CN101740033B - Audio coding method and audio coder - Google Patents

Abstract

The invention discloses an audio coding method and an audio coder method. The embodiment of the invention also provides a corresponding audio coder. As the feature that the frequency resolution characteristics of linear prediction (LP) and winding linear prediction (WLP) are very close to the critical band and masking characteristic in human auditory characteristic is utilized in the technical scheme of the invention, a psychoacoustics module is built, the masking threshold is obtained, and coding is carried out on audio signals according to the obtained masking threshold, thereby decreasing the complexity of building the psychoacoustic model, being realized easily, decreasing the hardware implementation cost of the psychoacoustic model, and lowering power consumption of hardware.

Description

A kind of audio coding method and audio coder

Technical field

The present invention relates to the audio encoding and decoding technique field, be specifically related to a kind of audio coding method and audio coder.

Background technology

In the audio coding technology, lose genuine audio coding technology and can obtain higher ratio of compression usually, but, need the degree of coding distortion in the control audio coding techniques in order to obtain good audio quality.Psychoacoustic model is a kind of mathematical model that generally is used to control the coding distortion degree.The mathematical model of the psychoacoustic model reflection human auditory perception characteristic that to be people on research human auditory system basis abstract comes out, it has reflected perception and the screening ability of human auditory system to audio frequency and noise.Parameter in the psychoacoustic model that specifically uses in the audio coding technology is generally masking threshold, this parameter be scrambler receive signal each frequency place on frequency domain be subjected to value that other all frequency components shelter and, this parameter is the bar curve on frequency domain.The frequency component that is in this curve below can not be felt that then this frequency component can be used zero bits of encoded by people's ear; On the other hand, shelter curve as if guaranteeing that quantizing noise is lower than when selecting quantization step, also do not discovered, so the big more frequency component quantization step of masking threshold can be big more by people's ear.Therefore with the foundation of masking threshold as quantization encoding, the sound quality after just can guaranteeing to compress.So, by psychoacoustic model, in to coding audio signal, can remove effectively and be included in the original audio and the incoherent signal content of human auditory, thereby can when obtaining high compression ratio, guarantee the quality of sound signal.

Referring to shown in Figure 1, be the application of psychoacoustic model in the audio coding technology.Wherein, this audio coding technology is the sensing audio encoding technology.As shown in Figure 1a, in audio coder, an input audio signal part enters the time frequency analysis module, and scrambler carries out the frequency domain parameter that conversion process obtains audio frequency to sound signal in this module; Some enters psychoacoustic model input audio signal, scrambler processes input signal and obtains masking threshold in this module, masking threshold is inputed to the Bit Allocation in Discrete module, and the Bit Allocation in Discrete module of scrambler is obtained bit distribution information to perceptual coding according to masking threshold; Quantification and coding module quantize and compressed encoding the frequency domain parameter from the output of time frequency analysis module according to the bit distribution information that obtains; Close the road module and will close the road processing with the coded message of coding module with from the bit distribution information Bit Allocation in Discrete module, that transmit as side information, form coded bit stream output from quantizing.

In audio decoder, referring to Fig. 1 b, shunt module is implemented to handle along separate routes to the coded bit stream that receives, and obtains coded message and Bit Allocation in Discrete side information respectively; Decoding and inverse quantization module are decoded and are carried out inverse quantization and handle according to obtaining coded message and Bit Allocation in Discrete side information, thereby obtain the frequency domain parameter of reconstruct; Last time-frequency synthesis module carries out inverse transformation with the reconstructed frequency domain parameter to be handled, and obtains the audio frequency time-domain signal output of reconstruct.

Scrambler carries out in the process to the coding of voice signal according to the masking threshold that obtains in the prior art, for obtaining the psychoacoustic model that masking threshold is set up, need carry out very complicated calculating, and be difficult for realizing, hardware device is required height, and consumed power is big.

Summary of the invention

The embodiment of the invention provides a kind of audio coding method and related device, and the technical scheme that the embodiment of the invention provides can reduce the complexity of setting up psychoacoustic model, but can arrive the technique effect similar to prior art, and promptly accuracy is similar.

The embodiment of the invention provides a kind of audio coding method, and this method comprises:

Receive time-domain audio signal;

Described sound signal is sampled;

Sound signal after the sampling is carried out linear prediction;

According to the result of described linear prediction, obtain the amplitude-frequency response of linear prediction filter;

To the sound signal linear prediction of curling after the sampling;

According to described curling linear prediction result, obtain the amplitude-frequency response of the linear prediction filter that curls;

According to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter, obtain the part and shelter curve;

According to the characteristic of definitely sheltering curve that the described part that obtains is sheltered curve and preset, obtain the overall situation and shelter curve;

According to the information that the described overall situation of obtaining is sheltered curve and the critical band that presets, obtain overall masking threshold;

According to the described overall masking threshold that obtains, the time-domain audio signal of described reception is encoded;

Wherein said according to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter, obtain the part and shelter curve and specifically comprise:

According to the control information of presetting, respectively the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter are carried out frequency compensation;

According to result after the described frequency compensation and the control information of presetting, obtain the part and shelter curve.

The embodiment of the invention also provides a kind of audio coder, comprising:

Receiving element is used to receive time-domain audio signal;

Sampling unit is used for described sound signal is sampled;

Linear prediction unit is used for the sound signal after the sampling is carried out linear prediction;

Obtain linear prediction filter amplitude-frequency response unit, be used for result, obtain the amplitude-frequency response of linear prediction filter according to described linear prediction;

The linear prediction unit of curling is used for the linear prediction of curling of the sound signal after the sampling;

Obtain the linear prediction filter amplitude-frequency response unit that curls, be used for, obtain the amplitude-frequency response of the linear prediction filter that curls according to described curling linear prediction result;

Obtain the part and shelter curved unit, be used for obtaining the part and sheltering curve according to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter;

Obtain the overall situation and shelter curved unit, be used for sheltering curve and the characteristic of definitely sheltering curve that presets, obtain the overall situation and shelter curve according to the described part that obtains;

Obtain the masking threshold unit, be used for sheltering the information of curve and the critical band that presets, obtain overall masking threshold according to the described overall situation of obtaining;

Audio coding unit is used for according to the overall masking threshold that obtains the time-domain audio signal of described reception being encoded;

The wherein said part that obtains is sheltered curved unit and is specifically comprised:

Frequency compensation unit is used for respectively the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter being carried out frequency compensation according to the control information of presetting;

Calculate the part and shelter curved unit, be used for obtaining the part and sheltering curve according to result after the described frequency compensation and the control information of presetting.

The embodiment of the invention also provides a kind of audio frequency watermark flush mounting, it is characterized in that, comprising:

Receiving element is used to receive time-domain audio signal;

Sampling unit is used for described sound signal is sampled;

Linear prediction unit is used for the sound signal after the sampling is carried out linear prediction;

Obtain linear prediction filter amplitude-frequency response unit, be used for result, obtain the amplitude-frequency response of linear prediction linear prediction filter according to described linear prediction;

The linear prediction unit of curling is used for the linear prediction of curling of the sound signal after the sampling;

Obtain the linear prediction filter amplitude-frequency response unit that curls, be used for, obtain the amplitude-frequency response of the linear prediction filter that curls according to described curling linear prediction result;

Obtain the part and shelter curved unit, be used for obtaining the part and sheltering curve according to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter;

Obtain the overall situation and shelter curved unit, be used for sheltering curve and the characteristic of definitely sheltering curve that presets, obtain the overall situation and shelter curve according to the described part that obtains;

Obtain the masking threshold unit, be used for sheltering the information of curve and the critical band that presets, obtain overall masking threshold according to the described overall situation of obtaining;

Watermark embeds the unit, is used for the overall masking threshold that obtains according to described, and watermark encoder is embedded in the input audio signal;

The wherein said part that obtains is sheltered curved unit and is specifically comprised:

Frequency compensation unit is used for respectively the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter being carried out frequency compensation according to the control information of presetting;

Calculate the part and shelter curved unit, be used for obtaining the part and sheltering curve according to result after the described frequency compensation and the control information of presetting.

The frequency discrimination characteristic that the embodiment of the invention is utilized linear prediction LP and curling linear prediction WLP is very near the characteristics of critical band in human auditory's characteristic and masking characteristics, the psychoacoustic model of setting up, obtain masking threshold, carry out the method for audio coding according to the masking threshold that obtains, reduced the complexity of setting up psychoacoustic model, be easy to realize, reduced psychoacoustic model, reduced the power consumption of hardware at hard-wired cost.

Description of drawings

Fig. 1 a is the composition simplified schematic diagram of prior art sound intermediate frequency scrambler;

Fig. 1 b is the composition simplified schematic diagram of prior art sound intermediate frequency demoder;

Fig. 2 is a kind of method flow simplified schematic diagram of setting up psychoacoustic model that the embodiment of the invention one provides;

Fig. 3 is a computing simplified schematic diagram of obtaining related function in the embodiment of the invention one in the curling linear prediction;

Fig. 4 a is the sound signal amplitude-frequency response figure of input coding device;

Fig. 4 b is wave filter amplitude-frequency response C _LP(f), WLP wave filter amplitude-frequency response C _WLP(f) and the power spectrum density synoptic diagram of sound signal;

Fig. 5 is that the part is sheltered curve and definitely sheltered curve map;

Fig. 6 a is the trajectory diagram of absolute masking threshold at log-domain;

Fig. 6 b is the trajectory diagram of absolute masking threshold at linear domain;

Fig. 7 is that the overall situation is sheltered curve map;

Fig. 8 a is a broken line type overall situation masking threshold synoptic diagram;

Fig. 8 b is a notch cuttype overall situation masking threshold synoptic diagram;

Fig. 9 is a kind of audio coding method flow process simplified schematic diagram that the embodiment of the invention two provides;

Figure 10 is that the logic of a kind of audio coder of providing of the embodiment of the invention three is formed synoptic diagram.

Embodiment

The embodiment of the invention provides a kind of audio coding method, and the embodiment of the invention also provides corresponding audio coder.Below be elaborated respectively.

Embodiment one

Present embodiment provides a kind of audio coding method, this method is to utilize the linear prediction (WLP that curls, Warped Linear Prediction) and conventional linear prediction (LP, Linear Prediction) the frequency resolution characteristic of Fen Xiing very near the characteristic of critical band in human auditory's characteristic and masking characteristics, is finally obtained masking threshold.Referring to shown in Figure 2, this method comprises:

Step 1: scrambler receives time-domain audio signal;

The time-domain audio signal that scrambler receives can be the mixed information of the various voice signals that can hear of voice signal, sound signal or various people's ear, and the frequency span of this sound signal can be heard frequency range (being that 0Hz is to 24000Hz) for people's ear usually.The sound signal that scrambler receives is the form of frame normally, and the length of a frame is generally between 5 milliseconds to 30 milliseconds.

Step 2: scrambler is sampled to the sound signal that receives, the sound signal x (n) after obtaining sampling;

In the present embodiment, the sample frequency that scrambler adopts the time-domain audio signal that receives usually in the step 2 comprises among 48kHz, 44.1kHz, 32kHz, 16kHz or the 8kHz each.

Step 3: scrambler carries out linear prediction LP to the sound signal after sampling;

Wherein, scrambler has got access to preferable linear filter coefficient to the result who carries out linear prediction of sound signal in the step 3.It will be appreciated that linear prediction LP utilizes in the past several sampled signals to predict current sampled signal, promptly normally utilize the linear combination of N sampled signal before the current time to estimate the time-domain audio signal x (n) that imports, n=1,2 ..., L, wherein, L is the length of frame.Be formulated as follows:

$\hat{x}\left(n\right)=\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_{k}x(n-k)---\left(1\right).$

Wherein, a _k, k=1,2 ..., N is the LP filter coefficient, N is the linear prediction exponent number.At a _k, k=1,2 ..., N is under the known situation, the system function of LP wave filter is then for as can be known, shown in the following formula:

$\hat{X}\left(z\right)=\left[\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_k{z}^{-k}\right]X\left(z\right)---\left(2\right).$

Therefore, filter coefficient a _kIt is the key that realizes the LP bank of filters.The concrete operations of obtaining preferable linear filter coefficient in step 3 comprise:

Steps A 1: obtain the autocorrelation function r (l) of input signal x (n),

Steps A 2: predict the outcome with linear prediction filter according to the sound signal x (n) after the original signal sampling

Between the minimum principle of difference (being predicated error e), obtain the canonical equation of LP;

Wherein, steps A 1 is formulated as: LP filter coefficient a _kOptimum value be $e=E\left[{|x\left(n\right)-\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_{k}x(n-k)|}^2\right]$ Minimum promptly only needs for k=1, and 2 ..., N makes Thus, can obtain the canonical equation of LP wave filter, expression formula is as follows:

$r\left(l\right)=-\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_{k}r(l-k)---\left(3\right).$

Steps A 3: according to the autocorrelation function r (l) that obtains in the steps A 1, utilize Paul levinson-Doby LevinsonDurbin algorithm solution formula (3), obtain LP filter coefficient a _k, this LP filter coefficient a _kBe preferable LP filter coefficient.

Step 4: scrambler obtains the amplitude-frequency response C of LP wave filter according to the result of linear prediction _LP(f);

Wherein, the result that scrambler carries out linear prediction according to the sound signal after the sampling of obtaining in the step 3 obtains amplitude-frequency response C _LP(f) specifically can be the concluding formula:

$A^{-1}\left(z\right)=\frac{G_{\mathrm{LP}}}{1-\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_k{z}^{-k}}---\left(4\right).$

Wherein, G _LPIt is the energy of input signal.With the A that obtains ^-1(z) use curve representation, be amplitude-frequency response C _LP(f).

Step 5: scrambler is to the back sound signal that obtains in the step 2 the sampling linear prediction WLP that curls;

Wherein, it will be appreciated that the linear prediction WLP that curls is the principle according to linear prediction LP, and the Forecasting Methodology of a kind of more approaching and human hearing characteristic that sums up.Scrambler is to get access to preferable filter coefficient according to the sound signal sampling point that obtains in the step 2 linear prediction result of curling.

Wherein, the linear prediction WLP that adopt to curl utilizes the linear combination of N sampled signal before the current time to estimate the time-domain audio signal x (n) that imports, n=1, and 2 ..., L, the employing formulate is as follows:

$\hat{X}\left(z\right)=\left[\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_{k}D{\left(z\right)}^k\right]X\left(z\right)---\left(5\right).$

Wherein, D (z) is the system function of all-pass filter, and D (z) is equivalent to the Postponement module in the WLP wave filter, and this module has similar effect to Postponement module among the LP.It is constant that the all-pass filter that uses among the WLP has amplitude response, the characteristic of the mapping situation of phase response decision frequency domain, and the date expression of its amplitude-frequency response is as follows:

$\hat{w}=w+2\arctan \left(\frac{\mathrm{\λ}\sin \left(w\right)}{1-\mathrm{\λ}\cos \left(w\right)}\right)---\left(6\right).$

Wherein, in order to guarantee the WLP wave filter to the approaching as much as possible human auditory properties of the mapping situation of frequency, parameter lambda need meet some requirements, and this condition is the following formula of reference specifically:

${\mathrm{\λ}}_{f_s}\≈1.0674{\left(\frac{2}{\mathrm{\π}}\arctan \left(0.06583f_s\right)\right)}^{1/2}-0.1916---\left(7\right).$

Wherein, f _sIt is the sample frequency of input signal.

By above explanation to all-pass filter, can obtain the system function D (z) of this all-pass filter, be formulated as follows:

$D\left(z\right)=\frac{z^{-1}-\mathrm{\λ}}{1-\mathrm{\λ}{z}^{-1}}---\left(8\right).$

The expression formula of its impulse response function is as follows:

Wherein, λ is a filter parameter.Need to prove that also this system function D (z) is preset in this scrambler according to above explanation, when scrambler can obtain better WLP filter coefficient a by D (z) when carrying out linear prediction _K2

To the middle actual input of D (z) and WLP wave filter and the relation (as formula (5)) between the prediction output, the expression formula of output on time domain of the K rank all-pass filter of this WLP wave filter is as follows by above:

d _k[x(n)]≡h(n)*h(n)*...*h(n)*x(n)(10)。

Wherein, h (n) is the impulse response of D (z), and " * " represents convolution algorithm, d _k[x (n)] is the output of k rank all-pass filter.

Therefore, in the step 5 the back sound signal linear prediction WLP concrete operations of curling of sampling are comprised:

Step B1: obtain each rank output of WLP wave filter and the related function r (k) between the input signal, k=0,1 ..., N-1, N and K are the exponent numbers of WLP wave filter, the exponent number of WLP wave filter is 10 usually.

Wherein, curl among the step B1 that related function can use auto-correlation network structure shown in Figure 3 to obtain in the linear prediction.Wherein, x (n) represents input audio signal, and D (z) is the single order all-pass filter; Input audio signal x (n) obtains the output signal d of each cascade all-pass filter respectively by the processing of the single order all-pass filter D (z) of cascade _k[x (n)], k=0,1 ... N-1.Then, add the output signal d that calculates input audio signal x (n) and each cascade rank all-pass filter D (z) with computing by taking advantage of _k[x (n)], k=0,1 ... the related function of N-1, promptly obtain r (k), k=0,1 .., N-1.

Step B2: similar to steps A 1, predict the outcome with linear filter according to the sound signal x (n) after the original signal sampling

Between the minimum principle of difference (being predicated error e), obtain the WLP canonical equation;

Wherein, steps A 1 usefulness formula is illustrated as: according to the principle of the predicated error mean square value minimum of WLP, that is:

$e=E\left[{|x\left(n\right)-\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_k{d}_k[x\left(n\right)\left]\right|}^2\right]---\left(11\right).$

Wherein, be minimum in order to make e, then as can be known

Therefore, can obtain the WLP canonical equation, its expression formula is as follows:

$E\left[d_j\right[x\left(n\right)\left]d_0\right[x\left(n\right)\left]\right]-\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_{k}E\left[d_k\right[x\left(n\right)\left]d_j\right[x\left(n\right)\left]\right]=0,j=\mathrm{0,1,2},...,N-1$

(12)。

Step B3: according to obtaining the canonical equation that obtains among related function r (k) and the step B2 among the step B1, utilize Levinson Durbin algorithm to find the solution canonical equation, obtain WLP filter coefficient a _K2

Step 6: scrambler obtains the amplitude-frequency response C of WLP wave filter according to linear prediction WLP that the sound signal after adopting is curled _WLP(f);

Wherein, scrambler obtains amplitude-frequency response C according to the sound signal after the sampling is carried out linear prediction in the step 6 _WLP(f) specifically can be according to the concluding formula that has been found that in the prior art:

$A^{-1}\left(z\right)=\frac{G_{\mathrm{WLP}}}{1-\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{a}_{k2}D{\left(z\right)}^k}---\left(13\right).$

Wherein, a _K2Be the coefficient of WLP wave filter, D (z) is the transition function of single order all-pass filter, G _WLPEnergy for input signal.Fig. 4 b has provided the amplitude-frequency response C of a WLP wave filter _WLP(f) example.

Above step 3 to step 6 is respectively the method for operating that adopts linear prediction and curling linear prediction, by above prediction, can obtain amplitude-frequency response C respectively _LP(f) and amplitude-frequency response C _WLP(f).Referring to shown in Figure 4, wherein, it is that a frame sampling speed is 48kHz that Fig. 4 a shows, length is the sound signal of 512 points; Fig. 4 b has shown the LP wave filter amplitude-frequency response C of the sound signal correspondence shown in Fig. 4 a _LP(f), WLP wave filter amplitude-frequency response C _WLP(f) and the power spectrum density of sound signal.As seen from the figure, the wave filter that adopts linear prediction to obtain has high frequency characteristics preferably, and the wave filter that adopts the linear prediction WLP that curls to obtain has low frequency characteristic preferably.

Step 7: according to the amplitude-frequency response C of the LP wave filter that obtains in the step 4 _LP(f) and WLP wave filter amplitude-frequency response C _WLP(f), obtain the part and shelter curve;

Wherein, basis is obtained C in the step 7 _LP(f) and C _WLP(f), obtaining the concrete grammar of sheltering curve in the part can be according to following formula:

C _p(f)＝C _LP’(f)+C _WLP’(f)＝C _LP(f)K _LP(f，C _qb)+C _WLP(f)K _WLP(f，C _qb)+B _XLP(C _qb)(dB)

(14)

Wherein, C _p(f) shelter curve for the part, K _LP(f, C _Qb) and K _WLP(f, C _Qb) be respectively C _LP(f) and C _WLP(f) frequency compensation function.Wherein, control information C _QbBe optionally, relevant with audio coding quality settings value or code rate setting value, so K _LP(f, C _Qb), K _WLP(f, C _Qb) also relevant with setting tonequality or code rate requirement, but K _LP(f, C _Qb), K _WLP(f, C _Qb) different; K _LP(f, C _Qb) be mainly used in and strengthen C _LP(f) low frequency characteristic, K _WLP(f, C _Qb) be mainly used in and strengthen C _WLP(f) high frequency characteristics can be obtained by empirical value in actual applications.B _XLP(C _Qb) be the relative deviation coefficient, its fundamental purpose is to be used for according to audio coding quality settings value or the local amplitude of sheltering curve of the whole adjustment of code rate setting value; Set when higher B such as audio coding quality settings value or code rate _XLP(C _Qb) can reduce the part and shelter profile amplitude, and set when low B when audio coding quality settings value or code rate _XLP(C _Qb) can improve the part and shelter profile amplitude.K _LP(f, C _Qb), K _{WLP (}F, C _Qb) and B _XLP(C _Qb) can obtain by empirical value in actual applications.Fig. 5 has provided the local synoptic diagram of sheltering curve, K among the figure _LP(f, C _Qb)=K _WLP(f, C _Qb)=0.5, B _XLP(C _Qb)=0.0.Promptly get C _LP(f) and C _WLP(f) mean value on each frequency is sheltered curve as the part.

By above explanation, obtain the concrete grammar of sheltering curve in the part as can be seen and can be: earlier to the amplitude-frequency response C of the LP wave filter that obtains step 7 _LP(f) and WLP wave filter amplitude-frequency response C _WLP(f) carry out frequency compensation respectively, promptly obtain C respectively _LP(f) K _LP(f, C _Qb) and C _WLP(f) K _WLP(f, C _Qb); According to frequency compensated result and the information that presets, obtain the part and shelter curve.Wherein, the information that presets can be according to audio coding quality settings value or the local information of sheltering the amplitude of curve of the whole adjustment of code rate setting value, as the relative deviation coefficient B _XLP(C _Qb).

Step 8:, obtain the overall situation and shelter curve C according to obtaining the characteristic of definitely sheltering curve that the part is sheltered curve and preset in the step 7 _g(f);

Wherein, absolute masking threshold (Absolute Threshold) is illustrated in that tone signal can be perceived by the human ear energy needed under the noise-free environment, represents with sound pressure level dB usually.Absolute masking threshold also with frequency dependence, its amplitude can obtain by following expression is approximate:

$T_{\mathrm{AT}}\left(f\right)=3.64{(f/1000)}^{-0.8}-6.5e^{-0.6{(f/1000-3.3)}^2}+{10}^{-3}{(f/1000)}^4,\left(\mathrm{dB}\right)---\left(15\right)$

Wherein, T _AT(f) be absolute masking threshold, it is the function of frequency, referring to absolute masking threshold shown in Fig. 6 a at the track of log-domain; Shown in Fig. 6 b, be the track of absolute masking threshold at linear domain.Though two kinds of track representation differences, essence is identical.

Be appreciated that by above explanation, if the energy of sound signal be lower than the part shelter curve and definitely shelter curve each, then this sound signal is not then discovered by people's ear, therefore, to shelter curve be that part on each frequency is sheltered curve and definitely sheltered the curve that maximal value is linked to be in the curve to the overall situation.As follows with equation expression:

C _g(f)＝max{C _p(f)，γ(C _qb)T _AT(f)}(dB) (20)

Wherein, 1.0＜=γ (C _Qb)＜=0.0 is absolute masking threshold matching factor, and it is the control information C that requires with reflection tonequality or code rate _QbBe correlated with.Be used for the part and shelter curve C _p(f) with absolute masking threshold T _ATThe matching treatment of sound pressure level (f).Wherein, C _QbBe optionally to be preset at parameter in the scrambler in embodiments of the present invention always.If without C _Qbγ (C then _Qb) value is 1.The overall situation of correspondence shown in Figure 7 is sheltered curve C _g(f).Shelter curve and absolute masking threshold with reference to part shown in Figure 5 simultaneously, shelter curve C thereby be more readily understood the overall situation that shows among Fig. 7 _g(f).

Step 9:, obtain overall masking threshold according to the information that the overall situation of obtaining in the step 8 is sheltered curve and the critical band that presets.

Wherein, what need to prove critical band (Critical Band) reflection is the another kind of auditory properties of people's ear, i.e. frequency analysis ability.People's ear can characterize with the bandpass filter of a series of high superposed the analysis ability of frequency, the amplitude-frequency response of bandpass filter is asymmetric and nonlinear, its frequency span is to increase along with the raising of frequency, different frequency in the same critical band, people's ear has apperceive characteristic much at one.Critical band is exactly the characteristic with a frequency function quantitative description sense of hearing bandpass filter.In Bark Bark territory, the frequency span of a critical band is generally a Bark Bark, and following formula can be with critical band by the conversion of linear frequency domain to the Bark territory:

$z\left(f\right)=13\arctan \left(0.00076f\right)+3.5\arctan \left[{\left(\frac{f}{7500}\right)}^2\right],\left(\mathrm{Bark}\right)---\left(21\right)$

Can obtain the approximate value of the frequency span of critical band by following formula:

BW _c(f)＝25+75[1+1.4(f/1000) ²] ^0.69(Hz) (22)

The critical band that also need to prove people's ear can be by experiment or other existing experience obtain, be preset in the scrambler.

Shelter the information of curve and the critical band that presets in step 9 according to the overall situation of obtaining, the concrete manner of execution of obtaining overall masking threshold can be: obtain interior critical band end points of each Bark and frequency band mid point and shelter curve C in the overall situation _g(f) amplitude value couples together the overall masking threshold T that form according to frequency order with straight line with described all amplitude values of obtaining _PSY(f).Shown in Fig. 8 a, this overall situation masking threshold T _PSY(f) be the form of broken line, can be with should overall situation masking threshold T _PSY(f) be called broken line type overall situation masking threshold.

Shelter the information of curve and the critical band that presets in step 9 according to the overall situation of obtaining, the concrete manner of execution of obtaining overall masking threshold also can be: shelter curve C by obtaining the overall situation in the critical band on each Bark _g(f) the amplitude value in the minimum value, each critical band is that the overall situation is sheltered curve C in this frequency band _g(f) minimum value in, the amplitude-frequency response that obtains are overall masking threshold T _PSY(f).Shown in Fig. 8 b, this overall situation masking threshold T _PSY(f) be notch cuttype in the drawings.This overall situation masking threshold T _PSY(f) can become notch cuttype overall situation masking threshold T _PSY(f).Because people's ear is better than resolving ability at high frequency to voice or sound signal in the resolving ability of low frequency, therefore, as can be seen, the critical band width is narrower at the low frequency place from the figure, in the height critical band wider width that occurs frequently.In fact, broken line type overall situation masking threshold also has these characteristics.

Also it will be appreciated that, shelter the information of curve and the critical band that presets in the step 9 according to the overall situation of obtaining, the concrete manner of execution of obtaining overall masking threshold is confined to two kinds described above incessantly, more than two kinds of methods be the fairly simple approximate overall masking threshold T that obtains _PSY(f), can also on each critical band, choose a plurality of frequencies certainly, obtain overall masking threshold T _PSY(f).

Above step 1 has realized a kind of method of setting up psychoacoustic model to the explanation of step 9, and scrambler can get access to overall masking threshold as the foundation that quantizes.This method according to the frequency discrimination characteristic of linear prediction LP and curling linear prediction WLP very near the characteristics of critical band in human auditory's characteristic and masking characteristics, the sampled audio signal that receives is carried out linear prediction LP and curling linear prediction WLP respectively, obtain the amplitude-frequency response of LP wave filter and the amplitude-frequency response of WLP wave filter, according to the amplitude-frequency response of the LP wave filter that obtains and the amplitude-frequency response of WLP wave filter, get access to the part and shelter curve; Absolute masking threshold and critical frequency bandwidth according to the part that obtains is sheltered curve, preset obtain overall masking threshold.

Step 10: according to the overall masking threshold that obtains, to coding audio signal.

The frequency discrimination characteristic of utilizing linear prediction LP and curling linear prediction WLP that present embodiment provides is very near the characteristics of critical band in human auditory's characteristic and masking characteristics, set up the method for psychoacoustic model, obtain overall masking threshold, according to the overall masking threshold that obtains to coding audio signal, reduced the complexity of setting up psychoacoustic model, be easy to realize, reduced psychoacoustic model, reduced the power consumption of hardware at hard-wired cost.

Embodiment two

The embodiment of the invention provides a kind of audio coding method, referring to shown in Figure 9, and with reference to audio coder shown in Figure 1.Obtaining the method for psychoacoustic model overall situation masking threshold in this audio coding method, is that a kind of method of setting up psychoacoustic model that provides among the embodiment one has been provided.The embodiment of the invention provides a kind of audio coding method to comprise:

Step H1: scrambler receives time-domain audio signal;

Wherein, the time-domain audio signal that receives of scrambler be with embodiment one in step 1 in the same step carried out.

Step H2: scrambler is set up psychoacoustic model according to the time-domain audio signal that receives, and obtains overall masking threshold;

Wherein, step H3's has the explanation of manner of execution in can reference example one.

Step H3: scrambler is encoded to the time-domain audio signal that receives according to the overall masking threshold that obtains in the step 2.

Wherein, a kind of audio coding method that provides among this embodiment that needs to understand, this audio coding method, the frequency discrimination characteristic of utilizing linear prediction LP and curling linear prediction WLP is very near the characteristics of critical band in human auditory's characteristic and masking characteristics, the psychoacoustic model of setting up has reduced the complexity of setting up psychoacoustic model, is easy to realize, reduce psychoacoustic model at hard-wired cost, reduced the power consumption of hardware.

Therefore, the method for setting up psychoacoustic model among the embodiment one can be applied in varying environment, and above embodiment two is wherein a kind of applied environments, promptly is applied in the audio coding.This psychoacoustic model can also be applied in the audio frequency watermark system, uses the psychoacoustic model of setting up that embodiment one provides in promptly a kind of audio frequency watermark embedding grammar.The audio frequency watermark system of the psychoacoustic model method of setting up that employing embodiment one provides, also have and reduce the complexity of setting up psychoacoustic model, be easy to realize, reduced psychoacoustic model, reduced the characteristics of the power consumption of hardware at hard-wired cost.

Embodiment three

Present embodiment provides a kind of audio coder, referring to shown in Figure 10, comprise: receiving element 10, sampling unit 20, linear prediction LP unit 30, obtain LP wave filter amplitude-frequency response unit 40, the linear prediction WLP unit 50 that curls, obtain WLP wave filter amplitude-frequency response unit 60, obtain the part and shelter curved unit 70, obtain the overall situation and shelter curved unit 80, obtain masking threshold unit 90 and audio coding unit 100.

Wherein, receiving element 10 receives time-domain audio signal, this time-domain audio signal that receives can be the mixed information of the various voice signals that can hear of voice signal, sound signal or various people's ear, the frequency span of this sound signal can be heard frequency range (being that 0Hz is to 24000Hz) for people's ear usually, sound signal is the form of frame normally, and the length of a frame is generally between 5 milliseconds to 30 milliseconds.

Adopt the 20 pairs of sound signals that receive in unit to adopt, the frequency of employing can be that 48kHz, 44.1kHz, 32kHz, 16kHz, 8kHz etc. are wherein any.The sampled speech signal that 30 pairs of linear prediction LP unit obtain carries out linear prediction LP, also can be described as the coefficient that obtains the LP wave filter according to the employing voice signal that obtains.

Wherein.LP unit 30 can also specifically comprise: first is obtained from related function unit 301, first obtains canonical equation unit 302 and first and obtains coefficient of linear prediction wave filter unit 303.Wherein first be obtained from related function unit 301, be used to obtain the autocorrelation function of the sound signal after the sampling; First obtains the principle of the difference minimum of canonical equation unit 302 between predicting the outcome according to the sound signal after the original signal sampling and linear filter, obtains the canonical equation of LP; First obtains LP filter system unit 303 is obtained from the autocorrelation function that obtains in the related function unit 301 according to first, utilizes Levinson Durbin algorithm to find the solution canonical equation, obtains the LP filter coefficient.

First obtains LP amplitude-frequency response unit 40 according to obtaining the LP filter coefficient in the LP unit 30, obtains the amplitude-frequency response of LP wave filter; Wherein, obtaining the amplitude-frequency response that obtains the LP wave filter in the LP amplitude-frequency response unit 40 obtains according to concluding formula (4).

The voice signal after the 50 pairs of samplings of obtaining in linear prediction WLP unit of the curling linear prediction WLP that curls also can be described as the coefficient that employing voice signal that basis obtains obtains the WLP wave filter.

Wherein, the linear prediction WLP unit 50 that curls is similar to LP unit 30, can also specifically comprise: second is obtained from related function unit 501, second obtains canonical equation unit 502 and second and obtains curling linear prediction filter coefficient elements 503.Wherein second be obtained from related function unit 501, be used to obtain the autocorrelation function that adopts sound signal; Second obtains the principle of the difference minimum of canonical equation unit 502 between predicting the outcome according to the sound signal after the original signal sampling and curling linear filter, obtains the WLP canonical equation; Second obtains WLP filter system unit 503 is obtained from the autocorrelation function that obtains in the related function unit 501 according to second, utilizes Levinson Durbin algorithm to find the solution canonical equation, obtains the WLP filter coefficient.

Obtain WLP wave filter amplitude-frequency response unit 60 according to obtaining the WLP filter coefficient in the WLP unit 50, obtain the amplitude-frequency response of WLP wave filter; Wherein, obtaining the amplitude-frequency response that obtains the WLP wave filter in the WLP amplitude-frequency response unit 60 obtains according to concluding formula (13).

Obtain the part and shelter curved unit 70 according to the amplitude-frequency response that obtains the LP wave filter that obtains in the LP wave filter amplitude-frequency response unit 40, amplitude-frequency response with obtaining the WLP wave filter that obtains in the WLP wave filter amplitude-frequency response unit 60 obtains the part and shelters curve.This obtains the part and shelters curved unit 70 and specifically can comprise: frequency compensation unit 701 and calculate the part and shelter curved unit 702.

Wherein, frequency compensation unit 701 is used for strengthening according to the control information of presetting the amplitude-frequency response C of LP wave filter _LPThe amplitude-frequency response C of low frequency characteristic (f) and reinforcement WLP wave filter _WLP(f) high frequency characteristics; Calculate local shelter curved unit 702 after according to the frequency compensation of obtaining in the frequency compensation unit 701 amplitude versus frequency characte and the relative deviation coefficient that presets, obtain the part and shelter curve, can be with reference to formula (14).

Obtain the overall situation shelter curved unit 80 according to obtain the part shelter the part that obtains in the curved unit 70 shelter curve and preset definitely shelter curve, obtain the overall situation and shelter curve, specifically can be with reference to formula (20).

Obtain the information that curve and the critical band that presets are sheltered according to the overall situation of obtaining in masking threshold unit 90, obtain overall masking threshold.

Audio coding unit 100, this audio coding unit are used for encoding to the received signal according to obtaining masking threshold.

Receiving element 10 in a kind of audio coder that the embodiment of the invention provides, sampling unit 20, linear prediction LP unit 30, obtain LP wave filter amplitude-frequency response unit 40, the linear prediction WLP unit 50 that curls, obtain WLP wave filter amplitude-frequency response unit 60, obtain the part and shelter curved unit 70, obtain the overall situation and shelter curved unit 80 and obtain masking threshold unit 90 and realize obtaining overall masking threshold jointly, can be included in a kind of audio frequency watermark flush mounting.This watermark flush mounting comprises that also watermark embeds the unit.

Wherein, this watermark embeds the unit according to the overall masking threshold that obtains, and watermark encoder is embedded in the input audio signal.

Explanation by above a kind of audio coder that present embodiment is provided, the frequency discrimination characteristic that this audio coder utilizes linear prediction LP and curling linear prediction WLP is very near the characteristics of critical band in human auditory's characteristic and masking characteristics, set up psychoacoustic model, reduced the complexity of setting up psychoacoustic model, be easy to realize, reduce psychoacoustic model at hard-wired cost, reduced the power consumption of hardware.

One of ordinary skill in the art will appreciate that all or part of step in the whole bag of tricks of the foregoing description is to instruct relevant hardware to finish by program, this program can be stored in the computer-readable recording medium, and storage medium can comprise: ROM, RAM, disk or CD etc.

More than a kind of audio coding method and related device that the embodiment of the invention provided are described in detail, used specific case herein principle of the present invention and embodiment are set forth, the explanation of above embodiment just is used for helping to understand method of the present invention and core concept thereof; Simultaneously, for one of ordinary skill in the art, according to thought of the present invention, the part that all can change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.

Claims (13)

1. an audio coding method is characterized in that, comprising:

Receive time-domain audio signal;

Described sound signal is sampled;

Sound signal after the sampling is carried out linear prediction;

According to the result of described linear prediction, obtain the amplitude-frequency response of linear prediction filter;

To the sound signal linear prediction of curling after the sampling;

According to described curling linear prediction result, obtain the amplitude-frequency response of the linear prediction filter that curls;

According to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter, obtain the part and shelter curve;

According to the characteristic of definitely sheltering curve that the described part that obtains is sheltered curve and preset, obtain the overall situation and shelter curve;

According to the information that the described overall situation of obtaining is sheltered curve and the critical band that presets, obtain overall masking threshold;

According to the described overall masking threshold that obtains, the time-domain audio signal of described reception is encoded;

Wherein said according to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter, obtain the part and shelter curve and specifically comprise:

According to the control information of presetting, respectively the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter are carried out frequency compensation;

According to result after the described frequency compensation and the control information of presetting, obtain the part and shelter curve.

2. method according to claim 1 is characterized in that, described to the sampling after sound signal carry out linear prediction, specifically comprise:

Obtain the autocorrelation function of the sound signal after the described sampling;

The value of difference is obtained linear prediction LP canonical equation between predicting the outcome according to the sound signal after the original signal sampling and linear prediction filter;

Find the solution described LP canonical equation according to described autocorrelation function, described LP canonical equation separate coefficient for described linear prediction filter.

3. method according to claim 1 is characterized in that, and is described to the sound signal linear prediction of curling after the sampling, specifically comprises:

Obtain the autocorrelation function of the sound signal after the described sampling;

The value of difference is obtained curling linear prediction WLP canonical equation between predicting the outcome according to sound signal after the original signal sampling and curling linear prediction filter;

Find the solution described WLP canonical equation according to described autocorrelation function, described WLP canonical equation separate coefficient for described curling linear prediction filter.

4. method according to claim 3 is characterized in that, the described autocorrelation function that obtains the sound signal after the described sampling specifically comprises:

Sound signal after the described sampling gets access to the output signal d of each cascade all-pass filter by the single order all-pass filter of the cascade of presetting _k[x (n)], k=0,1 ... N-1, the sound signal after wherein x (n) expression is sampled, n=1,2 ..., L, L are the length of frame, N represents the exponent number of curling linear prediction filter;

Output signal d according to the sound signal after the described sampling and each cascade rank all-pass filter _k[x (n)], k=0,1 ... N-1 is obtained from related function.

5. method according to claim 1 is characterized in that, describedly obtains the part and shelters curve according to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter, is specially:

The amplitude-frequency response of described linear prediction filter and the amplitude of amplitude-frequency response on same frequency of described curling linear prediction filter are averaged, and the curve of the described mean value that obtains is that curve is sheltered in the part.

6. method according to claim 1 is characterized in that, the characteristic of definitely sheltering curve that the part that described basis is obtained is sheltered curve and preset is obtained the overall situation and sheltered curve, specifically comprises:

Obtain the part and shelter curve and the value of definitely sheltering curve amplitude maximum on same frequency that presets, the curve that described maximal value forms on frequency is the overall situation and shelters curve.

7. method according to claim 1 is characterized in that, the characteristic of definitely sheltering curve that the part that described basis is obtained is sheltered curve and preset is obtained the overall situation and sheltered curve, specifically comprises:

According to the characteristic of definitely sheltering curve that absolute masking threshold matching factor, the part of presetting are sheltered curve and preset, obtain the overall situation and shelter curve.

8. according to each described method of claim 1 to 7, it is characterized in that the overall situation that described basis is obtained is sheltered the information of curve and the critical band that presets, and obtains overall masking threshold, specifically comprises:

Obtain critical band end points and frequency band mid point and shelter amplitude value on the curve, the described amplitude value of obtaining is connected to form overall masking threshold according to frequency order with straight line in the overall situation.

9. according to each described method of claim 1 to 7, it is characterized in that the overall situation that described basis is obtained is sheltered the information of curve and the critical band that presets, and obtains overall masking threshold, specifically comprises:

In critical band, obtain the minimum value that the overall situation is sheltered amplitude on the curve, shelter the minimum value of amplitude on the curve as overall masking threshold with the overall situation in the last critical band of each Bark.

10. an audio coder is characterized in that, comprising:

Receiving element is used to receive time-domain audio signal;

Sampling unit is used for described sound signal is sampled;

Linear prediction unit is used for the sound signal after the sampling is carried out linear prediction;

Obtain linear prediction filter amplitude-frequency response unit, be used for result, obtain the amplitude-frequency response of linear prediction filter according to described linear prediction;

The linear prediction unit of curling is used for the linear prediction of curling of the sound signal after the sampling;

Obtain the linear prediction filter amplitude-frequency response unit that curls, be used for, obtain the amplitude-frequency response of the linear prediction filter that curls according to described curling linear prediction result;

Obtain the part and shelter curved unit, be used for obtaining the part and sheltering curve according to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter;

Obtain the overall situation and shelter curved unit, be used for sheltering curve and the characteristic of definitely sheltering curve that presets, obtain the overall situation and shelter curve according to the described part that obtains;

Obtain the masking threshold unit, be used for sheltering the information of curve and the critical band that presets, obtain overall masking threshold according to the described overall situation of obtaining;

Audio coding unit is used for according to the overall masking threshold that obtains the time-domain audio signal of described reception being encoded;

The wherein said part that obtains is sheltered curved unit and is specifically comprised:

Frequency compensation unit is used for respectively the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter being carried out frequency compensation according to the control information of presetting;

Calculate the part and shelter curved unit, be used for obtaining the part and sheltering curve according to result after the described frequency compensation and the control information of presetting.

11. audio coder according to claim 10 is characterized in that, described linear prediction unit specifically comprises:

First is obtained from the related function unit, is used to obtain the autocorrelation function of the sound signal after the described sampling;

First obtains the canonical equation unit, and the value of the difference between being used for predicting the outcome according to the sound signal after the original signal sampling and linear prediction filter is obtained linear prediction LP canonical equation;

First obtains the coefficient of linear prediction wave filter unit, is used for according to described autocorrelation function, finds the solution described LP canonical equation, described LP canonical equation separate coefficient for described linear prediction filter.

12. audio coder according to claim 10 is characterized in that, described curling linear prediction unit specifically comprises:

Second is obtained from the related function unit, is used to obtain the autocorrelation function of the sound signal after the described sampling;

Second obtains the canonical equation unit, and the value of the difference between being used for predicting the outcome according to sound signal after the original signal sampling and curling linear prediction filter is obtained curling linear prediction WLP canonical equation;

Second obtains the coefficient of linear prediction wave filter unit, is used for according to described autocorrelation function, finds the solution described WLP canonical equation, described WLP canonical equation separate coefficient for described curling linear prediction filter.

13. an audio frequency watermark flush mounting is characterized in that, comprising:

Receiving element is used to receive time-domain audio signal;

Sampling unit is used for described sound signal is sampled;

Linear prediction unit is used for the sound signal after the sampling is carried out linear prediction;

Obtain linear prediction filter amplitude-frequency response unit, be used for result, obtain the amplitude-frequency response of linear prediction filter according to described linear prediction;

The linear prediction unit of curling is used for the linear prediction of curling of the sound signal after the sampling;

Obtain the linear prediction filter amplitude-frequency response unit that curls, be used for, obtain the amplitude-frequency response of the linear prediction filter that curls according to described curling linear prediction result;

Obtain the part and shelter curved unit, be used for obtaining the part and sheltering curve according to the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter;

Obtain the overall situation and shelter curved unit, be used for sheltering curve and the characteristic of definitely sheltering curve that presets, obtain the overall situation and shelter curve according to the described part that obtains;

Obtain the masking threshold unit, be used for sheltering the information of curve and the critical band that presets, obtain overall masking threshold according to the described overall situation of obtaining;

Watermark embeds the unit, is used for the overall masking threshold that obtains according to described, and watermark encoder is embedded in the input audio signal;

The wherein said part that obtains is sheltered curved unit and is specifically comprised:

Frequency compensation unit is used for respectively the amplitude-frequency response of described linear prediction filter and the amplitude-frequency response of described curling linear prediction filter being carried out frequency compensation according to the control information of presetting;

Calculate the part and shelter curved unit, be used for obtaining the part and sheltering curve according to result after the described frequency compensation and the control information of presetting.

Images