CN102339607A - Method and device for spreading frequency bands - Google Patents
Method and device for spreading frequency bands Download PDFInfo
- Publication number
- CN102339607A CN102339607A CN2010102330332A CN201010233033A CN102339607A CN 102339607 A CN102339607 A CN 102339607A CN 2010102330332 A CN2010102330332 A CN 2010102330332A CN 201010233033 A CN201010233033 A CN 201010233033A CN 102339607 A CN102339607 A CN 102339607A
- Authority
- CN
- China
- Prior art keywords
- frequency band
- high frequency
- statistical property
- property classification
- parameters
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 58
- 239000011159 matrix material Substances 0.000 claims abstract description 42
- 239000013598 vector Substances 0.000 claims abstract description 42
- 238000009499 grossing Methods 0.000 claims description 48
- 230000002123 temporal effect Effects 0.000 claims description 32
- 239000012141 concentrate Substances 0.000 claims description 6
- 230000007704 transition Effects 0.000 abstract description 14
- 230000008447 perception Effects 0.000 abstract description 8
- 238000012805 post-processing Methods 0.000 abstract description 5
- 238000012549 training Methods 0.000 description 13
- 230000003044 adaptive effect Effects 0.000 description 9
- 238000007493 shaping process Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 6
- 239000000284 extract Substances 0.000 description 6
- 238000013459 approach Methods 0.000 description 4
- 238000001228 spectrum Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 238000005086 pumping Methods 0.000 description 2
- 238000013139 quantization Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000017105 transposition Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 239000012792 core layer Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000012467 final product Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/038—Speech enhancement, e.g. noise reduction or echo cancellation using band spreading techniques
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The embodiment of the invention discloses a method and device for spreading frequency bands. Taking the implementation of the method for instance, the method comprises the following steps of: acquiring the characteristic vectors of low-frequency band signals; carrying out classification on the low-frequency band signals according to the characteristic vectors and the preset statistical characteristic classification characteristic vector set pairs and obtaining a statistical characteristic classification result; obtaining predicted high-frequency band parameters according to the statistical characteristic classification result, the characteristic vectors and the preset statistical characteristic classification state transition matrix; adjusting the predicted high-frequency band parameters according to the statistical characteristic classification result and the preset post-processing smooth factor set and obtaining adjusted high-frequency band parameters; and reconstructing high-frequency band signals. With the adoption of the method, the obtained parameter information for spreading the frequency bands can be more directive, the interframe transition is smoother, and the obtained spread signals have higher hearing perception.
Description
Technical field
The present invention relates to communication technical field, particularly a kind of method and apparatus of band spread.
Background technology
Along with the development of bearing technology, People more and more is not content with the quality of narrowband speech codec, thus audio coder & decoder (codec) progressively to the broadband (Windband, WB), ultra broadband (Super Windband, SWB) expansion.(the International Telecommunication Union of International Telecommunications Union (ITU) for example; ITU) released G.722, G.722.1, G.722.2, G.729.1 waited broadband encoding and decoding speech standard; 3G (Third Generation) Moblie partnership projects (Third Generation Partnership Project; 3GPP) released AMR-WB (Adaptive Multi-Rate Windband; AMR-WB) (be ITU G.722.2) this broadband voice encoding and decoding standard, 3GPP2 then released the variable Rate multi-mode wideband (Variable-Rate Multimode Windband, VMR-WB).G.711, WB&G.722 associating ultra broadband etc. ITU has proposed G.729.1&G.718 to unite ultra broadband recently again in addition.These standards are all from arrowband expansion, and core layer is generally Code Excited Linear Prediction (Code-Excited Linear-Prediction, CELP) coding, and broadband, ultra broadband partly use transition coding technological.Transition coding has a lot, and the discrete cosine transform of for example revising (Modified Discrete Cosine Transform, MDCT), the conversion code excitation (Transform Coded exciation, TCX) etc.
Band spread is used in the voice/audio coding field very widely, can improve the perceived quality of limit band voice/music effectively, and the quality enhancement techniques of on the terminal, using based on band spread is exactly one type of good application example.Band spreading technique also is applied in the embedded variable rate speech coding device by wide model, and the audio bandwidth that particularly when transmission channel conditions changes, produces switches.Common bandwidth switch mainly contain the arrowband (Narrow band, NB), broadband, ultra broadband, full band (Full Band, the switching between FB).
The method that realizes band spread can be divided into two kinds of the band spreads of band spread that side information is arranged and no side information.There is the band spread of side information to extract some characteristic informations of treating extending bandwidth, and these information products are delivered to decoding end, instruct decoding end to carry out the frequency band corresponding expansion at coding side.The band spread of no side information is called blind expansion again, need be in the coding side information extraction, and the information of the partial-band that only need obtain according to decoding end is through the information of the required extending bandwidth of the artificial generation of certain algorithm for estimating.
The method of band spread can also be divided into based on the expansion of time domain with based on the expansion of frequency domain.The time-domain information of the partial-band that normally obtains based on decoding end based on the expansion of time domain carries out obtaining the time-domain information of required extending bandwidth after the shaping of time domain and frequency domain, thereby realizes band spread.The frequency domain information of the partial-band that normally obtains based on decoding end based on the expansion of frequency domain carries out obtaining the frequency domain information of required extending bandwidth after the shaping of frequency domain and time domain, thereby realizes band spread.
At present, the processing that the band spreading technique of no side information generally carries out in time domain wherein has a kind of method to be based on the piecewise linear maps frequency range expanding method of statistical property.The performing step of this method:
1, extracts the eigenvector of the partial-band that decoding obtains;
2, compare through the statistical property characteristic of division vector set that training in advance before the eigenvector that extracts and the band spread is obtained, signal is classified; Above-mentioned training is meant: according to certain rule, between data are concentrated, extract useful information, use the guidance of these useful informations that these data are divided into different classes, represent with its corresponding useful information for of a sort data.
3, according to type corresponding predefined state-transition matrix of above-mentioned branch, obtain the parameter information of required extending bandwidth, thereby realize band spread.
The inventor finds in realizing process of the present invention: because signal is divided into limited several types; The parameter information of the extending bandwidth that therefore can generate has only limited several kinds; Can't adaptive signal characteristic widely, cause the interframe transition unsmooth, cause auditory perception poor.
Summary of the invention
The technical matters that the embodiment of the invention will solve provides a kind of method and apparatus of band spread, improves auditory perception.
For solving the problems of the technologies described above, the method embodiment of band spread provided by the present invention can realize through following technical scheme:
Obtain the eigenvector of low band signal;
Reach preset statistical property characteristic of division vector set to said low band signal is classified according to said eigenvector, obtain the statistical property classification results;
According to said statistical property classification results, eigenvector and preset statistical property classification state-transition matrix, the high frequency band parameters that obtains estimating;
According to said statistical property classification results and preset aftertreatment smoothing factor collection, the said high frequency band parameters of estimating is adjusted, obtain adjusted high frequency band parameters;
According to adjusted high frequency band parameters, rebuild high-frequency band signals.
A kind of device of band spread comprises:
The vector acquiring unit is used to obtain the eigenvector of low band signal;
Taxon is used for reaching preset statistical property characteristic of division vector set to said low band signal is classified according to said eigenvector, obtains the statistical property classification results;
Estimate the unit, be used for according to said statistical property classification results, eigenvector and preset statistical property classification state-transition matrix, the high frequency band parameters that obtains estimating;
Adjustment unit is used for according to said statistical property classification results and preset aftertreatment smoothing factor collection the said high frequency band parameters of estimating being adjusted, and obtains adjusted high frequency band parameters;
The signal reconstruction unit is used for according to adjusted high frequency band parameters, rebuilds high-frequency band signals.
Technique scheme has following beneficial effect: on the basis based on the piecewise linear maps band spread algorithm of statistical property, increased a self-adaptation aftertreatment; This method has effectively utilized the classified information that obtains in the piecewise linear maps band spread algorithm; The parameter information of the extending bandwidth that piecewise linear maps band spread algorithm is obtained carries out adaptive aftertreatment again by class; Make that the extending bandwidth parameter information that obtains is more targeted; The interframe transition is more level and smooth, and the signal that expands that obtains has higher auditory perception.
Description of drawings
In order to be illustrated more clearly in the technical scheme in the embodiment of the invention; The accompanying drawing of required use is done to introduce simply in will describing embodiment below; Obviously, the accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills; Under the prerequisite of not paying creative work property, can also obtain other accompanying drawing according to these accompanying drawings.
Fig. 1 is the embodiment of the invention one a method flow synoptic diagram;
Fig. 2 is the embodiment of the invention two method flow synoptic diagram;
Fig. 3 is the embodiment of the invention three method schematic flow sheets;
Fig. 4 is the embodiment of the invention four apparatus structure synoptic diagram;
Fig. 5 is the embodiment of the invention four apparatus structure synoptic diagram;
Fig. 6 is the embodiment of the invention four apparatus structure synoptic diagram;
Fig. 7 is the embodiment of the invention four apparatus structure synoptic diagram.
Embodiment
To combine the accompanying drawing in the embodiment of the invention below, the technical scheme in the embodiment of the invention is carried out clear, intactly description, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills are not making the every other embodiment that is obtained under the creative work prerequisite, all belong to the scope of the present invention's protection.
Embodiment one, and the embodiment of the invention provides a kind of method of band spread, and is as shown in Figure 1, comprising:
101: the eigenvector that obtains low band signal;
Particularly, above-mentioned eigenvector can comprise: temporal envelope and linear predictor coefficient, above-mentioned temporal envelope are represented the energy size of each subframe signal in the time domain, the resonance peak position and the amplitude of above-mentioned linear predictor coefficient expression signal.Certain above-mentioned eigenvector can also comprise: parameters such as the frequency domain envelope of low band signal, frequency-domain linear prediction coefficient will not limit this embodiment of the invention.
102: according to above-mentioned eigenvector and preset statistical property characteristic of division vector set above-mentioned low band signal is classified, obtain the statistical property classification results;
103: according to above-mentioned statistical property classification results, eigenvector and preset statistical property classification state-transition matrix, the high frequency band parameters that obtains estimating;
Particularly, in above-mentioned 103, according to statistical property classification results, eigenvector and preset statistical property classification state-transition matrix, the high frequency band parameters that obtains estimating comprises:
According to above-mentioned statistical property classification results, the corresponding state-transition matrix of the above-mentioned classification results of inquiry in preset statistical property classification state-transition matrix; The high frequency band parameters that obtains estimating according to above-mentioned state-transition matrix and above-mentioned eigenvector.
The statistical property classification results that certain above-mentioned basis obtains, the eigenvector of low band signal and preset statistical property classification state-transition matrix; The high frequency band parameters that obtains estimating; Other modes can also be arranged; For example following dual mode: (1) is inquired about the corresponding state-transition matrix of this classification results according to the statistical property classification results that obtains in preset statistical property classification state-transition matrix; Multiply by the eigenvector of low band signal, the high frequency band parameters that obtains estimating according to the corresponding state-transition matrix of the statistical property classification results that obtains.(2) according to the statistical property classification results that obtains, the corresponding state transitions map index value of this classification results of inquiry in preset statistical property classification state-transition matrix; Table look-up and obtain the corresponding high frequency band parameters of estimating according to obtaining state transitions map index value.
Particularly, above-mentioned high frequency band parameters can comprise: temporal envelope and frequency domain envelope; Above-mentioned temporal envelope is represented the energy size of each subframe signal in the time domain, and above-mentioned frequency domain envelope is represented the gain size of each subband signal in the frequency domain.Certain above-mentioned high frequency band parameters can also comprise: parameters such as the time domain linear predictive coefficient of high-frequency band signals, frequency-domain linear prediction coefficient will not limit this embodiment of the invention.
104: according to above-mentioned statistical property classification results and preset aftertreatment smoothing factor collection, the above-mentioned high frequency band parameters of estimating is adjusted, obtained adjusted high frequency band parameters;
If said high frequency band parameters comprises: temporal envelope and frequency domain envelope; Said aftertreatment factor parameter comprises: smoothing factor in the frame; Then said the high frequency band parameters of estimating is adjusted; Obtain adjusted high frequency band parameters can for: according to smoothing factor in the corresponding frame of statistical property classification; Temporal envelope parameter and frequency domain envelope parameters in the high frequency band parameters that adjustment is estimated obtain adjusted high frequency band parameters.
Particularly; Above-mentioned according to statistical property classification results and preset aftertreatment smoothing factor collection; The above-mentioned high frequency band parameters adjustment of estimating is comprised:, concentrate the corresponding aftertreatment factor parameter of the above-mentioned statistical property classification results of inquiry at preset aftertreatment smoothing factor according to above-mentioned statistical property classification results; According to the aftertreatment factor parameter corresponding with above-mentioned statistical property classification results, adjust the above-mentioned high frequency band parameters of estimating, obtain adjusted high frequency band parameters.
Certainly; Classification results that above-mentioned basis obtains and preset aftertreatment smoothing factor collection; The method that the high frequency band parameters of estimating is adjusted can also have other mode; Give an example for for example following 3: the aftertreatment factor parameter of this classification results correspondence of inquiry is concentrated according to the statistical property classification results that obtains in (1) at preset aftertreatment smoothing factor; According to the corresponding aftertreatment factor parameter of the statistical property classification results that obtains; According to the aftertreatment factor parameter carry out in the frame to the high frequency band parameters of estimating and/or interframe level and smooth; The high frequency band parameters that adjustment is adaptively estimated obtains adjusted high frequency band parameters.(2), concentrate the corresponding aftertreatment factor parameter of this classification results of inquiry at preset aftertreatment smoothing factor according to the statistical property classification results that obtains; According to the corresponding aftertreatment factor parameter of the statistical property classification results that obtains, the high frequency band parameters of estimating carried out parameter decay or strengthen according to the aftertreatment factor parameter, the high frequency band parameters that adjustment is adaptively estimated obtains adjusted high frequency band parameters.(3), concentrate the corresponding aftertreatment factor parameter of this classification results of inquiry at preset aftertreatment smoothing factor according to the statistical property classification results that obtains; According to the corresponding aftertreatment factor parameter of the statistical property classification results that obtains; According to the aftertreatment factor parameter carry out in the frame to the high frequency band parameters of estimating and/or interframe level and smooth; The line parameter of going forward side by side decay or enhancing; The high frequency band parameters that adjustment is adaptively estimated obtains adjusted high frequency band parameters.
105:, rebuild high-frequency band signals according to above-mentioned adjusted high frequency band parameters.
Particularly, above-mentioned aftertreatment factor parameter can comprise: smoothing factor and interframe smoothing factor in the frame.
More specifically, the high frequency band parameters that above-mentioned adjustment is estimated comprises: according to smoothing factor in the corresponding frame of statistical property classification, the temporal envelope parameter in the high frequency band parameters that adjustment is estimated; According to the corresponding interframe smoothing factor of statistical property classification, the frequency domain envelope parameters in the high frequency band parameters that preliminary adjustment is estimated; According to smoothing factor in the corresponding frame of statistical property classification, preliminary adjusted high frequency band frequency domain envelope parameters is adjusted again, obtain adjusted high frequency band parameters.
The executive agent of said method can be any device that carries out band spread; The method that the embodiment of the invention provides has increased a self-adaptation aftertreatment on the basis based on the piecewise linear maps band spread algorithm of statistical property; This method has effectively utilized the classified information that obtains in the piecewise linear maps band spread algorithm; The parameter information of the extending bandwidth that piecewise linear maps band spread algorithm is obtained carries out adaptive aftertreatment again by class; Make that the extending bandwidth parameter information that obtains is more targeted, the interframe transition is more level and smooth, and the signal that expands that obtains has higher auditory perception.
Embodiment two, and the method that the embodiment of the invention gives band spread is instance more specifically, and is as shown in Figure 2, comprises the steps:
201: decoding obtains low band signal;
202: the eigenvector X that extracts low band signal
f
Eigenvector can have various array modes, only needs to reflect that the characteristic of low band signal gets final product.For example, eigenvector can comprise the temporal envelope and the linear predictor coefficient of low band signal, also can comprise the temporal envelope and the frequency domain envelope of low band signal.
203: according to the eigenvector X of the low band signal that obtains
fAnd preset statistical property characteristic of division vector set X
F, j(j ∈ 1 ..., M}), low band signal is classified, obtain statistical property classification i.
Concrete realization can be used the method for vector quantization, and preset statistical property characteristic of division vector set as code book, is searched for and eigenvector X in code book
fThe code word that distance is minimum; This code word corresponding index in code book is statistical property classification i: code book is an array; Comprised tactic M the characteristic of division vector that classification is corresponding, each eigenvector is exactly a code word, and M eigenvector is exactly M code word.The position of this code word in code book represented in the index of code word, index is corresponding classification number.
Wherein || ||
2Square error is calculated in expression.
204: according to the statistical property classification i that obtains, the eigenvector X of low band signal
fAnd preset statistical property classification state-transition matrix collection H, estimate high frequency band parameters
Wherein
The statistical property classification state-transition matrix H that expression is corresponding to statistical property classification i
iCalculate transposition (property sort i is corresponding i vector H among the state-transition matrix H
i).Statistical property category set and statistical property classification state-transition matrix have kept one-to-one relationship when training out simultaneously.
Above-mentioned high frequency band parameters can have the different combinations mode, as long as can reflect the characteristic of high-frequency band signals.For example, high frequency band parameters can comprise the temporal envelope and the linear predictor coefficient of high-frequency band signals, also can comprise the temporal envelope and the frequency domain envelope of high-frequency band signals.The array mode that need to prove the eigenvector in high frequency band parameters and the preceding text is can be inconsistent, does not influence the realization of the embodiment of the invention.
205: according to statistical property classification i and preset aftertreatment factor set; Carry out the self-adaptation adjustment to estimating high frequency band parameters, obtain adjusted high frequency band parameters
Above-mentioned preset aftertreatment factor set can have the different combinations mode, can only comprise the interframe smoothing factor, also can comprise smoothing factor in interframe smoothing factor and the frame, can also comprise the different aftertreatment factors such as the state transition factor.Each parameter in the preset aftertreatment factor set can be directed against the different statistic property sort respectively, thereby embodies the adaptive characteristic of post-processing approach and fully use the characteristics of statistical property classification.
206: rebuild high-frequency band signals according to adjusted high frequency band parameters.
Rebuild high-frequency band signals according to the high frequency band parameters that obtains; The particular content that mainly comprises according to high frequency band parameters; For example, when high frequency band parameters has comprised temporal envelope and frequency domain envelope, can adopt the method for frequency domain band spread; The frequency domain spectra of low-frequency band is transformed to time domain after according to the frequency domain envelope shaping, carry out the high-frequency band signals that shaping obtains rebuilding according to temporal envelope again; Also can adopt the method for time domain band spread, the time domain pumping signal of low-frequency band is carried out transforming to frequency domain after the shaping according to temporal envelope, carry out shaping according to the frequency domain envelope again, time domain is returned in conversion more at last, the high-frequency band signals that obtains rebuilding.When high frequency band parameters has comprised the high frequency band linear predictor coefficient, can use the composite filter of the time domain pumping signal of low-frequency band, the high-frequency band signals that obtains rebuilding through high frequency band linear predictor coefficient formation.
Below to preset statistical property characteristic of division vector set, preset statistical property classification state-transition matrix collection and preset aftertreatment factor set, be that the statistical property according to a large amount of signals obtains, concrete training method is following:
At first, from training set, extract the low-frequency band eigenvector and the corresponding high frequency band parameters vector of each signal, form low-frequency band eigenvector training set and high frequency band parameters vector training set respectively; Above-mentioned training set is the data set that is used to train, and this data set is previously selected voice/audio language material.
Then, according to the different statistic characteristic, according to the method for cluster, training obtains low-frequency band feature vector set X from low-frequency band eigenvector training set
F, j(j ∈ 1 ..., M}), the cluster according to statistical property characteristic of division vector set obtains corresponding high frequency band parameters vector set Y simultaneously
F, j(j ∈ 1 ..., M}).
According to each statistical nature classification j ∈ 1 ..., the training data that M} is corresponding (the corresponding low-frequency band eigenvector X of statistical nature classification j
jWith the high frequency band parameters vector Y
i), calculate corresponding state-transition matrix:
H
j＝X
j +·Y
j＝(X
j TX
j)
-1X
j T·Y
j
X
j +Vector X is asked in expression
jPseudo-inverse operation, X
j +=(X
j TX
j)
-1X
j TEach statistical nature classification j ∈ 1 ..., the state-transition matrix that M} is corresponding has just constituted state-transition matrix collection H
j(j ∈ 1 ..., M}).
According to trained low-frequency band feature vector set and state-transition matrix collection; According to the method for classification linear mapping band spread, each signal estimates high frequency band vector
in the calculation training
Calculate each statistical nature classification corresponding reliable sex factor, and with it as the parameter in the aftertreatment factor set.Above-mentioned reliability factor is α
iThe scope of the aftertreatment factor is more wide in range, can only comprise α
i, also can comprise other factors except that reliability factor.The corresponding average error in classification of statistical nature classification i is:
Wherein
Expression belongs to n the 1st component estimating the high frequency band vector of i statistical nature classification,
Expression belongs to the 1st component of n actual high frequency band parameters vector of i statistical nature classification, and N is the number that belongs to all vectors of i statistical nature classification in the training set, N
yDimension for the high frequency band parameters vector.
Can obtain statistical nature classification i corresponding reliable sex factor according to the corresponding average error in classification of statistical nature classification i:
Wherein c is a constant.
In the practical application, can only calculate one group of reliability factor α
iAlso can calculate several groups of reliability factors such as α respectively to the different parameter that comprises in the high frequency band parameters vector
i, β
iDeng, form the reliability factor vector and with it as the aftertreatment factor set, will not limit counting the embodiment of the invention with the group of reliability factor.Use the instance of many group reliability factors, for example,, can use top method to calculate each statistical nature classification corresponding reliable sex factor note respectively and make α if comprised temporal envelope parameter and frequency domain envelope parameters in the high frequency band parameters vector
iAnd β
i, form reliability factor vector { α
i, β
i.
In above-mentioned instance, the low-frequency band feature vector set X that trains
F, j(j ∈ 1 ..., M}) be preset statistical property characteristic of division vector set, the state-transition matrix collection H that trains
j(j ∈ 1 ..., M}) being preset statistical property classification state-transition matrix collection, the aftertreatment factor set that trains is preset aftertreatment factor set.
The method that the embodiment of the invention provides has increased a self-adaptation aftertreatment on the basis based on the piecewise linear maps band spread algorithm of statistical property; This method has effectively utilized the classified information that obtains in the piecewise linear maps band spread algorithm; The parameter information of the extending bandwidth that piecewise linear maps band spread algorithm is obtained carries out adaptive aftertreatment again by class; Make that the extending bandwidth parameter information that obtains is more targeted; The interframe transition is more level and smooth, and the signal that expands that obtains has higher auditory perception.
Embodiment three, present embodiment provide one to be applied in the frequency expansion method of leniently taking ultra broadband in the ultra broadband demoder to; The method that it is understandable that present embodiment also can be applied to from the arrowband to the broadband; Expansion from the arrowband to the ultra broadband, present embodiment should not be construed as the qualification to the embodiment of the invention as an instance.Present embodiment medium and low frequency band signal is broadband signal, and signal bandwidth is 0～7KHz, and high-frequency band signals is ultra-broadband signal, and signal bandwidth is 7～14KHz.Composite signal sampling rate 32KHz, signal is a frame with 20ms, i.e. N=160 point/frame.The method of concrete band spread is as shown in Figure 3, comprises the steps:
301: decoding obtains broadband signal, and extracts the eigenvector X of broadband signal
f
According to wideband decoded method in the ultra broadband demoder, obtain the wideband decoded signal of present frame, note is made x
nIn the present embodiment, the eigenvector of broadband signal is that example describes with temporal envelope and the linear predictor coefficient that comprises broadband signal.At first, find the solution the N rank linear predictor coefficient LPC of broadband signal
Low={ LPC
Low(0), LPC
Low(1) ..., LPC
Low(K-1) }, concrete grammar can use Lay Wen Xun-Du Bin algorithm.K=64 in the present embodiment can certainly choose other exponent number.Then, calculate temporal envelope: a frame signal is divided into the L sub-frame, and each subframe N/L sampling point calculates the energy of each subframe respectively, with the signal temporal envelope E of the energy of L sub-frame
Low={ E
Low(0), E
Low(1) ..., E
Low(L-1) }, the energy E of i sub-frame wherein
Low(i):
L=8 in the present embodiment can certainly choose other time domain sub-frame division mode.
So, the eigenvector X of broadband signal
fCan remember work:
X
f＝{E
low(0)，E
low(1)，…，E
low(L-1)，LPC
low(0)，LPC
low(1)，…，LPC
low(N-1)}
302: according to the eigenvector X of the broadband signal that obtains
fAnd preset statistical property characteristic of division vector set X
F, j(j ∈ 1 ..., M}), broadband signal is classified, obtain statistical property classification i.
Obtain the concrete realization of statistical property classification i and can use the method for vector quantization, preset statistical property characteristic of division vector set as code book, is searched for and eigenvector X in code book
fThe code word that distance is minimum, the corresponding index of this code word are statistical property classification i:
Wherein || ||
2Square error, M=8 in the present embodiment are calculated in expression.
303: according to the statistical property classification i that obtains, the eigenvector X of broadband signal
fAnd preset statistical property classification state-transition matrix collection H, estimate the ultra broadband parameter
Wherein
The statistical property classification state-transition matrix H that expression is corresponding to statistical property classification i
iCalculate transposition.The ultra broadband parameter can have the different combinations mode, as long as can reflect the characteristic of ultra-broadband signal.In the present embodiment, the ultra broadband parameter has comprised the temporal envelope and the frequency domain envelope of ultra-broadband signal.Estimate ultra broadband parameter
and can remember work:
E wherein
High(0), E
High(1) ..., E
High(L-1) estimate temporal envelope parameter, E for ultra-broadband signal
High(i) the time domain energy parameter of expression i sub-frame, i=0 ..., L-1, L=8 in the present embodiment can certainly choose other time domain sub-frame division mode.G
High(0), G
High(1) ..., G
High(P-1) estimate frequency domain envelope parameters, G for ultra-broadband signal
High(i) the frequency domain gain factor of i subband of expression, i=0 ..., P-1, P=18 in the present embodiment can certainly choose other frequency domain sub-band division mode.
304: according to statistical property classification i and preset aftertreatment factor set; Carry out the self-adaptation adjustment to estimating the ultra broadband parameter, obtain adjusted ultra broadband parameter
Preset aftertreatment factor set can have the different combinations mode, can only comprise the interframe smoothing factor, also can comprise smoothing factor in interframe smoothing factor and the frame, can also comprise the different aftertreatment factors such as the state transition factor.Each parameter in the preset aftertreatment factor set has embodied the adaptive characteristic of post-processing approach and has fully used the characteristics of statistical property classification respectively to the different statistic property sort.In the present embodiment, the aftertreatment factor set has comprised smoothing factor collection and interframe smoothing factor collection in the frame.The interior smoothing factor note of the frame that statistical property classification i is corresponding is made α
i, interframe smoothing factor note is made β
iAccording to statistical property classification i and preset aftertreatment factor set, carry out the process of self-adaptation adjustment and comprise estimating the ultra broadband parameter:
(1) according to smoothing factor α in the corresponding frame of statistical property classification i
i, the temporal envelope parameter in the ultra broadband parameter is estimated in adjustment.
Wherein the L-1 of
expression former frame is individual without the ultra broadband temporal envelope of crossing adjustment, the adjusted ultra broadband temporal envelope of
expression present frame.
(2) according to the corresponding interframe smoothing factor β of statistical property classification i
i, the frequency domain envelope parameters in the ultra broadband parameter is estimated in preliminary adjustment.
Wherein
expression present frame is without the ultra broadband frequency domain envelope of crossing adjustment;
expression former frame is without the ultra broadband frequency domain envelope of crossing adjustment, and
representes the preliminary adjusted ultra broadband frequency domain envelope of present frame.
(3) according to statistical property classification i, preliminary adjusted ultra broadband frequency domain envelope parameters is adjusted again.
Wherein const is a constant factor, and const carries the assembly average of energy in the present embodiment for the ultra broadband frequency domain.
So adjusted ultra broadband parameter
note is done:
305: rebuild high-frequency band signals according to adjusted ultra broadband parameter.
Rebuild ultra-broadband signal according to the ultra broadband parameter that obtains, the particular content that mainly comprises according to the ultra broadband parameter.In the present embodiment; The ultra broadband parameter has comprised temporal envelope and frequency domain envelope, can adopt the method for frequency domain band spread, at first duplicates the frequency domain spectra in broadband and composes as ultra broadband; According to adjusted frequency domain envelope the ultra broadband spectrum is carried out shaping then; After the shaping ultra broadband spectrum signal is transformed to time domain, carry out the time domain shaping according to adjusted temporal envelope again, the high-frequency band signals that obtains rebuilding.
The method that the embodiment of the invention provides has increased a self-adaptation post-processing approach on the basis based on the piecewise linear maps band spread algorithm of statistical property; This method has effectively utilized the classified information that obtains in the piecewise linear maps band spread algorithm; The parameter information of the extending bandwidth that piecewise linear maps band spread algorithm is obtained carries out adaptive aftertreatment again by class; Make that the extending bandwidth parameter information that obtains is more targeted; The interframe transition is more level and smooth, and the signal that expands that obtains has higher auditory perception.
Embodiment four, the embodiment of the invention also provide a kind of device of band spread, and be as shown in Figure 4, comprising:
Vector acquiring unit 401 is used to obtain the eigenvector of low band signal;
Particularly, above-mentioned adjustment unit 404 as shown in Figure 5 comprises:
Aftertreatment factor query unit 501 is used for according to above-mentioned statistical property classification results, concentrates the corresponding aftertreatment factor parameter of the above-mentioned statistical property classification results of inquiry at preset aftertreatment smoothing factor;
Particularly, as shown in Figure 6, the above-mentioned unit 403 of estimating comprises:
Particularly, as shown in Figure 7, above-mentioned adjustment unit 404 comprises:
The 3rd adjustment unit 703 is used for according to smoothing factor in the corresponding frame of statistical property classification preliminary adjusted high frequency band frequency domain envelope parameters being adjusted again, obtains adjusted high frequency band parameters.
The device that the embodiment of the invention provides has increased a self-adaptation post-processing approach on the basis based on the piecewise linear maps band spread algorithm of statistical property; This method has effectively utilized the classified information that obtains in the piecewise linear maps band spread algorithm; The parameter information of the extending bandwidth that piecewise linear maps band spread algorithm is obtained carries out adaptive aftertreatment again by class; Make that the extending bandwidth parameter information that obtains is more targeted; The interframe transition is more level and smooth, and the signal that expands that obtains has higher auditory perception.
One of ordinary skill in the art will appreciate that all or part of step that realizes in the foregoing description method is to instruct relevant hardware to accomplish through program; Above-mentioned program can be stored in a kind of computer-readable recording medium; The above-mentioned storage medium of mentioning can be a ROM (read-only memory), disk or CD etc.
More than the method and apparatus of a kind of band spread that the embodiment of the invention provided has been carried out detailed introduction; Used concrete example among this paper 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 on embodiment and range of application, all can change, to sum up, this description should not be construed as limitation of the present invention.
Claims (11)
1. the method for a band spread is characterized in that, comprising:
Obtain the eigenvector of low band signal;
Reach preset statistical property characteristic of division vector set to said low band signal is classified according to said eigenvector, obtain the statistical property classification results;
According to said statistical property classification results, eigenvector and preset statistical property classification state-transition matrix, the high frequency band parameters that obtains estimating;
According to said statistical property classification results and preset aftertreatment smoothing factor collection, the said high frequency band parameters of estimating is adjusted, obtain adjusted high frequency band parameters;
According to adjusted high frequency band parameters, rebuild high-frequency band signals.
2. according to the said method of claim 1, it is characterized in that, said according to statistical property classification results and preset aftertreatment smoothing factor collection, the said high frequency band parameters adjustment of estimating is comprised:
According to said statistical property classification results, concentrate the corresponding aftertreatment factor parameter of the said statistical property classification results of inquiry at preset aftertreatment smoothing factor;
According to the aftertreatment factor parameter corresponding with said statistical property classification results, adjust the said high frequency band parameters of estimating, obtain adjusted high frequency band parameters.
3. according to claim 1 or 2 said methods, it is characterized in that said according to statistical property classification results, eigenvector and preset statistical property classification state-transition matrix, the high frequency band parameters that obtains estimating comprises:
According to said statistical property classification results, the corresponding state-transition matrix of the said classification results of inquiry in preset statistical property classification state-transition matrix;
The high frequency band parameters that obtains estimating according to said state-transition matrix and said eigenvector.
4. according to claim 1 or 2 said methods; It is characterized in that said eigenvector comprises: temporal envelope and linear predictor coefficient; Said temporal envelope is represented the energy size of each subframe signal in the time domain, the resonance peak position and the amplitude of said linear predictor coefficient expression signal.
5. according to claim 1 or 2 said methods, it is characterized in that said high frequency band parameters comprises: temporal envelope and frequency domain envelope; Said aftertreatment factor parameter comprises: smoothing factor in the frame; Said the high frequency band parameters of estimating is adjusted, is obtained adjusted high frequency band parameters and comprise:
According to smoothing factor in the corresponding frame of statistical property classification, temporal envelope parameter and frequency domain envelope parameters in the high frequency band parameters that adjustment is estimated obtain adjusted high frequency band parameters.
6. according to claim 1 or 2 said methods, it is characterized in that said aftertreatment factor parameter comprises: smoothing factor and interframe smoothing factor in the frame.
7. according to the said method of claim 6, it is characterized in that the high frequency band parameters that said adjustment is estimated comprises:
According to smoothing factor in the corresponding frame of statistical property classification, the temporal envelope parameter in the high frequency band parameters that adjustment is estimated;
According to the corresponding interframe smoothing factor of statistical property classification, the frequency domain envelope parameters in the high frequency band parameters that preliminary adjustment is estimated;
According to smoothing factor in the corresponding frame of statistical property classification, preliminary adjusted high frequency band frequency domain envelope parameters is adjusted again, obtain adjusted high frequency band parameters.
8. the device of a band spread is characterized in that, comprising:
The vector acquiring unit is used to obtain the eigenvector of low band signal;
Taxon is used for reaching preset statistical property characteristic of division vector set to said low band signal is classified according to said eigenvector, obtains the statistical property classification results;
Estimate the unit, be used for according to said statistical property classification results, eigenvector and preset statistical property classification state-transition matrix, the high frequency band parameters that obtains estimating;
Adjustment unit is used for according to said statistical property classification results and preset aftertreatment smoothing factor collection the said high frequency band parameters of estimating being adjusted, and obtains adjusted high frequency band parameters;
The signal reconstruction unit is used for according to adjusted high frequency band parameters, rebuilds high-frequency band signals.
9. said according to Claim 8 device is characterized in that said adjustment unit comprises:
Aftertreatment factor query unit is used for according to said statistical property classification results, concentrates the corresponding aftertreatment factor parameter of the said statistical property classification results of inquiry at preset aftertreatment smoothing factor;
The adjustment subelement is used for the basis aftertreatment factor parameter corresponding with said statistical property classification results, adjusts the said high frequency band parameters of estimating, and obtains adjusted high frequency band parameters.
10. according to Claim 8 or 9 said devices, it is characterized in that the said unit of estimating comprises:
The matrix query unit is used for according to said statistical property classification results, the corresponding state-transition matrix of the said classification results of inquiry in preset statistical property classification state-transition matrix;
Estimate subelement, be used for the high frequency band parameters that obtains estimating according to said state-transition matrix and said eigenvector.
11., it is characterized in that said adjustment unit comprises according to the said device of claim 10:
First adjustment unit is used for according to smoothing factor in the corresponding frame of statistical property classification, the temporal envelope parameter in the high frequency band parameters that adjustment is estimated;
Second adjustment unit is used for according to the corresponding interframe smoothing factor of statistical property classification, the frequency domain envelope parameters in the high frequency band parameters that preliminary adjustment is estimated;
The 3rd adjustment unit is used for according to smoothing factor in the corresponding frame of statistical property classification preliminary adjusted high frequency band frequency domain envelope parameters being adjusted again, obtains adjusted high frequency band parameters.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102330332A CN102339607A (en) | 2010-07-16 | 2010-07-16 | Method and device for spreading frequency bands |
| PCT/CN2011/075079 WO2011144130A1 (en) | 2010-07-16 | 2011-06-01 | Method and device for frequency band expanding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102330332A CN102339607A (en) | 2010-07-16 | 2010-07-16 | Method and device for spreading frequency bands |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102339607A true CN102339607A (en) | 2012-02-01 |
Family
ID=44991196
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010102330332A Pending CN102339607A (en) | 2010-07-16 | 2010-07-16 | Method and device for spreading frequency bands |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN102339607A (en) |
| WO (1) | WO2011144130A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015043161A1 (en) * | 2013-09-26 | 2015-04-02 | 华为技术有限公司 | Method and device for bandwidth extension |
| CN109788922A (en) * | 2016-10-14 | 2019-05-21 | 公立大学法人大阪府立大学 | Swallow diagnostic device and program |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1520590A (en) * | 2001-06-28 | 2004-08-11 | �ʼҷ����ֵ�������˾ | Wideband signal transmission system |
| CN1992533A (en) * | 2005-12-26 | 2007-07-04 | 索尼株式会社 | Signal encoding device and signal encoding method, signal decoding device and signal decoding method, program, and medium |
| CN101076853A (en) * | 2004-12-10 | 2007-11-21 | 松下电器产业株式会社 | Wide-band encoding device, wide-band lsp prediction device, band scalable encoding device, wide-band encoding method |
| WO2008101324A1 (en) * | 2007-02-23 | 2008-08-28 | Qnx Software Systems (Wavemakers), Inc. | High-frequency bandwidth extension in the time domain |
| US20080221905A1 (en) * | 2006-10-18 | 2008-09-11 | Markus Schnell | Encoding an Information Signal |
| CN101620854A (en) * | 2008-06-30 | 2010-01-06 | 华为技术有限公司 | Method, system and device for frequency band expansion |
| CN101751926A (en) * | 2008-12-10 | 2010-06-23 | 华为技术有限公司 | Signal coding and decoding method and device, and coding and decoding system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101304261B (en) * | 2007-05-12 | 2011-11-09 | 华为技术有限公司 | Method and apparatus for spreading frequency band |
| CN101770777B (en) * | 2008-12-31 | 2012-04-25 | 华为技术有限公司 | LPC (linear predictive coding) bandwidth expansion method, device and coding/decoding system |
-
2010
- 2010-07-16 CN CN2010102330332A patent/CN102339607A/en active Pending
-
2011
- 2011-06-01 WO PCT/CN2011/075079 patent/WO2011144130A1/en active Application Filing
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1520590A (en) * | 2001-06-28 | 2004-08-11 | �ʼҷ����ֵ�������˾ | Wideband signal transmission system |
| CN101076853A (en) * | 2004-12-10 | 2007-11-21 | 松下电器产业株式会社 | Wide-band encoding device, wide-band lsp prediction device, band scalable encoding device, wide-band encoding method |
| CN1992533A (en) * | 2005-12-26 | 2007-07-04 | 索尼株式会社 | Signal encoding device and signal encoding method, signal decoding device and signal decoding method, program, and medium |
| US20080221905A1 (en) * | 2006-10-18 | 2008-09-11 | Markus Schnell | Encoding an Information Signal |
| WO2008101324A1 (en) * | 2007-02-23 | 2008-08-28 | Qnx Software Systems (Wavemakers), Inc. | High-frequency bandwidth extension in the time domain |
| CN101620854A (en) * | 2008-06-30 | 2010-01-06 | 华为技术有限公司 | Method, system and device for frequency band expansion |
| CN101751926A (en) * | 2008-12-10 | 2010-06-23 | 华为技术有限公司 | Signal coding and decoding method and device, and coding and decoding system |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015043161A1 (en) * | 2013-09-26 | 2015-04-02 | 华为技术有限公司 | Method and device for bandwidth extension |
| CN104517610A (en) * | 2013-09-26 | 2015-04-15 | 华为技术有限公司 | Band spreading method and apparatus |
| US9666201B2 (en) | 2013-09-26 | 2017-05-30 | Huawei Technologies Co., Ltd. | Bandwidth extension method and apparatus using high frequency excitation signal and high frequency energy |
| CN104517610B (en) * | 2013-09-26 | 2018-03-06 | 华为技术有限公司 | The method and device of bandspreading |
| US10186272B2 (en) | 2013-09-26 | 2019-01-22 | Huawei Technologies Co., Ltd. | Bandwidth extension with line spectral frequency parameters |
| CN109788922A (en) * | 2016-10-14 | 2019-05-21 | 公立大学法人大阪府立大学 | Swallow diagnostic device and program |
| US11246526B2 (en) | 2016-10-14 | 2022-02-15 | University Public Corporation Osaka | Swallowing diagnosis apparatus and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2011144130A1 (en) | 2011-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11580998B2 (en) | Method and device for encoding a high frequency signal, and method and device for decoding a high frequency signal | |
| EP3910630B1 (en) | Transient speech or audio signal encoding method and device, decoding method and device, processing system and computer-readable storage medium | |
| CN102177426B (en) | Multi-resolution switched audio encoding/decoding scheme | |
| AU2007206167B2 (en) | Apparatus and method for encoding and decoding signal | |
| CN108711431B (en) | Method and apparatus for concealing frame errors | |
| EP2791937B1 (en) | Generation of a high band extension of a bandwidth extended audio signal | |
| EP2770503B1 (en) | Method and apparatus for concealing frame errors and method and apparatus for audio decoding | |
| US20070219785A1 (en) | Speech post-processing using MDCT coefficients | |
| CN106128473A (en) | For producing equipment and the method for bandwidth expansion signal | |
| EP2774145B1 (en) | Improving non-speech content for low rate celp decoder | |
| CN105706166B (en) | Audio decoder apparatus and method for decoding a bitstream | |
| Bessette et al. | Universal speech/audio coding using hybrid ACELP/TCX techniques | |
| CN104025189A (en) | Method for encoding voice signal, method for decoding voice signal, and apparatus using same | |
| CN103493129B (en) | For using Transient detection and quality results by the apparatus and method of the code segment of audio signal | |
| CN104978970A (en) | Noise signal processing and generation method, encoder/decoder and encoding/decoding system | |
| EP2951816B1 (en) | Noise filling without side information for celp-like coders | |
| CN105304090A (en) | Apparatus and method for encoding and decoding an audio signal using an aligned look-ahead portion | |
| CN106233112A (en) | Coding method and equipment and signal decoding method and equipment | |
| CN102339607A (en) | Method and device for spreading frequency bands | |
| CN101622668B (en) | Methods and arrangements in a telecommunications network | |
| Zhao et al. | A CNN postprocessor to enhance coded speech | |
| Park et al. | Artificial bandwidth extension of narrowband speech signals for the improvement of perceptual speech communication quality | |
| Ali et al. | A long term harmonic plus noise model for narrow-band speech coding at very low bit-rates | |
| CN1327408C (en) | A low bit-rate speech encoder | |
| Choi et al. | Efficient harmonic-CELP based hybrid coding of speech at low bit rates. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20120201 |