment of the invention technical scheme can also increase audio frequency or the detection of voice signal time-varying characteristics, adjusts according to testing result.

More than a kind of high-frequency reconstruction method, coding module and decoder module that the embodiment of the invention provided have been carried out detailed introduction; For one of ordinary skill in the art; Thought according to the embodiment of the invention; The part that on embodiment and range of application, all can change, in sum, this description should not be construed as limitation of the present invention.

Claims (8)

1. a high-frequency reconstruction method is characterized in that, comprising:

Sound signal is carried out Filtering Processing obtain low frequency sub-band and high-frequency sub-band;

Said sound signal is carried out specificity analysis in short-term, according to carrying out the different frequency bands of the corresponding selection as a result replication strategy of specificity analysis in short-term;

Obtain the correlativity of said low frequency sub-band and high-frequency sub-band according to said definite spectral band replication strategy, select the big low frequency sub-band of correlativity for high-frequency sub-band and duplicate frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band as optimum;

Said said sound signal is carried out specificity analysis in short-term, the different frequency bands of the corresponding selection as a result replication strategy of specificity analysis is specially according to carrying out in short-term: low frequency part average energy value, HFS average energy value, each sub belt energy value of low frequency part, each sub belt energy value of HFS of calculating sound signal; First weighted value of each sub belt energy value of low frequency part and low frequency part average energy value is compared; If the sub belt energy value of part subband that has low frequency part is smaller or equal to first weighted value of low frequency part average energy value; The strategy of selecting is: choose the low frequency sub-band of concentration of energy, duplicate said low frequency sub-band at the big high band of low frequency sub-band correlativity that choose and said concentration of energy; If the sub belt energy value of low frequency part subband is all greater than first weighted value of low frequency part average energy value, further second weighted value with HFS average energy value and low frequency part average energy value compares; If the HFS average energy value is smaller or equal to second weighted value of low frequency part average energy value, the strategy of selection is: choose whole low frequency sub-band, duplicate said low frequency sub-band what choose with the big high band of whole low frequency sub-band correlativity; If the HFS average energy value is greater than second weighted value of low frequency part average energy value, the strategy of selection is: high frequency is divided into a plurality of frequency bands that duplicate, duplicates frequency band for each and select the big low frequency sub-band of correlativity to duplicate.

2. high-frequency reconstruction method according to claim 1 is characterized in that:

Said sound signal is carried out further comprising after the Filtering Processing:

Sound signal after the Filtering Processing is carried out time-varying characteristics to be detected; Accordingly,

Further combine the result who carries out the time-varying characteristics detection to select optimum to duplicate frequency band.

3. high-frequency reconstruction method according to claim 1 is characterized in that:

The strategy of said selection further comprises:

As copy source, choose with the big low frequency sub-band of its correlativity and duplicate by said high-frequency sub-band of closing on low frequency sub-band with the said low frequency sub-band of choosing for the high-frequency sub-band that further will close on low frequency sub-band when duplicating.

4. high-frequency reconstruction method according to claim 2 is characterized in that:

Saidly sound signal after the Filtering Processing carried out time-varying characteristics detect, further combine to carry out result that time-varying characteristics detect and select optimum to duplicate frequency band to be specially:

Calculate the low frequency sub-band energy square error of present frame and former frame; If said energy square error is less than decision threshold; And the deviation frequency of the low frequency sub-band of concentration of energy then carries out optimum again and duplicate the frequency band selection, otherwise the optimum of selecting before continuing to use is duplicated frequency band greater than preset value.

5. high-frequency reconstruction method according to claim 1 is characterized in that:

The spectral band replication strategy that said basis is confirmed obtains the correlativity of said low frequency sub-band and high-frequency sub-band, selects the big low frequency sub-band of correlativity to duplicate frequency band as optimum for high-frequency sub-band and is specially:

Calculate the correlation function value of high-frequency sub-band and low frequency sub-band;

Auto-correlation function value according to high-frequency sub-band and low frequency sub-band calculates the envelope difference value between high-frequency sub-band and low frequency sub-band;

According to said correlation function value and said envelope difference value is that high-frequency sub-band selects the big low frequency sub-band of correlativity to duplicate frequency band as optimum.

6. a code device is characterized in that, comprises the analysis filter module, specificity analysis module and frequency band are selected module in short-term;

Said analysis filter module is used for that audio frequency is carried out Filtering Processing and obtains low frequency sub-band and high-frequency sub-band;

The said module of specificity analysis in short-term is used for said sound signal is carried out specificity analysis in short-term;

Said frequency band selects module to comprise replication strategy selection module and optimum frequency band selection module;

Said replication strategy is selected module, is used for the different frequency bands of the corresponding selection as a result replication strategy according to the said module analysis of specificity analysis in short-term;

Said optimum frequency band is selected module; Be used for obtaining the correlation of said low frequency sub-band and high-frequency sub-band based on said definite spectral band replication strategy; Select the big low frequency sub-band of correlation for high-frequency sub-band and duplicate frequency band, and output comprises the high-frequency reconstruction parameter information of the corresponding relation of selecting frequency band as optimum;

The said module of specificity analysis is in short-term carried out in short-term to said sound signal, and specificity analysis is specially: low frequency part average energy value, HFS average energy value, each sub belt energy value of low frequency part, each sub belt energy value of HFS of calculating sound signal;

The said replication strategy selection module basis different frequency bands of the corresponding selection as a result replication strategy of specificity analysis module analysis in short-term is specially: first weighted value of each sub belt energy value of low frequency part and low frequency part average energy value is compared; If the sub belt energy value of part subband that has low frequency part is smaller or equal to first weighted value of low frequency part average energy value; The strategy of selecting is: choose the low frequency sub-band of concentration of energy, duplicate said low frequency sub-band at the big high band of low frequency sub-band correlativity that choose and said concentration of energy; If the sub belt energy value of low frequency part subband is all greater than first weighted value of low frequency part average energy value, further second weighted value with HFS average energy value and low frequency part average energy value compares; If the HFS average energy value is smaller or equal to second weighted value of low frequency part average energy value, the strategy of selection is: choose whole low frequency sub-band, duplicate said low frequency sub-band what choose with the big high band of whole low frequency sub-band correlativity; If the HFS average energy value is greater than second weighted value of low frequency part average energy value, the strategy of selection is: high frequency is divided into a plurality of frequency bands that duplicate, duplicates frequency band for each and select the big low frequency sub-band of correlativity to duplicate.

7. code device according to claim 6 is characterized in that, said code device further comprises:

The time-varying characteristics detection module is used for that the sound signal after the Filtering Processing is carried out time-varying characteristics and detects; Accordingly, optimum frequency band selects module also to select optimum to duplicate frequency band according to the result that the time-varying characteristics detection module detects.

8. code device according to claim 6 is characterized in that,

The strategy that said replication strategy selects module to select further comprises: as copy source, choose with the big low frequency sub-band of its correlativity and duplicate by said high-frequency sub-band of closing on low frequency sub-band with the said low frequency sub-band of choosing for the high-frequency sub-band that further will close on low frequency sub-band when duplicating.

Images