阅读说明：本技术 用于选择第一编码算法与第二编码算法中的一个的装置 (For selecting one device in the first encryption algorithm and the second encryption algorithm ) 是由 埃曼努埃尔·拉维利 斯特凡·多赫拉 纪尧姆·福奇斯 埃莱尼·福托普洛 克里斯蒂安·赫尔姆里希 于 2014-01-28 设计创作，主要内容包括：本发明公开了一种用于选择第一编码算法和第二编码算法中的一个的装置,该算法用于编码一音频信号的一部分,以获得该音频信号的该部分的一经编码版本,该装置包含一第一估计器,其用于在实际上并不使用该第一编码算法编码及解码该音频信号的该部分的情况下,估计该音频信号的该部分的一第一质量测量,该第一质量测量与该第一编码算法相关联。提供用于在实际上并不使用该第二编码算法编码及解码该音频信项号的该部分的情况下,估计该音频信号的该部分的一第二质量测量的一第二估计器,该第二质量测量与该第二编码算法相关联。该装置包含用于基于该第一质量测量与该第二质量测量之间的一比较,选择该第一编码算法或该第二编码算法的一控制器。(The invention discloses a kind of one devices for selecting in the first encryption algorithm and the second encryption algorithm, the algorithm is used to encode a part of an audio signal, with obtain the audio signal the part once version of code, the device includes one first estimator, in the case that it is used for actually and without using the part of the first encryption algorithm encoding and decoding audio signal, estimate that one first mass measurement of the part of the audio signal, first mass measurement are associated with first encryption algorithm.In the case that the part for actually and without using the second encryption algorithm encoding and decoding audio believing item No. is provided, estimate one second estimator of one second mass measurement of the part of the audio signal, second mass measurement is associated with second encryption algorithm.The device includes for selecting first encryption algorithm or a controller of second encryption algorithm compared with based on one between first mass measurement and second mass measurement.)

1. a kind of to select one first encryption algorithm with one first characteristic and one second coding with one second characteristic One device (10) in algorithm, which is used to encode a part of an audio signal (40), to obtain the audio signal (40) the part once version of code, which includes:

One first estimator (12) is used for actually and without using the first encryption algorithm encoding and decoding audio signal The part in the case where, estimate one first mass measurement of the part of the audio signal, first mass measurement and this One encryption algorithm is associated；

One second estimator (14) is used for actually and without using the second encryption algorithm encoding and decoding audio signal The part in the case where, estimate one second mass measurement of the part of the audio signal, second mass measurement and this Two encryption algorithms are associated；And

One controller (16), compared with being used for based on one between first mass measurement and second mass measurement, selection should First encryption algorithm or second encryption algorithm.

2. device (10) as claimed in claim 1, wherein first encryption algorithm is to be better suited for music shape and noise-like letter Number an encryption algorithm, and second algorithm is the encryption algorithm for being better suited for voice shape and transient state shape signal.

3. device (10) as claimed in claim 2, wherein first encryption algorithm is a Transform Coding Algorithm, one based on encryption algorithm MDCT (modification discrete cosine transform) or a TCX (transform coded excitation) encryption algorithm, and wherein second encryption algorithm is One CELP (Code Excited Linear Prediction) encryption algorithm or an ACELP (Algebraic Code Excited Linear Prediction) encryption algorithm.

4. wherein first estimator and second estimator are configured as such as the device (10) of any one of claims 1 to 3 A part of a weighted version based on the audio signal, estimates respective quality measurement.

5. wherein first mass measurement and second mass measurement are to be somebody's turn to do such as the device (10) of any one of claims 1 to 4 The SNR (signal-to-noise ratio) or blockiness SNR of a part of one weighted version of audio signal.

6. such as the device (10) of any one of claims 1 to 5, wherein first estimator and second estimator (12,14) It is configured as the part energy of the weighted version based on the audio signal, and is based on when encoding the signal section by respective The introduced estimation distortion of algorithm, estimate respective mass measurement, wherein first estimator and second estimator (12, 14) it is configured to, upon the part energy of a weighted version of the audio signal, determines that the estimation is distorted.

7. wherein first estimator (12) is configured as determining when quantization such as the device (10) of any one of claims 1 to 6 When the part of the audio signal, an estimation quantizer of introducing is distorted by the quantizer for first encryption algorithm, and The part energy of a weighted version based on the audio signal and estimation quantizer distortion, estimate first mass measurement.

8. device (10) as claimed in claim 7, wherein first estimator (12) is configured as estimating the portion of the audio signal The global gain divided, so that when the quantizer and entropy coder coding using first encryption algorithm, audio letter Number the part will generate a given target bit rate, wherein first estimator (12) is configured to based on estimated Global gain, determine the estimation quantizer be distorted.

9. device (10) as claimed in claim 8, wherein first estimator (12) is configured as based on estimated global gain A power, determine the estimation quantizer be distorted.

10. device (10) as claimed in claim 9, wherein the quantizer for first encryption algorithm is a uniform scalar quantization Device, and wherein first estimator (12) is configured with formula D=G*G/12 and determines that the estimation quantizer is distorted, wherein D For estimation quantizer distortion and G is estimated global gain.

Technical field

The present invention relates to audio codings, more particularly, to suitching type audio coding, wherein for the difference of audio signal Part generates coded signal using different coding algorithm.

Background technique

The suitching type audio coder of the known different coding algorithm for determining the different piece for audio signal.Substantially and Speech, suitching type audio coder are provided in two different modes (that is, algorithm, such as ACELP (Algebraic Code Excited Linear Prediction (Algebraic Code Excited Linear Prediction；)) and TCX (transform coded excitation ACELP (Transform Coded Excitation；TCX it is switched between))).

(MPEG unifies speech audio coding (Unified Speech Audio Coding to MPEG USAC；USAC)) LPD mode is based on two the different modes ACELP and TCX.ACELP provides good quality for voice shape and transient state shape signal. TCX provides good quality for music shape and noise-like signal.Encoder frame by frame determines which kind of mode used.Made by encoder Decision it is most important for codec quality.Single erroneous decision can produce a large amount of pseudomorphisms, especially in low bit rate In the case of.

For determining that using the most direct method of which kind of mode be closed loop mode selection, that is, execute both of which Complete coding/decoding is then calculated based on audio signal and encoded/decoded audio signal quasi- for the selection of both of which Then (for example, blockiness SNR), and finally it is based on selection criterion selection mode.The method generally generates stable and firm determine It is fixed.However, it also requires large amount of complex, because both of which must be run at each frame.

To reduce complexity, alternative method is open loop model selection.Open loop is selected by not executing two kinds The complete coding/decoding of mode, but be used instead and select a mode to form by low-complexity selection criterion calculated. Then, by the complexity of least complex patterns (usually TCX) subtract calculate selection criterion needed for complexity and reduce most Poor complex.Large amount of complex is usually saved, this situation makes when the worst complex of codec suffers restraints, Such method is attractive.

AMR-WB+ standard (defined in 26.290 V6.1.0 2004-12 of international standard 3GPP TS) include for In 80ms frame, the open loop model selection that is determined between all combinations of ACELP/TCX20/TCX40/TCX80.It is described In the chapters and sections 5.2.4 of 3GPP TS 26.290.It is also described in " for taking action, multimedia less complex audio coding The meeting of (Low Complex Audio Encoding for Mobile, Multimedia), VTC 2006, Makinen et al. " It discusses in file, and traces 7,739,120 B2 of US7,747,430 B2 and US of the so far author of committee paper.

The open loop model selection of analysis of the US7,747,430 B2 announcement based on long-term forecast parameter.US 7,739, 120 B2 disclose the open loop model selection based on characteristics of signals, which indicates in the respective section of audio signal The type of audio content, wherein if this selection and infeasible, be based further on statistical appraisal and carry out for respective adjacent sections Selection.

Can two key steps the open loop model selection of AMR-WB+ is described.In the first key step, to audio Signal carries out several feature calculations, and standard deviation, low frequency/high-frequency energy ratio, gross energy, the ISP of such as energy level (are led Anti- spectrum is to (immittance spectral pair；ISP)) distance, pitch lag and gain, spectral tilt.Then, using letter It is single based on threshold classifier, these features are used to make a choice between ACELP and TCX.If in the first key step TCX is selected, then the second key step is determined between the possibility combination of TCX20/TCX40/TCX80 with closed loop manner.

2012/110448 A1 of WO is disclosed for transient detection result and quality results based on audio signal, has The method maked decision between two encryption algorithms of different characteristics.Magnetic hysteresis is applied in addition, disclosing, wherein magnetic hysteresis was dependent on the past Made selection, that is, the selection to made by the more early part of audio signal.

" for taking action, multimedia less complex audio coding (Low Complex Audio Encoding for Mobile, Multimedia), VTC 2006, Makinen et al. " committee paper in, loop to AMR-WB+ and open Road model selection is put back to be compared.Subjectivity listens to test instruction open loop model selection execution and is significantly relatively worse than loop Model selection.But it also shows, open loop model selection reduces by 40% worst complex.

Summary of the invention

The object of the present invention is to provide it is a kind of allow one first encryption algorithm with there is answering for good performance and reduction The improved method to make a choice between one second encryption algorithm of polygamy.

The embodiment of the present invention provides a kind of to select one first encryption algorithm with one first characteristic and have one One device in one second encryption algorithm of the second characteristic, which is used to encode a part of an audio signal, to obtain The audio signal the part once version of code, which includes:

One first estimator is used for actually and without using the first encryption algorithm encoding and decoding audio signal The part in the case where, estimate one first mass measurement of the part of the audio signal, first mass measurement and this One encryption algorithm is associated；

One second estimator is used for actually and without using the second encryption algorithm encoding and decoding audio signal The part in the case where, estimate one second mass measurement of the part of the audio signal, second mass measurement and this Two encryption algorithms are associated；And

One controller, compared with being used for based on one between first mass measurement and second mass measurement, selection should First encryption algorithm or second encryption algorithm.

The embodiment of the present invention provides a kind of to select one first encryption algorithm with one first characteristic and have one One method in one second encryption algorithm of the second characteristic, which is used to encode a part of an audio signal, to obtain The audio signal the part once version of code, this method includes:

In the case where actually and without using the part of the first encryption algorithm encoding and decoding audio signal, estimate One first mass measurement of the part of the audio signal is counted, first mass measurement is associated with first encryption algorithm；

In the case where actually and without using the part of the second encryption algorithm encoding and decoding audio signal, estimate One second mass measurement of the part of the audio signal is counted, second mass measurement is associated with second encryption algorithm；And

Compared with based on one between first mass measurement and second mass measurement, selects first encryption algorithm or be somebody's turn to do Second encryption algorithm.

The embodiment of the present invention is based on the recognition that can be by every in the first encryption algorithm of estimation and the second encryption algorithm One mass measurement, and based on one between first mass measurement and second mass measurement compared with select the coding calculate One in method, and implement the open loop selection with the efficiency of improvement.Estimate the mass measurement, that is, actually not The encoding and decoding audio signal is to obtain the mass measurement.Therefore, the mass measurement can be obtained by the complexity of reduction. Then, it the estimation mass measurement can be used to execute and select similar model selection with a closed loop mode.

In an embodiment of the present invention, implement the blockiness SNR first by lower complexity estimation ACELP and TCX An open loop model selection.And then, it is similar in closed loop mode selection, estimates blockiness using these SNR value executes the model selection.

The embodiment of the present invention be not similar to conducted in the open loop model selection of AMR-WB+ using once Feature+classifier methods of allusion quotation.But instead, the embodiment of the present invention attempts to estimate a mass measurement of each mode, and selects Select the mode for providing best in quality.

Detailed description of the invention

The embodiment of the present invention is described in further detail with reference to attached drawing, in which:

Fig. 1 shows the signal to select the embodiment of one device in the first encryption algorithm and the second encryption algorithm Figure；

Fig. 2 shows the schematic diagrames of the embodiment of the device for coded audio signal；

Fig. 3 shows the signal to select the embodiment of one device in the first encryption algorithm and the second encryption algorithm Figure；

Fig. 4 a and Fig. 4 b may indicate SNR and blockiness SNR.

Specific embodiment

In the following description, by same reference mark with reference to similar assembly/step in different schemas.It should be noted that In In schema, has been omitted from and understand the feature (such as, signal connection and fellow) of the present invention not necessarily.

Fig. 1 shows that (such as, ACELP is calculated to select the first encryption algorithm (such as, TCX algorithm) and the second encryption algorithm Method) in one device 10, such as coded audio signal a part encoder.Device 10 includes for estimating signal First estimator 12 of the first partial mass measurement.First mass measurement is associated with the first encryption algorithm.In other words, first If the estimation of estimator 12 is encoded and is decoded using the first encryption algorithm, the first mass measurement that the part of audio signal will have, And actually and without using the part of the first encryption algorithm encoding and decoding audio signal.Device 10 includes for estimating signal section Second estimator 14 of the second mass measurement divided.Second mass measurement is associated with the second encryption algorithm.In other words, second estimates If the estimation of gauge 14 is encoded and is decoded using the second encryption algorithm, the second mass measurement that the part of audio signal will have, and Actually and without using the part of the second encryption algorithm encoding and decoding audio signal.In addition, device 10 includes to based on the One mass measurement selects the controller 16 of the first encryption algorithm or the second encryption algorithm compared between the second mass measurement. Controller may include the output 18 for indicating selected encryption algorithm.

In one embodiment, the first characteristic associated with the first encryption algorithm is well suited for music shape and noise-like letter Number, and the second encoding characteristics associated with the second encryption algorithm are well suited for voice shape and transient state shape signal.In the present invention Embodiment in, the first encryption algorithm is audio coding algorithms (such as, Transform Coding Algorithm), such as (modification is discrete remaining by MDCT String converts (modified discrete cosine transform；MDCT (transition coding swashs for)) encryption algorithm, such as TCX Encourage) encryption algorithm.Other Transform Coding Algorithms can be based on FFT transform or any other transformation or filter group.Of the invention In embodiment, the second encryption algorithm is speech coding algorithm, such as, CELP (Code Excited Linear Prediction) encryption algorithm, and such as ACELP (Algebraic Code Excited Linear Prediction) encryption algorithm.

In embodiment, mass measurement indicates perceived quality measurement.It can be calculated as the subjective quality of the first encryption algorithm The single value of estimation, and for the second encryption algorithm subjective quality estimation single value.The comparison of these two values can be based only upon Selection provides the encryption algorithm of best estimated subjective quality.This situation is different from content conducted in AMR-WB+ standard, In In AMR-WB+ standard, calculate indicate signal different characteristics many features, and then application class device to determine which kind of is selected Algorithm.

In embodiment, a part estimation based on weights audios signal (also that is, weighted version of audio signal) is respective Quality measurement.It in embodiment, can be the audio signal filtered by weighted function by weights audios signal definition, wherein weighting function Can be weighting LPC filter A (z/g), wherein A (z) be LPC filter and g for the weight between 0 and 1 (such as, 0.68).It as a result is the good measurement that can obtain perceived quality by this method.It should be noted that determining LPC filtering in pre-conditioning stage Device A (z) and weighting LPC filter A (z/g), and it is also used in two kinds of encryption algorithms.In other embodiments, weighted function It can be linear filter, FIR filter or linear prediction filter.

In embodiment, mass measurement is blockiness SNR (signal-to-noise ratio (the signal to noise in weighted signal domain ratio；SNR)).It as a result is that the blockiness SNR in weighted signal domain indicates the good measurement of perceived quality, and therefore, it can It is used as quality measurement with beneficial manner.This blockiness SNR is also used in ACELP and TCX encryption algorithm the two to estimate to compile The mass measurement of code parameter.

Another quality measurement can be the SNR in weighted signal domain.Other mass measurements can be blockiness SNR, non-weighting The SNR of the corresponding part of audio signal in (not filtered also that is, by (weighting) LPC coefficient) signal domain.Other mass measurements can It is scramble spectrum distortion or miscellaneous screening than (noise-to-mask ratio；NMR).

By and large, compare to SNR sample-by-sample original audio signal and through handling audio signal (such as, voice signal). Its object is to measure the distortion of the wave coder of regeneration input waveform.SNR may be calculated to be shown in such as Fig. 4 a, wherein X (i) and y (i) is the original sample and treated sample by i index, and N is the total number of sample.Blockiness SNR calculates shorter The SNR value of section (such as, 1ms to 10ms, such as 5ms) is averaged, and not operation whole signal.SNR may be calculated such as to scheme It is shown in 4b, wherein N and M is respectively section length and number of sections.

In an embodiment of the present invention, the part of audio signal indicates that the audio obtained by windowing audio signal is believed Number frame, and for executing the selection of appropriate encryption algorithm by multiple successive frames that windowing audio signal obtains.With In lower specification, in conjunction with audio signal, term " part " and " frame " are used in a manner of commutative.In embodiment, each frame is drawn It is divided into subframe, and the SNR (being changed into dB) by calculating each subframe, and calculates being averaged for subframe SNR as unit of dB, and estimates Count the blockiness SNR of each frame.

Therefore, in embodiment, not estimate (blockiness) between input audio signal and decoded audio signal SNR, and estimate (blockiness) SNR between weighting input audio signal and the decoded audio signal of weighting.With regard to this related (section Property) for SNR, it can refer to the 5.2.3 chapter of AMR-WB+ standard (international standard 3GPP TS 26.290V6.1.0 2004-12).

In an embodiment of the present invention, the energy of a part based on weights audios signal, and be based on passing through respective algorithm Introduced estimation distortion estimation respectively mass measurement when encoded signal portion, wherein first and second estimator is configured as taking Certainly estimation distortion is determined in the energy of weights audios signal.

In an embodiment of the present invention, determine at the part of quantization audio signal by the quantization for the first encryption algorithm The introduced estimation quantizer distortion of device, and the energy of the part based on weights audios signal and estimation quantizer distortion determine the One mass measurement.In these embodiments, the global gain of the part of audio signal can be estimated, so that when by compiling for first When the quantizer and entropy coder of code algorithm encode, the part of audio signal will generate given target bit rate, wherein based on estimating Count the quantizer distortion that global gain determines estimation.In these embodiments, estimator can be determined based on the power of estimation gain Change device distortion.When the quantizer for the first encryption algorithm is uniform scalar quantizer, the first estimator can be configured to make Determined to estimate quantizer distortion with formula D=G*G/12, wherein D is the distortion of estimation quantizer and G is estimation global gain. If the first encryption algorithm uses another quantizer, then it can determine in different ways that quantizer is distorted from global gain.

Present inventors have recognized that can be by using the feature above for being in any combination thereof, come estimate in a suitable manner will be Using the first encryption algorithm (such as, TCX algorithm) encoding and decoding audio signal part when the mass measurement that obtains (such as, Blockiness SNR).

In an embodiment of the present invention, the first mass measurement is blockiness SNR, and the corresponding son based on weights audios signal Partial energy and estimation quantizer distortion is related to each of multiple subdivisions of the part of audio signal by calculating The estimation SNR of connection estimates blockiness SNR, and by calculating SNR associated with the subdivision of the part of weights audios signal Be averaged, to obtain the estimation blockiness SNR of the part of weights audios signal.

In an embodiment of the present invention, determine when using the part of adaptive codebook coded audio signal by being used for second The introduced estimation self-adaptive code book distortion of the adaptive codebook of encryption algorithm, and the energy of the part based on weights audios signal And the distortion of estimation self-adaptive code book, estimate the second mass measurement.

It, can be based on by advance for each of multiple subdivisions of part of audio signal in these embodiments The pitch lag determined in platform converts the version of the subdivision to past weights audios signal, approximate adaptive code This；Can estimation self-adaptive codebook gain so that minimize weights audios signal part subdivision with through approximate adaptive code Error between this；And can subdivision based on the part by the scaled weights audios signal of adaptive codebook gain with Energy through the error between approximate adaptive codebook determines the distortion of estimation self-adaptive code book.

It in an embodiment of the present invention, can be by the estimation self-adaptive of each subdivision of the part of the audio signal determined Code book distortion reduction invariant, to consider the distortion reduction realized by the innovation code book in the second encryption algorithm.

In an embodiment of the present invention, the second mass measurement is blockiness SNR, and the corresponding son based on weights audios signal Partial energy and the distortion of estimation self-adaptive code book, estimate section by calculating estimation SNR associated with each subdivision Property SNR, and estimate blockiness SNR by calculating the average of SNR associated with subdivision to obtain.

In an embodiment of the present invention, it is converted based on the pitch lag by being determined in pre-conditioning stage to past weighting The version of the part of audio signal, approximate adaptive codebook；Estimation self-adaptive codebook gain, so that minimizing weights audios signal Part with through the error between approximate adaptive codebook；And based on by the scaled weights audios of adaptive codebook gain The part of signal determines the distortion of estimation self-adaptive code book with through the energy between approximate adaptive codebook.It therefore, can be less Complicated sex determination estimation self-adaptive code book distortion.

Present inventors have recognized that can be by using the feature above for being in any combination thereof, come estimate in a suitable manner will be Using the second encryption algorithm (such as, ACELP algorithm) encoding and decoding audio signal part when the mass measurement that obtains it is (all Such as, blockiness SNR).

In an embodiment of the present invention, magnetic hysteresis mechanism is for comparing estimation mass measurement.This operation can be made to more stable Ground uses the decision of which kind of algorithm.Magnetic hysteresis mechanism may depend on estimation mass measurement (such as, difference therebetween) and other parameters, Such as, about the transient state in the number of the statistics, time anchor-frame that had previously determined, frame.It, can for these related magnetic hysteresis mechanism With reference to (for example) WO 2012/110448A1.

In an embodiment of the present invention, include device 10 for the encoder of coded audio signal, compiled for executing first The platform and platform for executing the second encryption algorithm of code algorithm encode wherein depending on the selection made by controller 16 Device is configured with the first encryption algorithm or the second encryption algorithm comes the part of coded audio signal.In the embodiment of the present invention In, the system for encoding and decoding includes the encoded version for being configured as receiving the part of audio signal, and for encoding The instruction of the algorithm of the part of audio signal, and the volume of the encoded version using the part of indicated algorithm decoding audio signal Code device and decoder.

Before the embodiment that the first estimator 12 and the second estimator 14 are described in detail referring to Fig. 3, describe to compile referring to Fig. 2 The embodiment of code device 20.

Encoder 20 include the first estimator 12, the second estimator 14, controller 16, pretreatment unit 22, switch 24, It is configured as executing the first encoder platform 26 of TCX algorithm, is configured as executing the second encoder platform 28 of ACELP algorithm And output interface 30.Pretreatment unit 22 can be the part of common USAC encoder, and can be configured to output LPC coefficient, add Weigh the set of LPC coefficient, weights audios signal and pitch lag.It is calculated it should be noted that these all parameters are all used for two kinds of codings Method, that is, TCX algorithm and ACELP algorithm.Therefore, it is not necessary to determine to calculate these parameters otherwise for open loop mode.It is opening It puts and is to save complexity using the advantage for the parameter having calculated that in circuit pattern decision.

Input audio signal 40 is provided in Input Online.Input audio signal 40 is applied to the first estimator 12, pre- place Manage both unit 22 and encoder platform 26,28.Pretreatment unit 22 handles input audio signal in conventional fashion, with export LPC coefficient and weighted LPC coefficients 42, and by 42 filtering audio signals 40 of weighted LPC coefficients to obtain weights audios signal 44. Pretreatment unit 22 exports the set 48 of weighted LPC coefficients 42, weights audios signal 44 and pitch lag.Such as it is familiar with this skill Patient is understood, weighted LPC coefficients 42 and the segmentation of weights audios signal 44 can be turned to frame or subframe.It can be by a suitable manner Windowing audio signal and be segmented.

In an embodiment of the present invention, quantified LPC coefficient or quantified weighted LPC coefficients can be used.Therefore, Ying Li Solution, term " LPC coefficient " are also intended to cover " quantified LPC coefficient ", and term " weighted LPC coefficients " is also intended to cover " weighting Quantized coefficient ".In this regard, it is notable that the TCX algorithm of USAC is using quantified weighted LPC coefficients with moulding MCDT frequency spectrum.

First estimator 12 receives audio signal 40, weighted LPC coefficients 42 and weights audios signal 44, based on above-mentioned each Person estimates the first quality measurement 46, and the first mass measurement is exported to controller 16.Second estimator 16 receives weights audios The set 48 of signal 44 and pitch lag estimates the second mass measurement 50 based on both above-mentioned, and the second mass measurement 50 is defeated Out to controller 16.As known to those who familiarize themselves with the technology, calculates and add in previous block (also that is, pretreatment unit 22) The set 48 of LPC coefficient 42, weights audios signal 44 and pitch lag is weighed, and therefore, above-mentioned each can be used without cost.

The comparison that controller is measured based on institute's quality of reception, the selection for making TCX algorithm or ACELP algorithm determine.Such as Referred to above to show, controller can use magnetic hysteresis mechanism when determining using which kind of algorithm.By means of defeated by 16 institute of controller in Fig. 2 The switch 24 that control signal 52 out controls schematically shows that the first encoder platform 26 of selection or second encoder are flat Platform 28.Control signal 52 indicates the first encoder platform 26 to be used or second encoder platform 28.Based on control signal 52, it is schematically indicated by the arrow 54 in Fig. 2, and include at least LPC coefficient, weighted LPC coefficients, audio signal, weighting Audio signal, the set of pitch lag required signal be applied to the first encoder platform 26 or second encoder platform 28.Selected encoder platform application is associated encryption algorithm, and indicates 56 or 58 outputs to output interface 30 for encoded. Output interface 30 can be configured to output coded audio signal, may include encoded expression 56 or 58, LPC coefficient or weighting LPC coefficient, the parameter for selected encryption algorithm and the information about selected encryption algorithm (and other data).

The specific embodiment for estimating first and second mass measurement is described referring now to Fig. 3, wherein first and second matter Measurement is the blockiness SNR in weighted signal domain.Fig. 3 shows in the form of step-by-step showing the flow chart respectively estimated One estimator 12 and the second estimator 14 and its functionality.

The estimation of TCX blockiness SNR

First (TCX) estimator connects audio signal 40 (input signal), weighted LPC coefficients 42 and weights audios signal 44 It receives as input.

In step 100, windowing is carried out to audio signal 40.It can be opened a window by the low overlapping sine-window of 10ms Mouthful.When past frame is ACELP, block size can increase 5ms, can be rectangle on the left of window, and can input from through windowing Signal removes responding through windowing zero pulse for ACELP composite filter.This situation is similar to content conducted in TCX algorithm. The frame for indicating the audio signal 40 of a part of audio signal is exported from step 100.

In a step 102, as MDCT (modification discrete cosine transform) transformation through windowing audio signal (also that is, obtained by Frame).At step 104, frequency spectrum moulding is executed and with weighted LPC coefficients moulding MDCT frequency spectrum.

In step 106, estimate global gain G, so that when being encoded by entropy coder (for example, arithmetic encoder), The Weighted spectral quantified by gain G will be generated to the R that sets the goal.Using term " global gain ", this is because a gain is Determine for entire frame.

Now explain the example of the implementation of global gain estimation.It is calculated it should be noted that the estimation of this global gain is suitable for TCX coding Method uses the embodiment of the scalar quantizer with arithmetic encoder.Assume that there is arithmetic encoder in MPEG USAC standard This scalar quantizer.

Initialization

Firstly, the variable by the initialization of following each for gain estimation:

1. en [i] is set as=9.0+10.0*log10 (c [4*i+0]+c [4*i+1]+c [4*i+2]+c [4*i+ 3]),

Wherein 0≤i < L/4, c [] are to the vector of the coefficient quantified, and L is the length of c [].

It is=fac by offset setting 2. fac is set as=128, and be by goal-setting=any value (for example, 1000)

Iteration

Then, by the following onblock executing of operation NITER times (for example, here, NITER=10).

1.fac=fac/2

2. offset=offset-fac

3.ener=0

4. being proceeded as follows for every i (wherein 0≤i < L/4):

If en [i]-deviates > 3.0, ener=ener+en [i]-offset

If deviate 5. ener > target=deviate+fac

The result of iteration is deviant.After iteration, global gain is estimated as G=10^ (offset/20).

Depending on used quantizer and entropy coder, estimate that the ad hoc fashion of global gain can change.In The scalar quantizer with arithmetic encoder is assumed in MPEG USAC standard.Different quantizers can be used in other TCX methods, And those who familiarize themselves with the technology should be understood that how to be directed to these different quantizers and estimate global gain.For example, AMR-WB+ is marked Grant leave of absence fixed use RE8 lattice quantizer.For this quantizer, the estimation of global gain can be estimated for such as 3GPP TS 26.290 Described in 5.3.5.7 chapter on page 34 of V6.1.0 2004-12, wherein it is assumed that fixed target bit rate.

It is after estimating global gain in step 106, distortion estimation occurs in step 108.More particularly, based on estimating Count the distortion of global gain half quantification device.In this embodiment it is assumed that using uniform scalar quantizer.Therefore, pass through single public affairs Formula D=G*G/12 determines quantizer distortion, and wherein D indicates determined quantizer distortion and G indicates estimation global gain.This feelings The high-speed that condition corresponds to uniform scalar quantizer distortion is approximate.

It is distorted based on determined quantizer, executes blockiness SNR in step 110 and calculate.It will be in each subframe of frame SNR is calculated as the weights audios signal energy in subframe and the ratio of distortion D (it is assumed that constant).For example, four are divided a frame into A continuous subframes (referring to fig. 4).Then, blockiness SNR is being averaged for the SNR of four subframes, and can be indicated with dB.

The method allows the first blockiness of acquisition to estimate when actually using TCX algorithm coding and decoding theme frame SNR, however not necessarily actually encoding and decoding audio signal, and therefore, this method has the complexity largely reduced and subtracts Few calculating time.

The estimation of ACELP blockiness SNR

Second estimator 14 receives the weights audios signal 44 having calculated that and the collection of pitch lag in pretreatment unit 22 Close 48.

As shown in step 112, in each subframe, by simply using weights audios signal and pitch lag T And approximate adaptive codebook.Pass through following formula approximation adaptive codebook

Xw (n-T), n=0 ..., N

Wherein xw is weights audios signal, and T is the pitch lag of corresponding subframe, and N is subframe lengths.Therefore, by making It is converted to the version of past subframe and approximate adaptive codebook with by T.Therefore, in an embodiment of the present invention, with extremely simple Mode approximation adaptive codebook.

In step 114, determine the adaptive codebook gain of each subframe.More particularly, in each subframe, estimation Codebook gain G, so that it minimizes the error between weights audios signal and approximated adaptive codebook.It can be by simply comparing Difference between two signals of more each sample simultaneously finds gain and carries out this operation so that these differences and it is minimum.

In step 116, determine the adaptive codebook distortion of each subframe.In each subframe, by adaptive codebook institute The distortion D of introducing is only by the error between the scaled weights audios signal of gain G and approximated adaptive codebook Energy.

The distortion that can be determined in set-up procedure 116 in optional step 118, to consider to innovate code book.It can will be used for The distortion of the innovation code book of ACELP algorithm is simply estimated as constant value.It is simply false in described embodiment of the invention Surely innovation code book will be distorted D and reduce invariant.It therefore, can be in step 118 by each subframe obtained in step 116 It is distorted the multiplication by constants factor, the invariant (such as, 0.055) of such as 0 to 1 rank.

In the step 120, the calculating of blockiness SNR occurs.In each subframe, SNR is calculated as weights audios signal Energy and the ratio for being distorted D.Then, blockiness SNR is the average value of the SNR of four subframes, and can be indicated with dB.

The method allows to estimate when actually using ACELP algorithm coding and decoding theme frame by the 2nd SNR of acquisition, However not necessarily actually encoding and decoding audio signal, and therefore, this method has the complexity and reduction largely reduced Calculate the time.

First and second estimator 12 and 14 will the estimation output of blockiness SNR 46,50 to controller 16, and controller 16 Based on estimation blockiness SNR 46,50, the decision which kind of algorithm is used to the associated section to audio signal is made.Controller Magnetic hysteresis mechanism can be used, optionally to make more stable decision.For example, can be made by slightly different tuner parameters With the mechanism for the magnetic hysteresis mechanism being identical in loop decision.This magnetic hysteresis mechanism can calculated value " dsnr ", which may depend on Blockiness SNR (such as, difference therebetween) and other parameters are estimated, such as about statistics, the time anchor-frame previously determined Transient state in number and frame.

In the case where no magnetic hysteresis mechanism, controller can select encryption algorithm by higher estimation SNR, and also even Two estimation SNR be higher than first estimation SNR, then select ACELP, and if first estimation SNR be higher than second estimation SNR, select TCX.In the case where there is magnetic hysteresis mechanism, controller can select encryption algorithm according to rule of making decision, and wherein acelp_snr is Second estimation SNR and tcx_snr are the first estimation SNR:

If acelp_snr+dsnr > tcx_snr, ACELP is selected, TCX is otherwise selected.

Therefore, the embodiment of the present invention permission estimates blockiness SNR in a manner of simple and is accurate and selects appropriate coding Algorithm.

In embodiments above, the average of the SNR of the respective subframe by calculating estimation estimates blockiness SNR.In In alternate embodiment, the SNR of entire frame can be estimated without dividing a frame into subframe.

When compared with loop selection, the embodiment of the present invention, which allows to largely reduce, calculates the time, this is because saving Slightly several steps required in loop selection.

Therefore, a large amount of steps and calculating time related to this can be saved by inventive method, while being allowed again good Ground executes the selection of appropriate encryption algorithm.

Although in the described in the text some aspects up and down of device, it will be clear that the description of corresponding method is also indicated in this respect, Wherein block or device correspond to the feature of method and step or method and step.Similarly, described in the context of method and step Aspect also indicate the description of corresponding block or project or the feature of corresponding intrument.

It can be by being configured as or sequencing, computer, one or more processors, one in order to provide described function Multi-microprocessor, field programmable gate array (FPGA), special application integrated circuit (ASIC) and so on or its group It closes to implement the embodiment of device described herein and its feature.

Method can be executed by (or use) hardware device (for example, microprocessor, programmable computer or electronic circuit) It is some or all of in step.In some embodiments, can thus device execute in most important method and step some or More persons.

Depending on certain implementations requirement, the embodiment of the present invention can be with hardware or software implementation.It can be used and store thereon Electronically readable controls signal, and cooperate (or can cooperate) with programmable computer system, so that executing the non-temporary of respective method When property storage media (such as, digital storage medium (for example, floppy disk, DVD, Blu-Ray, CD, ROM, PROM and EPROM, EEPROM or flash memory)) execute implementation.Therefore, digital storage medium can be computer-readable.

According to some embodiments of the present invention comprising with electronically readable control signal data medium, can with can journey Sequence computer system cooperation, so that executing one in method described herein.

By and large, it is the computer program product with program code that the embodiment of the present invention is implementable, works as computer When program product is run on computer, program code is operatively enabled to execute one in this method.Program code can (example As) be stored in machine-readable carrier.

Other embodiments include to be stored in machine-readable carrier, for executing one in method described herein Computer program.

In other words, therefore, the embodiment of inventive method is with for holding when computer program is run on computer The computer program of one program code in row method described herein.

Therefore, another embodiment of inventive method is comprising recording thereon, for executing side described herein The data medium (or digital storage medium or computer-readable media) of one computer program in method.Data medium, number Word storage media or record media are usually tangible and/or non-transient.

Therefore, another embodiment of the method for the present invention is to indicate for executing one in method described herein The data flow or signal sequence of computer program.Data flow or signal sequence can (for example) be configured as connecting via data communication It connects (for example, via internet) and transmits.

Another embodiment includes to be configured as or sequencing is to execute one processing in method described herein Component, for example, computer or programmable logical device.

Another embodiment includes to be equipped with thereon for executing one computer journey in method described herein The computer of sequence.

According to another embodiment of the present invention comprising being configured as to be used to execute one in method described herein A computer program transmission (for example, electronically or optical mode) is to the device or system of receiver.Receiver can (example As) it is computer, mobile device, memory devices or fellow.Device or system can be (for example) comprising for by computer program It is sent to the file server of receiver.

In some embodiments, programmable logical device (for example, field programmable gate array) can be used for executing this paper Described in method functionality in it is some or all of.In some embodiments, field programmable gate array can be with micro- place Device cooperation is managed, to execute one in method described herein.By and large, it is preferably executed by any hardware device This method.

In some embodiments, item once is provided:

1. a kind of to select one first encryption algorithm with one first characteristic and one second with one second characteristic One device (10) in encryption algorithm, which is used to encode a part of an audio signal (40), to obtain the audio The part of signal (40) once version of code, which includes:

One first estimator (12) is used for actually and without using the first encryption algorithm encoding and decoding audio In the case where the part of signal, estimate one first mass measurement of the part of the audio signal, first mass measurement with First encryption algorithm is associated；

One second estimator (14) is used for actually and without using the second encryption algorithm encoding and decoding audio In the case where the part of signal, estimate one second mass measurement of the part of the audio signal, second mass measurement with Second encryption algorithm is associated；And

One controller (16), compared with being used for based on one between first mass measurement and second mass measurement, choosing Select first encryption algorithm or second encryption algorithm.

2. wherein first encryption algorithm is to be better suited for music shape and noise-like signal such as the device (10) of item 1 An encryption algorithm, and second algorithm is the encryption algorithm for being better suited for voice shape and transient state shape signal.

3. wherein first encryption algorithm is a Transform Coding Algorithm, one based on encryption algorithm such as the device (10) of item 2 MDCT (modification discrete cosine transform) or a TCX (transform coded excitation) encryption algorithm, and wherein second encryption algorithm is one CELP (Code Excited Linear Prediction) encryption algorithm or an ACELP (Algebraic Code Excited Linear Prediction) encryption algorithm.

4. wherein first estimator and second estimator are configured as base such as the device (10) of any one of item 1 to 3 In a part of a weighted version of the audio signal, respective quality measurement is estimated.

5. wherein first mass measurement and second mass measurement are the sound such as the device (10) of any one of item 1 to 4 The SNR (signal-to-noise ratio) or blockiness SNR of a part of one weighted version of frequency signal.

6. such as the device (10) of any one of item 1 to 5, wherein first estimator and the second estimator (12, the 14) quilt It is configured to the part energy of a weighted version of the audio signal, and is based on when encoding the signal section by respective The introduced estimation distortion of algorithm, estimates respective mass measurement, wherein first estimator and second estimator (12,14) It is configured to, upon the part energy of a weighted version of the audio signal, determines that the estimation is distorted.

7. wherein first estimator (12) is configured as determining when quantization should such as the device (10) of any one of item 1 to 6 When the part of audio signal, an estimation quantizer of introducing is distorted by the quantizer for first encryption algorithm, and base It is distorted in the part energy of a weighted version of the audio signal and the estimation quantizer, estimates first mass measurement.

8. device (10) as claimed in claim 7, wherein first estimator (12) is configured as estimating the audio signal One global gain of the part, so that when the quantizer and entropy coder coding using first encryption algorithm, the sound The part of frequency signal will generate a given target bit rate, and wherein first estimator (12) is configured to based on institute The global gain of estimation determines that the estimation quantizer is distorted.

9. wherein first estimator (12) is configured as based on estimated global gain such as the device (10) of item 8 One power determines that the estimation quantizer is distorted.

10. such as the device (10) of item 9, wherein the quantizer for first encryption algorithm is a uniform scalar quantization Device, and wherein first estimator (12) is configured with formula D=G*G/12 and determines that the estimation quantizer is distorted, wherein D For estimation quantizer distortion and G is estimated global gain.

11. wherein first mass measurement is one of weights audios signal such as the device (10) of any one of item 7 to 10 The blockiness SNR divided, and wherein first estimator (12) is configured as based on the corresponding subdivision of weights audios signal An energy and estimated quantizer distortion, by calculate in multiple subdivisions of the part of the weights audios signal Each SNR estimated by associated one estimates blockiness SNR, and by calculating the portion with the weights audios signal The one of the associated SNR of the subdivision divided is average, to obtain the estimated section of the part of the weights audios signal Property SNR.

12. wherein second estimator (14) is configured as determining when use such as the device (10) of any one of item 1 to 11 When the part to encode the audio signal of one adaptive codebook, the adaptive codebook for second encryption algorithm will be introduced The distortion of one estimation self-adaptive code book, and wherein second estimator (14) is configured as the weighted version based on the audio signal A part an energy and the estimation self-adaptive code book distortion, estimate second mass measurement.

13. such as the device (10) of item 12, wherein for each in multiple subdivisions of the part of the audio signal A, which is configured as: based on being transformed by the pitch lag determined in a pre-conditioning stage One version of the subdivision of the weights audios signal gone, the approximate adaptive codebook；Estimate an adaptive codebook gain, makes One obtained between subdivision and the approximated adaptive codebook of the part of the weights audios signal minimizes the error；And Based on scaled by the adaptive codebook gain, subdivision of the part of the weights audios signal and approximated The energy of an error between adaptive codebook determines that the estimation self-adaptive code book is distorted.

14. wherein second estimator (14) is configured to the audio signal such as the device (10) of item 13 One invariant of estimation self-adaptive code book distortion reduction that each subdivision of the part determines.

15. wherein second mass measurement is the part of the weights audios signal such as the device (10) of item 13 or 14 One blockiness SNR, and wherein second estimator (14) is configured as the corresponding subdivision based on the weights audios signal The energy and estimation self-adaptive code book distortion, estimate that SNR estimates this by calculating associated with each subdivision one Blockiness SNR, and one by calculating the SNR associated with the subdivision is average, to obtain being somebody's turn to do for the weights audios signal Partial estimation blockiness SNR.

16. wherein second estimator (14) is configured as such as the device (10) of item 12: based on by a pre-conditioning stage The pitch lag that is determined and the version for being transformed into the part of the past weights audios signal, the approximate adaptive code This；An adaptive codebook gain is estimated, so that between the part of the weights audios signal and the approximated adaptive codebook One minimize the error；And it based on the part by the scaled weights audios signal of the adaptive codebook gain and is somebody's turn to do The energy of an error between approximated adaptive codebook determines that the estimation self-adaptive code book is distorted.

17. wherein the controller (16) is configured as comparing the estimation matter such as the device (10) of any one of item 1 to 16 A magnetic hysteresis is utilized when measurement.

18. a kind of device (20) for encoding a part of an audio signal, comprising such as any one of item 1 to 17 Device (10), one first encoder platform (26) for executing the first encryption algorithm, and for executing the second encryption algorithm A second encoder platform (28), wherein for coding the device (20) be configured to, upon by the controller (16) make Selection out encodes the part of the audio signal using first encryption algorithm or second encryption algorithm.

19. a kind of system for encoding and decoding, the system include such as item 18 for a device (20) for coding and One decoder, the decoder are configured as receiving the encoded version of the part of audio signal and for encoding the audio signal One instruction of the algorithm of the part, and decode using indicated algorithm the encoded version of the part of the audio signal.

20. a kind of to select one first encryption algorithm with one first characteristic and one second with one second characteristic One method in encryption algorithm, the algorithm are used to encode a part of an audio signal to obtain the portion of the audio signal Point once version of code, this method includes:

In the case where actually and without using the part of the first encryption algorithm encoding and decoding audio signal, estimate One first mass measurement of the part of the audio signal is counted, first mass measurement is associated with first encryption algorithm；

In the case where actually and without using the part of the second encryption algorithm encoding and decoding audio signal, estimate One second mass measurement of the part of the audio signal is counted, second mass measurement is associated with second encryption algorithm；With And

Compared with based on one between first mass measurement and second mass measurement, selects first encryption algorithm or be somebody's turn to do Second encryption algorithm.

21. wherein first encryption algorithm is to be better suited for music shape and noise-like signal such as the method for item 20 One encryption algorithm, and second algorithm is the encryption algorithm for being better suited for voice shape and transient state shape signal.

22. wherein first encryption algorithm is a Transform Coding Algorithm, one based on encryption algorithm such as the method for item 21 MDCT (modification discrete cosine transform) or a TCX (transform coded excitation) encryption algorithm, and wherein second encryption algorithm is one CELP (Code Excited Linear Prediction) encryption algorithm or an ACELP (Algebraic Code Excited Linear Prediction) encryption algorithm.

23. such as the method for any one of item 20 to 22, wherein a part of the weighted version based on the audio signal, estimates Count first quality measurement and second quality measurement.

24. wherein first mass measurement and second mass measurement are the sound such as the method for any one of item 20 to 23 The SNR (signal-to-noise ratio) or blockiness SNR of a part of one weighted version of frequency signal.

25. such as the method for any one of item 20 to 24, it includes: a part of the weighted version based on the audio signal Energy, and based on when encoding the signal section by respective algorithm it is introduced one estimation be distorted, to estimate respective matter Measurement；And the energy of a part of the weighted version depending on the audio signal, determine that (108, the 116) estimation is lost Very.

26. such as the method for item 20 or 25, it includes: (108) are determined when quantifying the part of the audio signal, for being somebody's turn to do One estimation quantizer of introducing is distorted by one quantizer of the first encryption algorithm；And the weighted version based on the audio signal One energy of a part and estimation quantizer distortion determine the mass measurement.

27. such as the method for item 26, it includes: a global gain of the part of estimation (106) audio signal, so that working as When being encoded by a quantizer for first encryption algorithm and an entropy coder, the part of the audio signal will generate one Given target bit rate；And the distortion of (108) the estimation quantizer is determined based on estimated global gain.

28., it includes the power based on estimated global gain, determining that the estimation quantizer loses such as the method for item 27 Very.

29. such as the method for item 28, wherein the quantizer is a uniform scalar quantizer, wherein using formula D=G*G/12 Determine that the estimation quantizer is distorted, wherein D is estimation quantizer distortion and G is estimated global gain.

30. wherein first mass measurement is one of the weights audios signal such as the method for any one of item 26 to 29 The blockiness SNR through LPC filtered version divided, and this method includes: based on the corresponding subdivision of weights audios signal One energy and the estimation quantizer distortion, by calculate with it is each in multiple subdivisions of the part of the weights audios signal SNR estimated by a associated one estimates first blockiness SNR；And by calculating the portion with the weights audios signal The one of the associated SNR of the subdivision divided is average, to obtain the estimated section of the part of the weights audios signal Property SNR.

31. such as the method for any one of item 20 to 30, it includes: determine to work as using an adaptive codebook to encode the audio When the part of signal, the adaptive codebook for second encryption algorithm will introduce estimation self-adaptive code book distortion (116)；And an energy of a part of the weighted version based on the audio signal and the estimation self-adaptive code book are distorted, estimation Second mass measurement.

32. such as the method for item 31, it includes: for each of multiple subdivisions of the part of the audio signal, Based on the subdivision for being transformed into the past weights audios signal by the pitch lag determined in a pre-conditioning stage A version, approximate (112) adaptive codebook；Estimate (114) one adaptive codebook gains, so that the weights audios signal One between the subdivision of the part and the approximated adaptive codebook minimizes the error；And based on by the adaptive codebook Gain is scaled, one between the subdivision and approximated adaptive codebook of the part of the weights audios signal The energy of error determines the distortion of (116) the estimation self-adaptive code book.

33. it includes the estimations for determining each subdivision of the part of the audio signal certainly such as the method for item 32 Adapt to (118) one invariant of code book distortion reduction.

34. wherein second mass measurement is an area of the part of the weights audios signal such as the method for item 32 or 33 Section property SNR, and this method includes: energy and the estimation self-adaptive of the corresponding subdivision based on the weights audios signal Code book distortion estimates blockiness SNR by calculating an estimation SNR associated with each subdivision；And by calculate with The one of the associated SNR of the subdivision is average, to obtain the estimation blockiness SNR of the part of the weights audios signal.

35. such as the method for item 31, it includes: based on being transformed by the pitch lag determined in a pre-conditioning stage One version of the part of the past weights audios signal, the approximate adaptive codebook；Estimate an adaptive codebook gain, makes One obtained between part and the approximated adaptive codebook of the weights audios signal minimizes the error；And it is based on by this certainly Adapt to the mistake between part and the approximated adaptive codebook of the scaled weights audios signal of codebook gain The energy of difference determines that the estimation self-adaptive code book is distorted.

36. it includes utilize a magnetic hysteresis when comparing the estimation mass measurement such as the method for any one of item 20 to 35.

37. one kind has when running on a computer, for executing one of the method such as any one of item 20 to 36 The computer program of program code.

Embodiments described above is merely illustrative the principle of the present invention.It should be understood that those who familiarize themselves with the technology will be aobvious and easy See to configuration described herein and the modification and variation of details.Therefore, it is intended only to be limited to next apply for a patent model The scope enclosed, and be not only restricted to describe and explain presented specific detail by embodiment herein.