[go: up one dir, main page]

WO2016082278A1 - 主成分分析pca映射模型的编解码方法及装置 - Google Patents

主成分分析pca映射模型的编解码方法及装置 Download PDF

Info

Publication number
WO2016082278A1
WO2016082278A1 PCT/CN2014/095393 CN2014095393W WO2016082278A1 WO 2016082278 A1 WO2016082278 A1 WO 2016082278A1 CN 2014095393 W CN2014095393 W CN 2014095393W WO 2016082278 A1 WO2016082278 A1 WO 2016082278A1
Authority
WO
WIPO (PCT)
Prior art keywords
frequency band
coefficient
vector
encoded
mapping
Prior art date
Application number
PCT/CN2014/095393
Other languages
English (en)
French (fr)
Inventor
吴超刚
潘兴德
Original Assignee
北京天籁传音数字技术有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 北京天籁传音数字技术有限公司 filed Critical 北京天籁传音数字技术有限公司
Publication of WO2016082278A1 publication Critical patent/WO2016082278A1/zh

Links

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
    • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
    • G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
    • G10L19/022—Blocking, i.e. grouping of samples in time; Choice of analysis windows; Overlap factoring
    • 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
    • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
    • G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
    • G10L19/032—Quantisation or dequantisation of spectral components

Definitions

  • the present invention relates to the field of audio processing technologies, and in particular, to a codec method and apparatus for a Principal Components Analysis (PCA) mapping model.
  • PCA Principal Components Analysis
  • mapping matrix performs quantization coding. Since the number of mapping matrices that need to be encoded is large, the coding rate of the PCA mapping model is too high.
  • the invention provides a codec method and device for a PCA mapping model, which effectively reduces the coding rate of the PCA mapping model.
  • the present invention provides a coding method of a PCA mapping model, the method comprising:
  • a first mapping matrix Determining, for each of the frequency band groups, a first mapping matrix, where the first mapping matrix is a mapping matrix of a set of PCA mapping models shared by each frequency band in the frequency band group;
  • the first mapping matrix is quantized and encoded.
  • the present invention provides a decoding method of a PCA mapping model, the method comprising:
  • mapping matrix is a mapping matrix determined by performing frequency band combining processing on each frequency band after frequency band division, and obtaining each frequency band group, and determining each frequency band group in each frequency band group.
  • the present invention provides an encoding apparatus for a PCA mapping model, the apparatus comprising:
  • a frequency band combining unit configured to perform frequency band combining processing on each frequency band after frequency band division, to obtain each frequency band group
  • a matrix determining unit configured to determine a first mapping matrix for each of the frequency band groups obtained by the frequency band combining unit, the first mapping matrix being a group of PCA mappings common to each frequency band in the frequency band group The mapping matrix of the model;
  • a coding unit configured to perform quantization coding on the first mapping matrix determined by the matrix determining unit.
  • the present invention provides a decoding apparatus for a PCA mapping model, the apparatus comprising:
  • a vector determining unit configured to determine a vector encoded in the mapped mapping matrix
  • a decoding unit configured to decode the encoded coefficients in the vector determined by the vector determining unit to obtain a reconstructed value of the coefficient
  • a vector reconstruction unit configured to reconstruct the vector according to a reconstructed value of a coefficient obtained by the decoding unit
  • a matrix reconstruction unit configured to reconstruct the mapping matrix according to the vector reconstructed by the vector reconstruction unit, where the mapping matrix performs frequency band combining processing on each frequency band divided by the frequency band, and after obtaining each frequency band group, is the frequency band group A mapping matrix determined for each band group in .
  • mapping matrix, the first mapping matrix is a mapping matrix of a set of PCA mapping models shared by each frequency band in the frequency band group, and then the first mapping matrix is quantized and encoded.
  • the embodiment of the present invention does not encode the mapping matrix corresponding to each frequency band after the frequency band division, but uses the frequency band combination processing to change the number of mapping matrices that need to be encoded from the original
  • the mapping matrix corresponding to each frequency band is reduced to a mapping matrix corresponding to each frequency band group, thereby effectively reducing the coding rate.
  • FIG. 1 is a flowchart of a coding method of a PCA mapping model according to an embodiment of the present invention
  • FIG. 2 is a schematic structural diagram of an apparatus for encoding a PCA mapping model according to another embodiment of the present invention.
  • FIG. 1 is a flowchart of a coding method of a PCA mapping model according to an embodiment of the present invention.
  • frequency band combination processing is performed on a mapping matrix of each frequency band after frequency band division, and then a coding matrix of a selected coding is quantized and encoded.
  • the method includes:
  • step 101 band combining processing is performed on each frequency band divided by the frequency band to obtain each frequency band group.
  • the frequency band combination processing may be performed on each frequency band after the frequency band division according to the characteristics of the frequency band signal and/or the psychoacoustic model and/or the model parameter similarity, and each frequency band group is obtained.
  • the frequency band combination processing may be specifically performed by using any one of the following methods or any combination of the following manners:
  • the first manner the energy of the adjacent two frequency bands is compared, when the energy of one frequency band is lower than
  • the second method is to calculate the masking threshold of a certain frequency band according to the psychoacoustic model, when When the energy of the frequency band is lower than the masking threshold, the frequency band is combined with the adjacent frequency band, and the two frequency bands are divided into one frequency band group.
  • the mapping matrix between two adjacent frequency bands is calculated. Distance, when the maximum distance is less than the distance threshold, The two or more frequency bands are combined to divide the two or more frequency bands into one frequency band group.
  • Step 102 Determine a first mapping matrix for each of the frequency band groups, where the first mapping matrix is a mapping matrix of a group of PCA mapping models shared by each frequency band in the frequency band group.
  • the first mapping matrix when the first mapping matrix is determined for the frequency band group, one mapping matrix may be selected as the first mapping matrix in the mapping matrix corresponding to each frequency band in the frequency band group, for example, a mapping matrix corresponding to the frequency band with the highest frequency band energy may be selected.
  • the mapping matrix As the first mapping matrix; the mapping matrix can also be recalculated for the band group.
  • a plurality of manners may be adopted to determine a first mapping matrix for each frequency band group.
  • Step 103 Perform quantization coding on the first mapping matrix.
  • the mapping matrix is composed of a series of coefficients. In order to further reduce the coding rate, in the embodiment of the present invention, all the coefficients in the first mapping matrix are not quantized, but the partial coefficients are selected according to the characteristics of the PCA mapping model. coding.
  • the number of packets to be encoded according to the dimension of the PCA analysis and the multi-channel sound signal may be selected, and coefficients that need to be encoded are selected from the first mapping matrix and quantized.
  • the vector to be encoded in the first mapping matrix may be determined according to the PCA packet number and the grouping condition selected to be encoded in the multi-channel sound signal; and the coefficients that need to be encoded in the vector are quantized and encoded.
  • the mapping matrix is composed of a series of coefficients
  • the number of packets to be encoded according to the dimension analyzed by the PCA and the multi-channel sound signal can be selected, and the coefficients to be encoded are selected from the mapping matrix and quantized.
  • the coefficients to be encoded are selected from the mapping matrix and quantized.
  • the relationship between the coefficients of the mapping matrix not all matrix coefficients need to be quantized, some do not need to be encoded, and can be calculated according to the already encoded coefficient values, and some only need to encode symbol bits.
  • mapping matrix W(t, k) is a 2*2 matrix with 4 coefficients, where t is the frame (or subframe) number and k is the frequency number.
  • W(t,k) can be expressed by the following formula:
  • W(t,k) is a unit orthogonal matrix that satisfies:
  • mapping matrix W(t, k) is a 4*4 matrix with 16 coefficients, and W(t, k) can be expressed by:
  • W(t,k) is a unit orthogonal matrix that satisfies:
  • a11, a12, a13, a14 need to be encoded. Because satisfied Therefore, three important coefficients can be selected from the four coefficients a11, a12, a13, and a14 for quantization coding, and the fourth coefficient is only coded or not coded, and the absolute value is obtained by the first three coefficients.
  • the basis for selection may be the absolute value of the coefficient or the positional relationship, and the like. For example, if a14 is selected to encode only the sign bit, and the remaining coefficients are quantized, the absolute value of a14 can be calculated by the formula. Calculated. For example, the coefficient with the largest absolute value is only coded, and the remaining coefficients are quantized. If the solution of W(t, k) ensures that the largest absolute value of each vector is positive or negative, then The coefficient with the largest absolute value is not encoded, but the remaining coefficients are quantized.
  • a coefficient can be selected from a11, a12, a13, a14 to perform only symbol bit coding or no coding, and the remaining three coefficients are quantized and encoded; and for A21, a22, a23, a24, two coefficients can be selected for quantization coding.
  • the other two coefficients are derived by using the above relational expression. For example, if a21 and a22 are selected for quantization coding, then a23 and a24 satisfy:
  • Selecting a coefficient from a11, a12, a13, a14, and a21, a22, a23, and a24, respectively, only performs symbol bit encoding, and quantizes the remaining six coefficients, such as selecting a14 and a24.
  • the absolute value of a14 is obtained by a11, a12, and a13
  • the absolute value of a24 is obtained by a21, a22, and a23.
  • W(t, k) is a matrix of M*M, having M*M coefficients
  • W(t, k) is a unit orthogonal matrix, which can specifically represent For the following formula:
  • the quantization coding of the coefficients of the mapping matrix W(t, k) may be performed by a scalar coding method or a vector coding method; the coefficients of W(t, k) may be directly encoded, It can be encoded in a transform form of W(t, k).
  • the step of performing quantization coding on the mapping matrix W(t, k) may include:
  • Step 1 Determine a vector to be encoded in the mapping matrix W(t, k) according to the PCA packet number M and the grouping of the multi-channel sound signal selected for encoding.
  • Step 2 pair vector The coefficients that need to be encoded are quantized and encoded.
  • step 2 the Dataposflag and aq are quantized and encoded.
  • performing quantization coding on the coefficients to be encoded in the vector may include: determining first location information and second location information according to the size relationship of each coefficient in the vector.
  • the first position information is used to indicate a position of a coefficient having a smallest absolute value
  • the second position information is used to indicate a position of a coefficient having a second smallest absolute value
  • a coefficient having a minimum absolute value in the vector an absolute value
  • the small coefficients, the first location information, and the second location information are quantized.
  • Step 1 right In a11, a12, and A21 in the quantization coding, and get the reconstructed value
  • the formula is as follows:
  • Step 3 solve the following equation group Get two sets of solutions
  • the foregoing process utilizes a coefficient vector of a mapping matrix. with They are all unit vectors and are orthogonal to each other. Due to the error in the quantization process, there may be no solution to the equation group or the quantization error of ⁇ a22, a23 ⁇ is large, which causes instability of the mapping matrix, so you can choose to use only Coefficient vector 1 and It is a property of unit vectors, and does not take advantage of the nature of vector orthogonality.
  • the specific coding steps are as follows:
  • Step 1 right In a11, a12, and A21, a22 in the quantization coding;
  • Step 1 according to In the magnitude relationship of each coefficient, the position information minindex11 and minindex12 are determined, wherein minindex11 is the position of the coefficient with the smallest absolute value, and minindex12 is the position of the coefficient with the second smallest absolute value; In the magnitude relationship of each coefficient, the position information minindex21 and minindex22 are determined, wherein minindex21 is the position of the coefficient with the smallest absolute value, and minindex22 is the position of the coefficient with the second smallest absolute value;
  • Step 2 encoding minindex11, minindex12, minindex21, and minindex22, and performing quantitative encoding on the minimum and minimum coefficients of the absolute value;
  • minindex11, minindex12, minindex21, and minindex22 may be combined by two or more to quantize, or Huffman coding isentropic coding method may be used to reduce the code rate.
  • the coding method of the PCA mapping model in the embodiment of the present invention first performs band combination processing on each frequency band after frequency band division, obtains each frequency band group, and then determines the frequency band for each frequency band group in each frequency band group.
  • a mapping matrix wherein the first mapping matrix is a mapping matrix of a set of PCA mapping models shared by each frequency band in the frequency band group, and then the first mapping matrix is quantized and encoded.
  • the mapping matrix corresponding to each frequency band after the frequency band division is not encoded, but the frequency band combination processing is performed, and the number of mapping matrices to be encoded is corresponding to each original frequency band.
  • the mapping matrix is reduced to the mapping matrix corresponding to each band group, thereby effectively reducing the coding rate.
  • the coding method of the PCA mapping model provided by the embodiment of the present invention is adapted.
  • the embodiment of the present invention further provides a decoding method of the PCA mapping model, and the decoding method may specifically include the following processing procedure:
  • Step one determining a coded vector in the mapped mapping matrix
  • Step two decoding the encoded coefficients in the vector to obtain a reconstructed value of the coefficients
  • Step three reconstructing the vector according to the reconstructed value of the coefficient
  • Step 4 reconstruct the mapping matrix according to the vector, where the mapping matrix performs frequency band combining processing on each frequency band divided by the frequency band, and after obtaining each frequency band group, determining for each frequency band group in each frequency band group Mapping matrix.
  • the decoding method may further include: decoding the encoding of the location identifier to obtain a location identifier, where the location The identifier is used to indicate the position of the encoded coefficient in the vector; the reconstructing the vector according to the reconstructed value of the coefficient may specifically include: reconstructing the vector according to the position identifier and the reconstructed value of the coefficient .
  • the coefficient when the number of the PCA packets is 3, the coefficient includes a coefficient having the smallest absolute value and a coefficient having a small absolute value in the vector, before the reconstructing the vector according to the reconstructed value of the coefficient,
  • the method may further include: decoding the encoding of the first location information and the encoding of the second location information to obtain first location information and second location information, where the first location information is used to indicate a location of a coefficient with a minimum absolute value, The second location information is used to indicate a location of a coefficient with a small absolute value;
  • the reconstructing the vector according to the reconstructed value of the coefficient may specifically include: reconstructing a value according to a coefficient with a minimum absolute value in the vector, a reconstructed value of a coefficient having a second smallest absolute value, the first position information, and the second position information determining a reconstructed value of a coefficient having the largest absolute value among the vectors; and a reconstructed value of a coefficient having the smallest absolute value in the vector Reconstruction
  • the decoding of the mapping matrix W(t, k) may include the following steps:
  • Step 1 on the vector The coefficients that need to be encoded are decoded
  • Step 2 according to the decoded vector Reconstruct the mapping matrix W(t, k).
  • Step 1 decoding the location identifiers Dataposflag and aq;
  • Step 2 according to Dataposflag and aq, determine If Dataposflag is 1, then otherwise
  • Step 3 Reconstruct W(t, k).
  • Step 1 decoding is obtained middle A21, sign13 and sign bit selectflag;
  • Step 3 solve the following equation group to get two sets of solutions
  • Step 5 get And reconstruct W(t,k).
  • Step 1 decoding is obtained middle middle And the sign bits sign13, s ign23;
  • Step 2 according to s ign13, sign23 operation with And reconstruct
  • Step 1 decoding to obtain minindex11, minindex12, minindex21, minindex22, and the reconstruction values aq11, aq12, aq21, aq22 of the absolute minimum and the second smallest;
  • Step 2 calculated according to the reconstruction values aq11, aq12, aq21, aq22 with The coefficients aq13, aq23 of the maximum value of each absolute value.
  • Step 3 reconstruct according to the decoded position information inindex11, minindex12, minindex21, minindex22 and aq11, aq12, aq21, aq22, aq13, aq23
  • FIG. 2 is a schematic structural diagram of an encoding apparatus of a PCA mapping model according to an embodiment of the present invention, where the apparatus includes:
  • the frequency band combining unit 201 is configured to perform band combining processing on each frequency band divided by the frequency band to obtain each frequency band group;
  • a matrix determining unit 202 configured to determine a first mapping matrix for each of the frequency band groups obtained by the frequency band combining unit 201, where the first mapping matrix is a group shared by each frequency band in the frequency band group a mapping matrix of the PCA mapping model;
  • the coding unit 203 is configured to perform quantization coding on the first mapping matrix determined by the matrix determining unit 202.
  • the frequency band combining unit 201 is specifically configured to perform frequency band combination processing on each frequency band after frequency band division according to characteristics of a frequency band signal and/or a psychoacoustic model and/or a model parameter similarity. Obtain each band group.
  • the frequency band combining unit 201 specifically includes:
  • a first frequency band combining subunit for comparing energy amounts of two adjacent frequency bands, and when the energy of one frequency band is lower than an energy threshold calculated according to energy of the adjacent frequency band, combining the two frequency bands, Frequency bands are divided into one frequency band;
  • a second frequency band combining subunit configured to calculate a masking threshold of a frequency band according to a psychoacoustic model, and when the energy of the frequency band is lower than a masking threshold, combining the frequency band with an adjacent frequency band, and dividing the two frequency bands into one Band group;
  • a third frequency band combining sub-unit configured to calculate a distance between mapping matrices of two adjacent frequency bands, and when the maximum distance is less than a distance threshold, combining the two or several frequency bands, the two or Several frequency bands are divided into one frequency band group.
  • the mapping matrix is composed of a series of coefficients
  • the encoding unit 203 is specifically configured to select a number of packets to be encoded according to the dimension analyzed by the PCA and the multi-channel sound signal, and select the coding required from the first mapping matrix.
  • the coefficients are quantized and encoded.
  • the coding unit 203 specifically includes:
  • a vector determining subunit configured to determine a vector to be encoded in the mapping matrix W(t, k) according to a PCA packet number M and a grouping condition selected to be encoded in the multichannel sound signal;
  • a coding subunit configured to perform quantization coding on the coefficients that need to be coded in the vector determined by the vector determining subunit.
  • the embodiment of the present invention further provides a decoding device of a PCA mapping model, where the device includes:
  • a vector determining unit configured to determine a vector encoded in the mapped mapping matrix
  • a decoding unit configured to decode the encoded coefficients in the vector determined by the vector determining unit to obtain a reconstructed value of the coefficient
  • a vector reconstruction unit configured to reconstruct the vector according to a reconstructed value of a coefficient obtained by the decoding unit
  • a matrix reconstruction unit configured to reconstruct the mapping matrix according to the vector reconstructed by the vector reconstruction unit, where the mapping matrix performs frequency band combining processing on each frequency band divided by the frequency band, and after obtaining each frequency band group, is the frequency band group A mapping matrix determined for each band group in .
  • the decoding unit is further configured to: when the number of PCA packets is 2, decode the encoding of the location identifier to obtain a location identifier before the vector reconstruction unit reconstructs the vector according to the reconstructed value of the coefficient
  • the location identifier is used to indicate a location of the encoded coefficient in the vector
  • the vector reconstruction unit is specifically configured to: reconstruct the vector according to the location identifier obtained by the decoding unit and the reconstructed value of the coefficient.
  • the decoding unit is further configured to: when the number of PCA packets is 3, the coefficient includes a coefficient with a minimum absolute value and a coefficient with a second smallest absolute value in the vector, Decoding the coefficient to reconstruct the vector, decoding the encoding of the first location information and the encoding of the second location information to obtain first location information and second location information, where the first location information is used to indicate that the absolute value is minimum The position of the coefficient, the second position information is used to indicate the position of the coefficient with the second smallest absolute value;
  • the vector reconstruction unit is specifically configured to: reconstruct a value of a coefficient with a minimum absolute value among the vectors obtained by the decoding unit, a reconstruction value of a coefficient with a second smallest absolute value, the first location information, and the second location information Determining a reconstructed value of a coefficient having the largest absolute value in the vector; a reconstructed value of a coefficient having a smallest absolute value in the vector, a reconstructed value of a coefficient having a second smallest absolute value, and a reconstructed value of a coefficient having the largest absolute value in the vector And the first location information and the second location information reconstruct the vector.
  • the steps of a method or algorithm described in connection with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both.
  • the software module can be placed in random access memory (RAM), memory, read only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or technical field. Any other form of storage medium known.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Computational Linguistics (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)

Abstract

一种主成分分析PCA映射模型的编解码方法及装置，该编码方法包括：对频带划分后的各频带进行频带组合处理,获得各频带组（101）；为所述各频带组中的每个频带组确定第一映射矩阵，所述第一映射矩阵为所述频带组中的各频带共用的一组PCA映射模型的映射矩阵（102）；对所述第一映射矩阵进行量化编码（103）。由上可见，本实施例在对PCA映射模型进行编码时，不是对频带划分后的每个频带对应的映射矩阵进行编码，而是通过频带组合处理，将需要编码的映射矩阵数量由原来的每个频带对应的映射矩阵减少为每个频带组对应的映射矩阵，从而有效地降低了编码码率。

Description

主成分分析PCA映射模型的编解码方法及装置 技术领域
本发明涉及音频处理技术领域，尤其涉及主成分分析(PCA，Principal Components Analysis)映射模型的编解码方法及装置。
背景技术
随着科技的发展，出现了多种对声音信号的编码技术，上述声音通常指的是语音、音乐、自然声音和人工合成声音等人耳可感知的信号在内的数字声音。其中，在对多声道声音信号进行编码时，通常会涉及到PCA映射模型的编码。
现有技术中，在对多声道声音信号进行编码时，先要对多声道声音信号进行频带划分，相应地，在对PCA映射模型进行编码时，要对划分后的每个频带对应的映射矩阵进行量化编码，由于需要编码的映射矩阵数量较多，因此导致PCA映射模型的编码码率过高。
发明内容
本发明提供了一种PCA映射模型的编解码方法及装置，有效降低了PCA映射模型的编码码率。
为实现上述目的，第一方面，本发明提供了一种PCA映射模型的编码方法，所述方法包括：
对频带划分后的各频带进行频带组合处理,获得各频带组；
为所述各频带组中的每个频带组确定第一映射矩阵，所述第一映射矩阵为所述频带组中的各频带共用的一组PCA映射模型的映射矩阵；
对所述第一映射矩阵进行量化编码。
第二方面，本发明提供了一种PCA映射模型的解码方法，所述方法包括：
确定被编码的映射矩阵中被编码的矢量；
对所述矢量中的被编码的系数进行解码获得所述系数的重建值；
根据所述系数的重建值重建所述矢量；
根据所述矢量重建所述映射矩阵，所述映射矩阵为对频带划分后的各频带进行频带组合处理,获得各频带组后，为所述各频带组中的每个频带组确定的映射矩阵。
第三方面，本发明提供了一种PCA映射模型的编码装置，所述装置包括：
频带组合单元，用于对频带划分后的各频带进行频带组合处理,获得各频带组；
矩阵确定单元，用于为所述频带组合单元获得的各频带组中的每个频带组确定第一映射矩阵，所述第一映射矩阵为所述频带组中的各频带共用的一组PCA映射模型的映射矩阵；
编码单元，用于对所述矩阵确定单元确定的第一映射矩阵进行量化编码。
第四方面，本发明提供了一种PCA映射模型的解码装置，所述装置包括：
矢量确定单元，用于确定被编码的映射矩阵中被编码的矢量；
解码单元，用于对所述矢量确定单元确定的矢量中的被编码的系数进行解码获得所述系数的重建值；
矢量重建单元，用于根据所述解码单元获得的系数的重建值重建所述矢量；
矩阵重建单元，用于根据所述矢量重建单元重建的矢量重建所述映射矩阵，所述映射矩阵为对频带划分后的各频带进行频带组合处理,获得各频带组后，为所述各频带组中的每个频带组确定的映射矩阵。
本发明实施例的PCA映射模型的编码方法，先要对频带划分后的各频带进行频带组合处理,获得各频带组，然后为各频带组中的每个频带组确定第一 映射矩阵，第一映射矩阵为频带组中的各频带共用的一组PCA映射模型的映射矩阵，再对第一映射矩阵进行量化编码。由上可见，本发明实施例在对PCA映射模型进行编码时，不是对频带划分后的每个频带对应的映射矩阵进行编码，而是通过频带组合处理，将需要编码的映射矩阵数量由原来的每个频带对应的映射矩阵减少为每个频带组对应的映射矩阵，从而有效地降低了编码码率。
附图说明
图1为本发明一个实施例中PCA映射模型的编码方法流程图；
图2为本发明另一个实施例中PCA映射模型的编码装置结构示意图。
具体实施方式
下面通过附图和实施例，对本发明的技术方案做进一步的详细描述。
图1为本发明一个实施例中PCA映射模型的编码方法流程图，该实施例中首先对频带划分后的各频带的映射矩阵进行频带组合处理，然后对选定编码的映射矩阵进行量化编码，该方法包括：
步骤101，对频带划分后的各频带进行频带组合处理,获得各频带组。
其中，可以根据频带信号的特点和/或心理声学模型和/或模型参数相似度对频带划分后的各频带进行频带组合处理,获得各频带组。
本发明实施例中，可以具体采用下述任意一种方式或者下述方式的任意组合来进行频带组合处理：第一种方式，比较相邻两个频带的能量大小，当一个频带的能量低于根据相邻频带能量计算得出的能量阈值时，对这两个频带进行组合，将这两个频带划分到一个频带组；第二种方式，根据心理声学模型计算某一频带的掩蔽阈值，当频带的能量低于掩蔽阈值时，将这一频带与相邻频带进行组合，将这两个频带划分到一个频带组；第三种方式，计算相邻两个或几个频带的映射矩阵之间的距离，当最大距离小于距离阈值时，将 这两个或几个频带进行组合，将这两个或几个频带划分到一个频带组。
步骤102，为所述各频带组中的每个频带组确定第一映射矩阵，所述第一映射矩阵为所述频带组中的各频带共用的一组PCA映射模型的映射矩阵。
其中，为频带组确定第一映射矩阵时，可以在该频带组中的每个频带对应的映射矩阵中选取一个映射矩阵作为第一映射矩阵，例如，可以选取频带能量最高的频带对应的映射矩阵作为第一映射矩阵；也可以针对该频带组重新计算获得映射矩阵。本发明实施例中，可以采取多种方式来为每个频带组确定第一映射矩阵。
步骤103，对所述第一映射矩阵进行量化编码。
其中，映射矩阵由一系列系数组成，为了进一步降低编码码率，本发明实施例中不对第一映射矩阵中的所有系数进行量化编码，而是根据PCA映射模型的特点，从中选取部分系数进行量化编码。
具体地，可以根据PCA分析的维度和多声道声音信号进行编码的分组个数，从所述第一映射矩阵中选择需要进行编码的系数并进行量化编码。
进一步地，可以根据PCA分组数和多声道声音信号中被选择进行编码的分组情况，确定第一映射矩阵中需要被编码的矢量；对矢量中需要编码的系数进行量化编码。
下面针对映射矩阵的量化编码进行详细说明。
由于映射矩阵由一系列系数组成，本发明实施例中可以根据PCA分析的维度和多声道声音信号进行编码的分组个数，从映射矩阵中选择需要进行编码的系数并进行量化编码。根据映射矩阵的系数之间的关系可知，并不是所有的矩阵系数都需要进行量化编码，有些不需要进行编码，可以根据已经编码的系数值运算得到，有些则只需要编码符号位。通过对系数进行组织和选择，可以达到降低编码码率的目的。
当对两个声道信号进行PCA分析时，映射矩阵W(t,k)是2*2的矩阵，有4个系数，其中，t为帧(或子帧)序号，k为频率序号。
W(t,k)可以通过下式来表示：
Figure PCTCN2014095393-appb-000001
W(t,k)为单位正交矩阵，满足：
Figure PCTCN2014095393-appb-000002
因此，W(t,k)可表达为
Figure PCTCN2014095393-appb-000003
由上可见，只需要对β或其转换形式如cosβ或sinβ等进行编码。
当对四个声道信号进行PCA分析时，映射矩阵W(t,k)是4*4的矩阵，有16个系数，W(t,k)可以通过下式来表示：
Figure PCTCN2014095393-appb-000004
W(t,k)为单位正交矩阵，满足：
Figure PCTCN2014095393-appb-000005
当只对多声道声音信号中的第一主成分进行编码时，只需要对a11、a12、a13、a14进行编码。因为满足
Figure PCTCN2014095393-appb-000006
所以可从a11、a12、a13、a14四个系数中选择三个重要的系数进行量化编码，而第四个系数仅进行符号位编码或不编码，其绝对值由前三个系数求解得到。选择的依据可以是系数的绝对值大小或位置关系等。比如选择a14仅编码符号位，而其余系数进行量化编码，a14的绝对值可由公式
Figure PCTCN2014095393-appb-000007
计算得到。比如选择绝对值最大的系数只进行符号编码，而其余系数进行量化编码；若W(t,k)的求解过程中保证了每个向量中绝对值最大的系数是正值或负值，则对绝对值最大的 系数不进行编码，而对其余系数进行量化编码。
当对第二多声道声音信号中的第一、第二主成分进行编码时，则需要对a11、a12、a13、a14、a21、a22、a23、a24进行编码。因为满足
Figure PCTCN2014095393-appb-000008
所以有
Figure PCTCN2014095393-appb-000009
因此可从a11、a12、a13、a14中选择一个系数仅仅进行符号位编码或不编码，而对其余3个系数进行量化编码；而对于a21、a22、a23、a24可选择2个系数进行量化编码，另外2个系数则利用上述关系式推导得出，比如选择a21、a22进行量化编码，则a23、a24满足：
Figure PCTCN2014095393-appb-000010
求解此方程式可得一组解或两组解
Figure PCTCN2014095393-appb-000011
当得到两组解时，则需要判断哪组解符合原数据a23、a24，若
Figure PCTCN2014095393-appb-000012
符合，则令selectflag＝0；否则，令selectflag＝1，Selectflag也需要进行编码。由于系数量化过程中存在误差，有时会使得上述方程式组无解，或者求解得到的
Figure PCTCN2014095393-appb-000013
与原始数据存在较大误差。此时，可不利用条件
Figure PCTCN2014095393-appb-000014
而仅仅利用
Figure PCTCN2014095393-appb-000015
从a11、a12、a13、a14及a21、a22、a23、a24中分别选择一个系数仅仅进行符号位编码，而对其余6个系数进行量化编码，比如选择对a14和a24 仅进行符号位编码时，a14绝对值由a11、a12、a13求解得到，a24绝对值由a21、a22、a23求解得到。
普遍地，当对M个声道信号进行PCA分析时，W(t,k)是M*M的矩阵，有M*M个系数，W(t,k)为单位正交矩阵，具体可以表示为下式：
Figure PCTCN2014095393-appb-000016
当对多声道声音信号中的u组信号进行编码时，则只需对
Figure PCTCN2014095393-appb-000017
的系数进行编码。利用矩阵系数满足的条件：
Figure PCTCN2014095393-appb-000018
降低所需要量化的系数的个数，对一些系数仅编码符号位或不编码。
本发明实施例中，对映射矩阵W(t,k)的系数进行量化编码可以采用标量编码方式，也可以采用矢量编码的方法；既可以直接对W(t,k)的系数进行编码，也可在W(t,k)的某变换形式进行编码。
对映射矩阵W(t,k)进行量化编码的步骤可以包括：
步骤1，根据PCA分组数M和多声道声音信号中被选择进行编码的分组情况，确定映射矩阵W(t,k)中需要被编码的矢量
Figure PCTCN2014095393-appb-000019
步骤2，对矢量
Figure PCTCN2014095393-appb-000020
中需要编码的系数进行量化编码。
本发明实施例中，当PCA分组数为2时，编码方法还可以包括：确定位置标识，所述位置标识用于指示所述第一系数；在所述对所述第一系数进行量化编码时，对所述位置标识进行量化编码。例如，当PCA分组数M＝2时，选择对矢量
Figure PCTCN2014095393-appb-000021
进行编码，具体步骤如下：
步骤1，确定位置标识Dataposflag，如果a11绝对值小于a12的绝对值，则Dataposflag为1，待量化数据aq＝a11，否则Dataposflag为0,待量化数据aq＝a12；
步骤2，对Dataposflag和aq进行量化编码。
本发明实施例中，当PCA分组数为3时，对矢量中需要编码的系数进行量化编码，具体可以包括：根据所述矢量中各系数的大小关系，确定第一位置信息和第二位置信息，所述第一位置信息用于指示绝对值最小的系数的位置，所述第二位置信息用于指示绝对值次小的系数的位置；对所述矢量中绝对值最小的系数、绝对值次小的系数、所述第一位置信息和所述第二位置信息进行量化编码。
例如，当PCA分组数M＝3、多声道声音信号中第一和第二主成分被选择编码时，具体步骤如下：
步骤1，对
Figure PCTCN2014095393-appb-000022
中的a11、a12，以及
Figure PCTCN2014095393-appb-000023
中的a21进行量化编码，并得到重建值
Figure PCTCN2014095393-appb-000024
步骤2，对a13的符号位sign13进行编码，并计算得到a13的重建值
Figure PCTCN2014095393-appb-000025
如果a13是正数，则sign13＝1，否则sign13＝0；
Figure PCTCN2014095393-appb-000026
的计算公式如下：
Figure PCTCN2014095393-appb-000027
步骤3，求解如下方程式组
Figure PCTCN2014095393-appb-000028
得到两组解
Figure PCTCN2014095393-appb-000029
步骤4，比较{a22、a23}与
Figure PCTCN2014095393-appb-000030
如果
Figure PCTCN2014095393-appb-000031
更接近{a22、a23},则selectflag＝0；否则，selectflag＝1。对selectflag进行编码。
本发明实施例中，上述过程利用了映射矩阵的系数矢量
Figure PCTCN2014095393-appb-000032
和
Figure PCTCN2014095393-appb-000033
都是单位矢量且相互正交的性质，由于量化过程存在误差，可能会使得方程式组无解或者使得{a22、a23}量化误差很大，带来映射矩阵不稳定等问题，因此可以 选择只利用系数矢量
Figure PCTCN2014095393-appb-000034
1和
Figure PCTCN2014095393-appb-000035
都是单位矢量这一性质，而不利用矢量相互正交的性质，此时具体编码步骤如下：
步骤1，对
Figure PCTCN2014095393-appb-000036
中的a11、a12，以及
Figure PCTCN2014095393-appb-000037
中的a21、a22进行量化编码；
步骤2，对a13和a23的符号位s ign13、s ign23进行编码，如果a13是正数，则s ign13＝1，否则s ign13＝0；如果a23是正数，则s ign23＝1，否则s ign23＝0。
当PCA分组数M＝3、多声道声音信号中第一和第二主成分被选择编码时，选择对映射矩阵W(t,k)中的矢量
Figure PCTCN2014095393-appb-000038
和
Figure PCTCN2014095393-appb-000039
进行编码，还可以先对系数进行排序然后再进行量化，具体步骤如下：
步骤1，根据
Figure PCTCN2014095393-appb-000040
中各系数的大小关系，确定位置信息minindex11和minindex12,其中，minindex11是绝对值最小的系数的位置，minindex12是绝对值次小的系数的位置；根据
Figure PCTCN2014095393-appb-000041
中各系数的大小关系，确定位置信息minindex21和minindex22，其中，minindex21是绝对值最小的系数的位置，minindex22是绝对值次小的系数的位置；
步骤2，对minindex11、minindex12、minindex21和minindex22进行编码，并对绝对值最小和次小的系数进行量化编码；
其中，为提高编码效率，minindex11、minindex12、minindex21和minindex22可以两个或多个组合在一起进行量化，也可以采用哈夫曼编码等熵编码方法来降低码率。
此时，也可以进一步利用
Figure PCTCN2014095393-appb-000042
和
Figure PCTCN2014095393-appb-000043
正交的性质来进一步减少需要量化编码的系数的个数，具体过程与前述按照原始顺序进行编码的过程中类似，在此不再赘述。
由上述处理过程可知，本发明实施例的PCA映射模型的编码方法，先要对频带划分后的各频带进行频带组合处理,获得各频带组，然后为各频带组中的每个频带组确定第一映射矩阵，第一映射矩阵为频带组中的各频带共用的一组PCA映射模型的映射矩阵，再对第一映射矩阵进行量化编码。由上可见， 本发明实施例在对PCA映射模型进行编码时，不是对频带划分后的每个频带对应的映射矩阵进行编码，而是通过频带组合处理，将需要编码的映射矩阵数量由原来的每个频带对应的映射矩阵减少为每个频带组对应的映射矩阵，从而有效地降低了编码码率。
与本发明实施例提供的PCA映射模型的编码方法向适应，本发明实施例还提供了PCA映射模型的解码方法，该解码方法具体可以包括下述处理过程：
步骤一，确定被编码的映射矩阵中被编码的矢量；
步骤二，对所述矢量中的被编码的系数进行解码获得所述系数的重建值；
步骤三，根据所述系数的重建值重建所述矢量；
步骤四，根据所述矢量重建所述映射矩阵，所述映射矩阵为对频带划分后的各频带进行频带组合处理,获得各频带组后，为所述各频带组中的每个频带组确定的映射矩阵。
优选地，当所述PCA分组数为2时，所述根据所述系数的重建值重建所述矢量之前，所述解码方法还可以包括：对位置标识的编码进行解码获得位置标识，所述位置标识用于指示被编码的系数在所述矢量中的位置；所述根据所述系数的重建值重建所述矢量，具体可以包括：根据所述位置标识和所述系数的重建值重建所述矢量。
优选地，当所述PCA分组数为3，所述系数包括所述矢量中绝对值最小的系数和绝对值次小的系数时，所述根据所述系数的重建值重建所述矢量之前，所述方法还可以包括：对第一位置信息的编码和第二位置信息的编码进行解码获得第一位置信息和第二位置信息，所述第一位置信息用于指示绝对值最小的系数的位置，所述第二位置信息用于指示绝对值次小的系数的位置；所述根据所述系数的重建值重建所述矢量，具体可以包括：根据所述矢量中绝对值最小的系数的重建值、绝对值次小的系数的重建值、所述第一位置信息和所述第二位置信息确定所述矢量中绝对值最大的系数的重建值；根据所述矢量中绝对值最小的系数的重建值、绝对值次小的系数的重建值、所述矢量 中绝对值最大的系数的重建值、所述第一位置信息和所述第二位置信息重建所述矢量。
具体地，映射矩阵W(t,k)的解码可以包括如下步骤：
步骤1，对矢量
Figure PCTCN2014095393-appb-000044
中需要编码的系数进行解码；
步骤2，根据解码的矢量
Figure PCTCN2014095393-appb-000045
重构映射矩阵W(t,k)。
例如，当PCA分组数M＝2时，具体解码步骤如下：
步骤1，对位置标识Dataposflag和aq进行解码；
步骤2，根据Dataposflag和aq，确定
Figure PCTCN2014095393-appb-000046
若Dataposflag为1，则
Figure PCTCN2014095393-appb-000047
否则
Figure PCTCN2014095393-appb-000048
步骤3，重构W(t,k)。
Figure PCTCN2014095393-appb-000049
当PCA分组数M＝3，多声道声音信号中第一和第二主成分被选择编码时，解码具体步骤如下：
步骤1，解码得到
Figure PCTCN2014095393-appb-000050
中的
Figure PCTCN2014095393-appb-000051
中的a21，sign13以及符号位selectflag；
步骤2，根据
Figure PCTCN2014095393-appb-000052
sign13运算得到
Figure PCTCN2014095393-appb-000053
Figure PCTCN2014095393-appb-000054
步骤3，求解如下方程式组得到两组解
Figure PCTCN2014095393-appb-000055
Figure PCTCN2014095393-appb-000056
步骤4，如果selectflag＝1，则用
Figure PCTCN2014095393-appb-000057
替换
Figure PCTCN2014095393-appb-000058
Figure PCTCN2014095393-appb-000059
步骤5，得到
Figure PCTCN2014095393-appb-000060
并重构W(t,k)。
Figure PCTCN2014095393-appb-000061
Figure PCTCN2014095393-appb-000062
Figure PCTCN2014095393-appb-000063
当选择只利用系数矢量
Figure PCTCN2014095393-appb-000064
和
Figure PCTCN2014095393-appb-000065
都是单位矢量这一性质，而不利用矢量相互正交的性质进行量化编码时，对应的解码具体步骤如下：
步骤1，解码得到
Figure PCTCN2014095393-appb-000066
中的
Figure PCTCN2014095393-appb-000067
中的
Figure PCTCN2014095393-appb-000068
以及符号位sign13、s ign23；
步骤2，根据
Figure PCTCN2014095393-appb-000069
s ign13、sign23运算得到
Figure PCTCN2014095393-appb-000070
和
Figure PCTCN2014095393-appb-000071
并重构
Figure PCTCN2014095393-appb-000072
Figure PCTCN2014095393-appb-000073
Figure PCTCN2014095393-appb-000074
Figure PCTCN2014095393-appb-000075
步骤3，根据
Figure PCTCN2014095393-appb-000077
重构W(t,k)；
Figure PCTCN2014095393-appb-000078
当选择先对系数进行排序然后量化编码的方法时，对应的解码具体步骤 如下：
步骤1，解码得到minindex11、minindex12、minindex21、minindex22，以及绝对值最小和次小的系数的重建值aq11、aq12、aq21、aq22；
步骤2，根据重建值aq11、aq12、aq21、aq22，计算
Figure PCTCN2014095393-appb-000079
和
Figure PCTCN2014095393-appb-000080
中各自的绝对值最大值的系数aq13、aq23。
Figure PCTCN2014095393-appb-000081
Figure PCTCN2014095393-appb-000082
步骤3，根据解码得到的位置信息inindex11、minindex12、minindex21、minindex22及aq11、aq12、aq21、aq22、aq13、aq23来重构
Figure PCTCN2014095393-appb-000083
Figure PCTCN2014095393-appb-000084
Figure PCTCN2014095393-appb-000085
步骤4，根据
Figure PCTCN2014095393-appb-000086
重构W(t,k)。
Figure PCTCN2014095393-appb-000087
图2为本发明一个实施例中PCA映射模型的编码装置结构示意图，该装置包括：
频带组合单元201，用于对频带划分后的各频带进行频带组合处理,获得各频带组；
矩阵确定单元202，用于为所述频带组合单元201获得的各频带组中的每个频带组确定第一映射矩阵，所述第一映射矩阵为所述频带组中的各频带共用的一组PCA映射模型的映射矩阵；
编码单元203，用于对所述矩阵确定单元202确定的第一映射矩阵进行量化编码。
优选地，所述频带组合单元201，具体用于根据频带信号的特点和/或心理声学模型和/或模型参数相似度对频带划分后的各频带进行频带组合处理, 获得各频带组。
优选地，所述频带组合单元201具体包括：
第一频带组合子单元，用于比较相邻两个频带的能量大小，当一个频带的能量低于根据相邻频带能量计算得出的能量阈值时，对这两个频带进行组合，将这两个频带划分到一个频带组；和/或
第二频带组合子单元，用于根据心理声学模型计算某一频带的掩蔽阈值，当频带的能量低于掩蔽阈值时，将这一频带与相邻频带进行组合，将这两个频带划分到一个频带组；和/或
第三频带组合子单元，用于计算相邻两个或几个频带的映射矩阵之间的距离，当最大距离小于距离阈值时，将这两个或几个频带进行组合，将这两个或几个频带划分到一个频带组。
优选地，映射矩阵由一系列系数组成，所述编码单元203，具体用于根据PCA分析的维度和多声道声音信号进行编码的分组个数，从所述第一映射矩阵中选择需要进行编码的系数并进行量化编码。
优选地，所述编码单元203具体包括：
矢量确定子单元，用于根据PCA分组数M和多声道声音信号中被选择进行编码的分组情况，确定映射矩阵W(t,k)中需要被编码的矢量；
编码子单元，用于对所述矢量确定子单元确定的矢量中需要编码的系数进行量化编码。
相应地，本发明实施例还提供了PCA映射模型的解码装置，所述装置包括：
矢量确定单元，用于确定被编码的映射矩阵中被编码的矢量；
解码单元，用于对所述矢量确定单元确定的矢量中的被编码的系数进行解码获得所述系数的重建值；
矢量重建单元，用于根据所述解码单元获得的系数的重建值重建所述矢量；
矩阵重建单元，用于根据所述矢量重建单元重建的矢量重建所述映射矩阵，所述映射矩阵为对频带划分后的各频带进行频带组合处理,获得各频带组后，为所述各频带组中的每个频带组确定的映射矩阵。
优选地，所述解码单元还用于：当所述PCA分组数为2时，在所述矢量重建单元根据所述系数的重建值重建所述矢量之前，对位置标识的编码进行解码获得位置标识，所述位置标识用于指示被编码的系数在所述矢量中的位置；
所述矢量重建单元具体用于：根据所述解码单元获得的位置标识和所述系数的重建值重建所述矢量。
优选地，所述解码单元还用于：当所述PCA分组数为3，所述系数包括所述矢量中绝对值最小的系数和绝对值次小的系数时，在所述矢量重建单元根据所述系数的重建值重建所述矢量之前，对第一位置信息的编码和第二位置信息的编码进行解码获得第一位置信息和第二位置信息，所述第一位置信息用于指示绝对值最小的系数的位置，所述第二位置信息用于指示绝对值次小的系数的位置；
所述矢量重建单元具体用于：根据所述解码单元获得的矢量中绝对值最小的系数的重建值、绝对值次小的系数的重建值、所述第一位置信息和所述第二位置信息确定所述矢量中绝对值最大的系数的重建值；根据所述矢量中绝对值最小的系数的重建值、绝对值次小的系数的重建值、所述矢量中绝对值最大的系数的重建值、所述第一位置信息和所述第二位置信息重建所述矢量。
专业人员应该还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每 个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
结合本文中所公开的实施例描述的方法或算法的步骤可以用硬件、处理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
以上所述的具体实施方式，对本发明的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施方式而已，并不用于限定本发明的保护范围，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (15)

  1. 一种主成分分析PCA映射模型的编码方法，其特征在于，所述方法包括：
    对频带划分后的各频带进行频带组合处理,获得各频带组；
    为所述各频带组中的每个频带组确定第一映射矩阵，所述第一映射矩阵为所述频带组中的各频带共用的一组PCA映射模型的映射矩阵；
    对所述第一映射矩阵进行量化编码。
  2. 如权利要求1所述的方法，其特征在于，所述对频带划分后的各频带进行频带组合处理,获得各频带组，具体包括：
    根据频带信号的特点和/或心理声学模型和/或模型参数相似度对频带划分后的各频带进行频带组合处理,获得各频带组。
  3. 如权利要求1所述的方法，其特征在于，所述对频带划分后的各频带进行频带组合处理,获得各频带组，具体包括:
    比较相邻两个频带的能量大小，当一个频带的能量低于根据相邻频带能量计算得出的能量阈值时，对这两个频带进行组合，将这两个频带划分到一个频带组；和/或
    根据心理声学模型计算某一频带的掩蔽阈值，当频带的能量低于掩蔽阈值时，将这一频带与相邻频带进行组合，将这两个频带划分到一个频带组；和/或
    计算相邻两个或几个频带的映射矩阵之间的距离，当最大距离小于距离阈值时，将这两个或几个频带进行组合，将这两个或几个频带划分到一个频带组。
  4. 如权利要求1所述的方法，其特征在于，映射矩阵由一系列系数组成，所述对所述第一映射矩阵进行量化编码，具体包括:
    根据PCA分组数和多声道声音信号中被选择进行编码的分组情况，确定所述第一映射矩阵中需要被编码的矢量；
    对所述矢量中需要编码的系数进行量化编码。
  5. 如权利要求4所述的方法，其特征在于，所述对所述矢量中需要编码的系数进行量化编码，具体包括:
    根据所述第一映射矩阵为单位正交矩阵的性质或者根据所述第一映射矩阵为单位矩阵的性质，从所述矢量中选择第一系数，对所述第一系数进行量化编码，对所述矢量中其余的系数不编码或只进行符号位编码。
  6. 如权利要求5所述的方法，其特征在于，所述PCA分组数为2，所述方法还包括：
    确定位置标识，所述位置标识用于指示所述第一系数；
    在所述对所述第一系数进行量化编码时，对所述位置标识进行量化编码。
  7. 如权利要求4所述的方法，其特征在于，所述PCA分组数为3，所述对所述矢量中需要编码的系数进行量化编码，具体包括：
    根据所述矢量中各系数的大小关系，确定第一位置信息和第二位置信息，所述第一位置信息用于指示绝对值最小的系数的位置，所述第二位置信息用于指示绝对值次小的系数的位置；
    对所述矢量中绝对值最小的系数、绝对值次小的系数、所述第一位置信息和所述第二位置信息进行量化编码。
  8. 一种主成分分析PCA映射模型的解码方法，其特征在于，所述方法包括：
    确定被编码的映射矩阵中被编码的矢量；
    对所述矢量中的被编码的系数进行解码获得所述系数的重建值；
    根据所述系数的重建值重建所述矢量；
    根据所述矢量重建所述映射矩阵，所述映射矩阵为对频带划分后的各频带进行频带组合处理,获得各频带组后，为所述各频带组中的每个频带组确定的映射矩阵。
  9. 如权利要求8所述的方法，其特征在于，所述PCA分组数为2，所述 根据所述系数的重建值重建所述矢量之前，所述方法还包括：
    对位置标识的编码进行解码获得位置标识，所述位置标识用于指示被编码的系数在所述矢量中的位置；
    所述根据所述系数的重建值重建所述矢量，具体包括：根据所述位置标识和所述系数的重建值重建所述矢量。
  10. 如权利要求8所述的方法，其特征在于，所述PCA分组数为3，所述系数包括所述矢量中绝对值最小的系数和绝对值次小的系数，所述根据所述系数的重建值重建所述矢量之前，所述方法还包括：
    对第一位置信息的编码和第二位置信息的编码进行解码获得第一位置信息和第二位置信息，所述第一位置信息用于指示绝对值最小的系数的位置，所述第二位置信息用于指示绝对值次小的系数的位置；
    所述根据所述系数的重建值重建所述矢量，具体包括：
    根据所述矢量中绝对值最小的系数的重建值、绝对值次小的系数的重建值、所述第一位置信息和所述第二位置信息确定所述矢量中绝对值最大的系数的重建值；
    根据所述矢量中绝对值最小的系数的重建值、绝对值次小的系数的重建值、所述矢量中绝对值最大的系数的重建值、所述第一位置信息和所述第二位置信息重建所述矢量。
  11. 一种主成分分析PCA映射模型的编码装置，其特征在于，所述装置包括：
    频带组合单元，用于对频带划分后的各频带进行频带组合处理,获得各频带组；
    矩阵确定单元，用于为所述频带组合单元获得的各频带组中的每个频带组确定第一映射矩阵，所述第一映射矩阵为所述频带组中的各频带共用的一组PCA映射模型的映射矩阵；
    编码单元，用于对所述矩阵确定单元确定的第一映射矩阵进行量化编码。
  12. 如权利要求11所述的装置，其特征在于，所述频带组合单元，具体用于根据频带信号的特点和/或心理声学模型和/或模型参数相似度对频带划分后的各频带进行频带组合处理,获得各频带组。
  13. 如权利要求11所述的装置，其特征在于，所述频带组合单元具体包括:
    第一频带组合子单元，用于比较相邻两个频带的能量大小，当一个频带的能量低于根据相邻频带能量计算得出的能量阈值时，对这两个频带进行组合，将这两个频带划分到一个频带组；和/或
    第二频带组合子单元，用于根据心理声学模型计算某一频带的掩蔽阈值，当频带的能量低于掩蔽阈值时，将这一频带与相邻频带进行组合，将这两个频带划分到一个频带组；和/或
    第三频带组合子单元，用于计算相邻两个或几个频带的映射矩阵之间的距离，当最大距离小于距离阈值时，将这两个或几个频带进行组合，将这两个或几个频带划分到一个频带组。
  14. 如权利要求11所述的装置，其特征在于，映射矩阵由一系列系数组成，所述编码单元，具体用于根据PCA分组数和多声道声音信号中被选择进行编码的分组情况，确定所述第一映射矩阵中需要被编码的矢量；对所述矢量中需要编码的系数进行量化编码。
  15. 一种主成分分析PCA映射模型的解码装置，其特征在于，所述装置包括：
    矢量确定单元，用于确定被编码的映射矩阵中被编码的矢量；
    解码单元，用于对所述矢量确定单元确定的矢量中的被编码的系数进行解码获得所述系数的重建值；
    矢量重建单元，用于根据所述解码单元获得的系数的重建值重建所述矢量；
    矩阵重建单元，用于根据所述矢量重建单元重建的矢量重建所述映射矩 阵，所述映射矩阵为对频带划分后的各频带进行频带组合处理,获得各频带组后，为所述各频带组中的每个频带组确定的映射矩阵。
PCT/CN2014/095393 2014-11-28 2014-12-29 主成分分析pca映射模型的编解码方法及装置 WO2016082278A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201410710991.2 2014-11-28
CN201410710991.2A CN105632505B (zh) 2014-11-28 2014-11-28 主成分分析pca映射模型的编解码方法及装置

Publications (1)

Publication Number Publication Date
WO2016082278A1 true WO2016082278A1 (zh) 2016-06-02

Family

ID=56047346

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/095393 WO2016082278A1 (zh) 2014-11-28 2014-12-29 主成分分析pca映射模型的编解码方法及装置

Country Status (2)

Country Link
CN (1) CN105632505B (zh)
WO (1) WO2016082278A1 (zh)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1248824A (zh) * 1998-09-17 2000-03-29 松下电器产业株式会社 音频信号编码装置和方法以及解码装置和方法
CN101105940A (zh) * 2007-06-27 2008-01-16 北京中星微电子有限公司 音频编解码的量化方法、反变换方法及音频编解码装置
JP2008185845A (ja) * 2007-01-30 2008-08-14 National Institute Of Advanced Industrial & Technology １次元信号の変換値からのｈｌａｃ特徴抽出方法および装置
CN101401152A (zh) * 2006-03-15 2009-04-01 法国电信公司 通过多通道音频信号的主分量分析进行编码的设备和方法
CN102150207A (zh) * 2008-07-24 2011-08-10 Dts（英属维尔京群岛）有限公司 通过二维变换压缩音频比例因子
CN102682779A (zh) * 2012-06-06 2012-09-19 武汉大学 面向3d音频的双声道编解码方法和编解码器
CN103544957A (zh) * 2012-07-13 2014-01-29 华为技术有限公司 音频信号的比特分配的方法和装置

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TW411723B (en) * 1996-11-15 2000-11-11 Koninkl Philips Electronics Nv A mono-stereo conversion device, an audio reproduction system using such a device and a mono-stereo conversion method
FI115329B (fi) * 2000-05-08 2005-04-15 Nokia Corp Menetelmä ja järjestely lähdesignaalin kaistanleveyden vaihtamiseksi tietoliikenneyhteydessä, jossa on valmiudet useisiin kaistanleveyksiin
JP4805541B2 (ja) * 2002-04-10 2011-11-02 コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ ステレオ信号の符号化
BR0305434A (pt) * 2002-07-12 2004-09-28 Koninkl Philips Electronics Nv Métodos e arranjos para codificar e para decodificar um sinal de áudio multicanal, aparelhos para fornecer um sinal de áudio codificado e um sinal de áudio decodificado, sinal de áudio multicanal codificado, e, meio de armazenagem
CN1942929A (zh) * 2004-04-05 2007-04-04 皇家飞利浦电子股份有限公司 多信道编码器
US7860721B2 (en) * 2004-09-17 2010-12-28 Panasonic Corporation Audio encoding device, decoding device, and method capable of flexibly adjusting the optimal trade-off between a code rate and sound quality
BRPI0517987B1 (pt) * 2004-11-04 2021-04-27 Koninklijke Philips N. V. Dispositivo de codificação de canal de áudio, dispositivo de decodificação de canal de áudio, e método para converter um primeiro número de canais de áudio de entrada em um segundo número de canais de áudio de saída
JP4943418B2 (ja) * 2005-03-30 2012-05-30 コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ スケーラブルマルチチャネル音声符号化方法
CN100420155C (zh) * 2005-08-03 2008-09-17 上海杰得微电子有限公司 宽带音频压缩编码器的频带组划分方法
US7831434B2 (en) * 2006-01-20 2010-11-09 Microsoft Corporation Complex-transform channel coding with extended-band frequency coding
FR2898725A1 (fr) * 2006-03-15 2007-09-21 France Telecom Dispositif et procede de codage gradue d'un signal audio multi-canal selon une analyse en composante principale
WO2009125046A1 (en) * 2008-04-11 2009-10-15 Nokia Corporation Processing of signals
US8452587B2 (en) * 2008-05-30 2013-05-28 Panasonic Corporation Encoder, decoder, and the methods therefor

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1248824A (zh) * 1998-09-17 2000-03-29 松下电器产业株式会社 音频信号编码装置和方法以及解码装置和方法
CN101401152A (zh) * 2006-03-15 2009-04-01 法国电信公司 通过多通道音频信号的主分量分析进行编码的设备和方法
JP2008185845A (ja) * 2007-01-30 2008-08-14 National Institute Of Advanced Industrial & Technology １次元信号の変換値からのｈｌａｃ特徴抽出方法および装置
CN101105940A (zh) * 2007-06-27 2008-01-16 北京中星微电子有限公司 音频编解码的量化方法、反变换方法及音频编解码装置
CN102150207A (zh) * 2008-07-24 2011-08-10 Dts（英属维尔京群岛）有限公司 通过二维变换压缩音频比例因子
CN102682779A (zh) * 2012-06-06 2012-09-19 武汉大学 面向3d音频的双声道编解码方法和编解码器
CN103544957A (zh) * 2012-07-13 2014-01-29 华为技术有限公司 音频信号的比特分配的方法和装置

Also Published As

Publication number Publication date
CN105632505A (zh) 2016-06-01
CN105632505B (zh) 2019-12-20

Similar Documents

Publication Publication Date Title
ES2689566T3 (es) Compresión de las representaciones descompuestas de un campo sonoro
JP2019529979A (ja) インデックスコーディング及びビットスケジューリングを備えた量子化器
JP7257965B2 (ja) デジタルオーディオ信号における差分データ
US10789964B2 (en) Dynamic bit allocation methods and devices for audio signal
CN107895580A (zh) 一种音频信号的重建方法和装置
JP4981122B2 (ja) 抑制されたベクトル量子化
US9413388B1 (en) Modified huffman decoding
WO2022227212A1 (zh) 基于联邦学习的语音表征模型训练方法、装置、设备及介质
WO2015165264A1 (zh) 处理信号的方法及设备
CN101266795A (zh) 一种格矢量量化编解码的实现方法及装置
WO2016082278A1 (zh) 主成分分析pca映射模型的编解码方法及装置
CN106033672B (zh) 确定声道间时间差参数的方法和装置
US9583113B2 (en) Audio compression using vector field normalization
CN103119649A (zh) 用于基于上下文进行无损编码和解码的方法和设备
CN101626242B (zh) 改进的霍夫曼解码方法及装置
WO2019243670A1 (en) Determination of spatial audio parameter encoding and associated decoding
CN102708872B (zh) 3d音频中水平方位参数码本获取方法
CN104282308A (zh) 频域包络的矢量量化方法和装置
CN102760442B (zh) 一种3d音频中水平方位参数量化方法
WO2016023322A1 (zh) 多声道声音信号编码方法、解码方法及装置
CN120226075A (zh) 参数化空间音频编码
CN110709927A (zh) 编码装置、解码装置、平滑化装置、逆平滑化装置、其方法及程序

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 14906679

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 14906679

Country of ref document: EP

Kind code of ref document: A1