CN1527282A - Method and apparatus for expansible coding and decoding code audio-frequency data - Google Patents
Method and apparatus for expansible coding and decoding code audio-frequency data Download PDFInfo
- Publication number
- CN1527282A CN1527282A CNA031650376A CN03165037A CN1527282A CN 1527282 A CN1527282 A CN 1527282A CN A031650376 A CNA031650376 A CN A031650376A CN 03165037 A CN03165037 A CN 03165037A CN 1527282 A CN1527282 A CN 1527282A
- Authority
- CN
- China
- Prior art keywords
- bit
- coding
- information
- unit
- layer
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; 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
- G10L19/04—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 predictive techniques
- G10L19/16—Vocoder architecture
- G10L19/18—Vocoders using multiple modes
- G10L19/24—Variable rate codecs, e.g. for generating different qualities using a scalable representation such as hierarchical encoding or layered encoding
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11B—INFORMATION STORAGE BASED ON RELATIVE MOVEMENT BETWEEN RECORD CARRIER AND TRANSDUCER
- G11B20/00—Signal processing not specific to the method of recording or reproducing; Circuits therefor
- G11B20/10—Digital recording or reproducing
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B14/00—Transmission systems not characterised by the medium used for transmission
- H04B14/02—Transmission systems not characterised by the medium used for transmission characterised by the use of pulse modulation
- H04B14/04—Transmission systems not characterised by the medium used for transmission characterised by the use of pulse modulation using pulse code modulation
Abstract
An audio coding method capable of controlling a bit rate, a coding device and a decoding device are provided to supply a more delicate FGS(Fine Grain Scalability) with a little amount of data, and to configure a frequency resolving power as being similar to a transfer function of ears of human being, thereby realizing high sound quality in a low layer. A coding device converts PCM audio data(701), and quantizes the converted data according to a quantization step size of a coding band(702). The coding device codes additional information including scale factor information and arithmetic coding model information(704), divides plural quantization samples corresponding to a basic layer to map the divided samples on a bit plane, and arithmetically codes the samples(705). The device counts the number of coded bits. If the number of the counted bits exceeds a usable bit range(706), the device stops coding a current layer. The device codes all layers including a target layer(707).
Description
Technical field
The present invention relates to the Code And Decode voice data, especially specifically, relate to and be used for coding audio data, so that the audio bitstream of coding has the method and apparatus of telescopic bit rate, and the method and apparatus that is used for decoding audio data.
Background technology
Because the development of Digital Signal Processing recently, in most cases sound signal is stored as digital signal and reproduction usually.DAB storing/restoring device is transformed into pulse code modulation (pcm) to sound signal by sampling and quantification, just digital signal.By such operation, DAB storage/reproduction device is at information storage medium, and such as storage pcm audio data in CD (CD) and the Digital video disc (DVD), and response user's order is reproduced the signal of storage so that the user can listen voice data.With respect to the analogy method of using fine groove (LP) record or tape, stored digital/reconstructing method has improved audio quality widely, and has significantly reduced the deterioration that is caused by the long memory cycle.Yet because a large amount of numerical datas, numerical approach is existing problems aspect storage and transmission.
For addressing this problem, various compression methods are used to the compression digital audio frequency signal.
In motion picture expert group (MPEG), perhaps among the AC-2/AC-3 by the Dolby exploitation, use psychoacoustic model to reduce data volume by ISO standardsization.As a result of, data volume can be reduced effectively and no matter the characteristic of signal how.In other words, MPEG/ audio standard or AC-2/AV-3 method can provide almost identical with CD audio quality with the bit rate of 64～384 Kbps only, this bit rate be previous digital coding method bit rate 1/6 to 1/8.
Yet in these methods, search is applicable to the optimum condition of fixed bit rate and then carries out and quantizes and coding.Therefore, if when sending bit stream by network because the very poor transmission bandwidth that makes of network condition is lowered, may occur disconnecting with suitable service and can not offer the user again.In addition, the bit stream that is transformed into reduced size when bit stream expectation should be carried out encoding process reducing the size of bit stream again, and increase required calculated amount when more being applicable to the mobile device with limited memory capacity.
For addressing this problem, applicant of the present invention has proposed korean patent application No.97-61298, on November 19th, 1997, title " uses the method and apparatus of the scalable bit rate audio coding/decoding of position section algorithm coding (BSAC) ", this patent is authorized to Korean patent No. No.261253 on April 17th, 2000.According to the BSAC technology, have the high bit rate bitstream encoded and can be become bit stream, and can only be reconstructed with bit stream partly with low bit rate.Therefore, when network over loading, perhaps during the poor performance of demoder, or the user by only using the bit stream of part, can offer the user to the service with certain audio quality when asking low bit rate, although along with the decline of bit rate, quality can descend inevitably pro rata.
Yet, because adopting, the BSAC technology revises discrete cosine transform (MDCT) converting audio frequency signal, the low layer distortion becomes more serious.Reason is as follows.That is, though MDCT has reduced audio quality in layer step-down, the frequency resolution of MDCT is constant.Therefore, from psychoacoustic model, exceedingly increased about the frequency resolution of the insensitive frequency band of people's ear.
Summary of the invention
Invention provides a kind of method and apparatus that is used for telescopically (with scalability) audio data encoding and decoding, is that fine granular scalability (FGS) provides lower complexity by this method and apparatus.
According to one aspect of the present invention, a kind of method of voice data being carried out scalable coding is provided, this method comprises encodes to comprising corresponding to the scaling factor information of ground floor and the additional information of encoding model information, the reference encoder model information, according to from MSB (highest significant position) to LSB (least significant bit (LSB)) and the order from the low frequency to the high frequency, with the K bit is that the unit pair a plurality of quantized samples corresponding with ground floor are carried out arithmetic coding, wherein K is the integer more than or equal to 1, and, finish up to the coding of multilayer by repeat these steps by 1 ordinal number that increases layer at every turn.
According to another aspect of the present invention, a kind of coding method is provided, it comprises, to voice data cut into slices (slicing), make the voice data of section corresponding with a plurality of layers, obtain calibration segment information and the coding section information corresponding with each layer of multilayer, according to calibration segment information corresponding and coding section information the additional information that comprises scaling factor information and encoding model information is encoded with ground floor, quantize to obtain quantized samples by reference scaling factor information pair and ground floor corresponding audio data, the reference encoder model information, according to the order from MSB to LSB and from the low frequency to the high frequency, with the K bit is that unit carries out arithmetic coding to resulting a plurality of quantized samples, wherein K is an integer more than or equal to 1, and, finish up to the coding of multilayer by repeat these steps by 1 ordinal number that increases layer at every turn.
Before the additional information coding, this method also can comprise, obtain the available bits scope in each layer of a plurality of layers, wherein in the arithmetic coding of a plurality of quantized samples that obtained, the number of coded-bit is counted, if the bit of being counted outnumber bit range corresponding to this bit, coding just stops, even if the number of the bit of after each quantized samples all is encoded, being counted less than this bit range, the bit that is encoded not yet after the inclined to one side sign indicating number of low layer is finished just is encoded into the scope that this bit range is allowed.
The section of voice data comprises, carries out the wavelet transform of voice data, with reference to cutoff frequency the wavelet transform data is cut into slices, so that the data of section are corresponding to a plurality of layers.
And the coding of additional information can comprise the differential coding to scaling factor information and encoding model information.
Arithmetic coding can comprise a plurality of quantized samples are mapped on the bit-planes, and is that unit carries out arithmetic coding to sample with the K bit in the bit range of allowing in the layer corresponding with sample according to the order from MSB to LSB and from the low frequency to the high frequency.
According to another aspect of the present invention, a kind of method that voice data with the hierarchy coding is carried out telescopic decoding is provided, this method comprises, the scaling factor information that comprises corresponding ground floor and the additional information of encoding model information are decoded, the reference encoder model information, according to the order from MSB to LSB and from the low frequency to the high frequency, with the K bit is that unit carries out arithmetic decoding to voice data, and acquisition quantized samples, wherein K is an integer more than or equal to 1, with reference to the resulting quantized samples of scaling factor information inverse quantization, the inverse quantization sample is carried out anti-wavelet transform, and, finish up to the decoding of multilayer by repeat these steps by 1 ordinal number that increases layer at every turn.
The decoding of additional information can comprise the differential coding to scaling factor information and encoding model information.The arithmetic decoding of voice data can comprise according to the order from MSB to LSB and from the low frequency to the high frequency, is that unit carries out arithmetic decoding to voice data with the K bit, and obtains quantized samples from the bit of decoding.
According to another aspect of the present invention, a kind of device that voice data with the hierarchy coding is carried out telescopic decoding is provided, it comprises the deblocking unit, it is decoded to comprising corresponding to the scaling factor information of ground floor and the additional information of encoding model information, and reference encoder model information, is that unit decodes to voice data according to the order from MSB to LSB and from the low frequency to the high frequency with the K bit, and obtains quantized samples, wherein K be one more than or equal to 1 integer; Inverse quantization unit, it carries out inverse quantization with reference to scaling factor information to the quantized samples that is obtained; With the inverse transformation unit, it carries out anti-wavelet transform to the inverse quantization sample.
To be unit with the K bit according to the order from MSB to LSB and from the low frequency to the high frequency in the bit range that equivalent layer is allowed preferably carry out the difference decoding to scaling factor information and encoding model information in the deblocking unit, order is from MSB to LSB and from the low frequency to the high frequency, and the bit-planes that is arranged from the bit of decoding obtains quantized samples.
According to another aspect of the present invention, provide a kind of voice data is carried out telescopic apparatus for encoding, it comprises, converter unit, it carries out wavelet transform to voice data, quantifying unit, it is with reference to scaling factor information, the wavelet transform voice data corresponding with every layer quantized, and output quantized samples, and encapsulation unit, it is encoded to comprising corresponding to every layer the scaling factor information and the additional information of encoding model information, and the reference encoder model information, according to the order from MSB to LSB and from the low frequency to the high frequency, with the K bit is that unit carries out arithmetic coding to a plurality of quantized samples from quantifying unit, wherein K be one more than or equal to 1 integer.
Encapsulation unit obtains calibration segment information and the coding section information corresponding to each layer of a plurality of layers, and according to the calibration segment information coding section information corresponding to each layer the additional information that comprises scaling factor information and encoding model information being encoded.And, the number of encapsulation unit counting coded-bit, if the bit of being counted outnumber bit range corresponding to this bit, just stop coding, even if the scope that the bits of encoded of encoding not yet after the number of the bit of being counted after quantized samples all is encoded less than the bit range corresponding to this bit, is then finished the low layer coding allows to this bit range.Encapsulation unit carries out differential coding to scaling factor information and encoding model information.
Encapsulation unit is mapped in a plurality of quantized samples on the bit-planes, and is that unit carries out arithmetic coding to sample with the K bit in the bit range of allowing in corresponding with sample layer according to the order from MSB to LSB and from the low frequency to the high frequency.
Description of drawings
By being described in detail with reference to the attached drawings its most preferred embodiment, above-mentioned purpose of the present invention and advantage will become more apparent, wherein:
Fig. 1 is the calcspar of the code device of most preferred embodiment of the present invention;
Fig. 2 is the calcspar of the decoding device of most preferred embodiment of the present invention;
Fig. 3 is the structural drawing of frame, and described frame is formed on bitstream encoded in the hierarchy so that can the control bit rate;
Fig. 4 is the detail drawing of the structure of additional information;
Fig. 5 is the reference diagram that Code And Decode method of the present invention is explained on principle ground;
Fig. 6 is a reference diagram of more specifically explaining coding method of the present invention;
Fig. 7 is a process flow diagram of explaining the coding method of most preferred embodiment of the present invention;
Fig. 8 is a process flow diagram of explaining the coding/decoding method of most preferred embodiment of the present invention; With
Fig. 9 is a process flow diagram of explaining the coding/decoding method of another most preferred embodiment of the present invention.
Embodiment
With reference to figure 1, according to the present invention, code device so that can control the bit rate of bitstream encoded, and comprises converter unit 11, tonequality unit 12, quantifying unit 13 and bit-envelope unit 14 with the hierarchy coding audio data.
Quantifying unit 13 according to corresponding to the scaling factor information of sound signal in each frequency range scalar quantization sound signal, make the shielding thresholding that the level of quantizing noise is provided less than tonequality unit 12 in the frequency range, so that the people can not perceive noise.Then, the sample of quantifying unit 13 output quantifications.In other words, the noise-shielding ratio (NMR) of the ratio by the noise that uses the shielding thresholding that calculates in the tonequality unit 12 and produce as each frequency range, quantifying unit 13 is carried out quantifications, makes that the NMR value in the full frequency band is 0dB or littler.0dB or littler NMR value mean that the people can not the perception quantizing noise.
Bit-envelope unit 14 codings belong to every layer quantized samples and additional information, and with hierarchy encapsulation coded signal.Calibration segment information during additional information comprises every layer, coding section information, their scaling factor information and encoding model information.Calibration segment information and coding section information can be packaged into header message, and then are sent to decoding device.Otherwise calibration segment information and coding section information can be encoded and be packaged into every layer additional information, and then send to decoding device.Calibration segment information and coding section information can not be sent to decoding device, because they are pre-stored in the decoding device in some cases.
More particularly, when the additional information that comprises scaling factor information corresponding with ground floor and encoding model information is carried out differential coding, 14 references and ground floor corresponding codes model information of bit-envelope unit, with some bits is that unit cuts into slices to quantized samples, and according to the order from the highest significant position to the least significant bit (LSB) sample of bit is encoded.Two components that are encoded are additional information and quantized samples, and additional information is more important than quantizing sample.The reasons are as follows.Just, even quantized samples is accurately transmitted, yet when the additional information that comprises calibration segment information, quantization step etc. was not correctly transmitted, it was reconstructed into full of prunes signal.But, when additional information only by accurate reconstruct the time, total then can be by the reconstruct profile, even slight error is arranged in quantized samples.Therefore, additional information at first is encoded, and then, quantized samples is encoded.When additional information and quantized samples were encoded, if the bit number that the number of the bit of being counted is allowed more than or equal to each layer, coding just stopped, and the formation of bit stream also stops.Bian Ma quantized samples is not allowed coding in the layer that bit number also has living space at it yet.Because the bit range and the number of plies are suitably to determine according to the size of voice data.Even all quantized samples are not encoded, and the formation of bit stream owing to coding during the variation of quantity of destination layer be terminated because important (significant) information at first is encoded, audio quality can seriously not become bad yet.According to present embodiment, in to the quantized samples coding, the 14 pairs of a plurality of quantized samples in bit-envelope unit are carried out bit section (bit slicing), and to be unit with the K bit carry out arithmetic coding to the sample through the bit section, and order is from the low frequency to the high frequency (K is an integer more than or equal to 1).Then, in the second layer, identical process is repeated to carry out.That is, coding is finished up to the coding of a plurality of predetermined layer along with the increase of the number of plies is carried out.Hierarchy by bitstream encoded of the present invention will be explained below.
The calibration segment information is meant the information that is used for more suitably carrying out according to the frequency characteristic of sound signal quantification.When frequency field was divided into a plurality of frequency ranges and a suitable scaling factor and is assigned to each frequency range, the indication of calibration segment information was corresponding to every layer calibration section.Like this, every layer belongs at least one calibration section.Each calibration section has the scaling factor of a distribution.Usually, form because calibration section is a auditory properties according to the people, the bandwidth of calibration section is less when low frequency, and it increases along with the increase of frequency.
In addition, coding section information is meant the information that is used for more suitably carrying out according to the frequency characteristic of sound signal lossless coding.When frequency field was divided into a plurality of frequency ranges and suitable encoding model and is assigned to each frequency range, the indication of coding section information was corresponding to every layer coding section.Because the calibration bandwidth is with frequency change, so be difficult to obtain uniform probability distribution.Therefore, make up several frequency signals, so that the signal of each section has similar probability distribution.Every group of this composite signal is called a coding section.Usually, when low frequency a plurality of calibration paragraphs on a coding section.Along with frequency increases, the calibration section overlaps with coding section.In practice, calibration section and coding section separate by experience, and the scaling factor and the encoding model of correspondence are determined according to each section respectively with it.Particularly, the present invention is based on wavelet transform, and divide coding section, so that corresponding with the final node of the tree construction of wavelet transform.This will describe in detail afterwards.
Fig. 2 is the calcspar of the decoding device of most preferred embodiment of the present invention.
With reference to figure 2, the decoding device decoded bits flows to by network condition, and the performance of decoding device and the determined destination layer of user's selection make that the bit rate of bit stream can Be Controlled.Decoding device comprises deblocking unit 21, inverse quantization unit 22 and inverse transformation unit 23.
Deblocking unit 21 deblocking bits flow to destination layer, and the bit stream in every layer is decoded.That is, the additional information that comprises scaling factor information corresponding with each layer and encoding model information is decoded, and then according to resulting encoding model information, the coded quantization sample that belongs to this layer is decoded, and quantized samples is by reconstruct.In the present embodiment, 21 pairs of scaling factor information in deblocking unit and arithmetic coding model information are carried out differential decoding, according to the order from MSB to LSB and from the low frequency to the high frequency, are that unit carries out arithmetic decoding to the coded quantization sample with the K bit, carry out bit combination then, obtain quantized samples thus.Here, K is an integer more than or equal to 1.Bit combination is the anti-process of the bit section carried out during the coding.
Simultaneously, from the header message of bit stream, or, obtain calibration segment information and coding section information by the additional information of decoding in every layer.On the other hand, decoding device store calibration segment information and coding section information in advance.Inverse quantization unit 22 is according to scaling factor information inverse quantization corresponding with each sample and the quantized samples in each layer of reconstruct.Inverse transformation unit 23 carries out frequency/time map by wavelet transform to reconstructed sample, so that output is as the sample of the pcm audio data of time domain.
Fig. 3 is the structural drawing of frame, and described frame forms with the hierarchy bitstream encoded, and making can the control bit rate.
With reference to figure 3, be encoded into hierarchy according to the frame of bit stream of the present invention by mapping quantized samples and additional information, to obtain fine granular scalability (FGS).In other words, the low layer bit stream is included in the enhancement layer bit-stream of hierarchy.The additional information that needs in every layer is assigned to every layer, is encoded then.
The stem zone that is used to store header message is placed in the front of bit stream, and the information of relevant layer 0 is packed the stem zone after then, then, belongs to packed in order as the information of layer 1-N of enhancement layer.Layer from the stem zone to layer 0 information is known as basic unit, and the layer from the stem zone to layer 1 information is known as layer 1 and the layer from the stem zone to layer 2 information is known as layer 2.Equally, the superiors represent from stem zone to a layer N information, in other words, from basic unit to layer N as enhancement layer.Additional information and coding audio data are stored as each layer information.For example, additional information 2 and coded quantization sample are stored stratification 2 information.Here, N is an integer more than or equal to 1.
Fig. 4 is the detail drawing of the structure of additional information.
With reference to figure 4, additional information and coded quantization sample are stored as any additional information, and in the present embodiment, additional information comprises the arithmetic coding model information, quantizing factor information, additional information and other additional information of relevant channel.The arithmetic coding model information is the index information of arithmetic coding model, should be used to encodes or decodes belong to corresponding to this information the layer quantized samples.Scaling factor information indication quantization step, this step-length are used to quantize or inverse quantization belongs to voice data corresponding to the layer of information.The additional information of relevant channel be relevant channel such as the stereosonic information of M/S.Other additional information is the relevant stereosonic flag information of M/S that whether uses.
In the present embodiment, 14 pairs of arithmetic coding model informations in bit-envelope unit and scaling factor information are carried out lossless coding.Quantize section because each calibration section has one, for quantizing factor is encoded, belong to minimum value in the middle of quantizing factor of each calibration section at first by lossless coding, then, the difference between minimum value and another quantizing factor is by lossless coding.For the lossless coding of additional information, arithmetic coding or Arithmetic coding can be used.In the present embodiment, arithmetic coding is performed.Equally, corresponding to the arithmetic coding model of each coding section and relevant information of allowing bit range can by with quantization step in identical coding method, promptly differential coding is encoded.
In the present embodiment, the losslessly encoding of additional information is carried out in bit deblocking unit 21, and additional information comprises arithmetic coding model information and scaling factor information.Quantize section because each calibration section has one, for quantizing factor is decoded, belong to minimum value in quantizing factor of each calibration section at first by losslessly encoding, then, the difference between this minimum value and another quantizing factor is by losslessly encoding.According to being used to additional information is carried out the identical mode of code device of lossless coding with arithmetic coding, the decoding device of present embodiment is carried out arithmetic coding.Equally, corresponding to the arithmetic coding model of each coding section and relevant information of allowing bit range by with quantizing factor in identical coding/decoding method decoded.
Fig. 5 is a reference diagram of explaining coding method of the present invention and coding/decoding method.
With reference to Fig. 5, represent the spectral line that quantized samples is formed with the oblique line rectangle.Reference symbol A represents every layer boundary line, and B represents the division boundary line corresponding to each terminal node of the tree construction of wavelet transform.
In the wavelet transform that coding method of the present invention is used, frequency transformation is to use tree construction to carry out, so that representation class is similar to the voice data of the shape of the bank of filters corresponding with people's ear characteristics.The final node of tree construction is corresponding with the calibration section of arithmetic coding respectively.Therefore, each last node is just corresponding with scaling factor.
Coding section is the unit of transfer that is used for the arithmetic coding model information of arithmetic coding, and it can be determined according to code efficiency.For example, suppose that the calibration section of final node is identical with coding section.Then, layer and final node shine upon mutually, as shown in Figure 5.Because the data corresponding to each final node appear on the time shaft of same frequency band, so be not separated when the layering corresponding to the data segment of final node.
Determine layer 0 so that coding is carried out up to a frequency band (a).Determine layer 1 so that coding is carried out up to a frequency band (b).Determine layer 2 so that coding is carried out up to a frequency band (c).Determine layer 3 so that coding is carried out up to a frequency band (d).Determine layer 4 so that coding is carried out up to a frequency band (e).Determine layer 5 so that coding is carried out up to a frequency band (f).Determine layer 6 so that coding is carried out up to a frequency band (g).Determine layer 7 so that coding is carried out up to a frequency band (h).
At first, the quantized samples that belongs to layer 0 utilizes the corresponding codes model to be encoded in predetermined bit range.Simultaneously, as the additional information of layer 0, the quantized samples that belongs to layer 0 is by arithmetic coding.Be that bit number was counted when unit encoded to quantized samples with some bits.If the bit number of being counted surpasses the bit range of allowing, the coding of layer 0 just stops, and layer 1 is by arithmetic coding.Among the quantized samples of layer 0, when also having living space in the bit number of allowing in layer 0 and 1, uncoded quantized samples is encoded.
Then, the quantized samples that belongs to layer 1 utilizes the corresponding codes model to be encoded.Simultaneously, the additional information of layer 1 is by arithmetic coding.Even if also have living space in the bit range of after all samples corresponding with layer 1 are encoded, allowing, in the layer 0 not yet the bit of coding just be encoded, up to counting down to the bit of allowing.If the bit number that coding is counted surpasses the bit range of allowing, the coding of layer 1 just stops, and the coding of layer 2 is activated.Same process is repeated, and finishes up to the coding of layer 7.
If all change that sample is encoded most and the bit range of allowing of not considering each layer, promptly, even if number of coded bits surpass allow bit range after all quantized samples be encoded (this means that some bit in the bit range of allowing of one deck is used to encode to working as anterior layer down), then common situation is that belonging to down, the quantized samples of one deck can not be encoded.Therefore, under scalable decoding situation, if decoding is at low layer, rather than on all decoded layers, carry out, because scope is not encoded up to all quantized samples of preset frequency, fluctuation may appear in the quantized samples of decoding, cause " Birdy " effect, make audio quality can become bad.
When decoding was carried out in the mode opposite with cataloged procedure, bit number was according to allowing that bit range counts.Therefore, the decoding of predetermined layer is regularly put and can be identified.
When coding when carrying out on the spectral line of direction from msb to Isb, the data on the same bit-planes on the final node of the tree construction of wavelet transform should be encoded together.For example, suppose that following quantized samples is on certain final node.
00000000101010110101
11111100000000000000
00001100110000000110
In the coding based on MDCT, quantized samples is grouped into following 5 3*4 bit-planes, and coding carries out downwards to the right then.
0000?0000?1010?1011?0101
1111?1100?0000?0000?0000
0000?1100?1100?0000?0110
In the coding based on wavelet, quantized samples is considered to be on the bit-planes, is unit with some a plurality of bits then, encodes successively from MSB to LSB and from the low frequency to the high frequency.As mentioned above, MSB 00000000101010110101 is that unit is encoded from left to right with the K bit, ensuing then bit 11111100000000000000 is that unit is encoded from left to right with the K bit, and last, LSB00001100110000000110 is that unit is encoded with the K bit.Here, K is an integer more than or equal to 1.Particularly, if K equals 1, binary coding is performed, because carry out arithmetic coding in the present invention.
Fig. 6 is a reference diagram of more specifically explaining coding method of the present invention.
With reference to figure 6, code device is stored respectively by quantizing to be converted to the sample X1 of frequency-region signal, X2, X3, X4 ... and the quantized samples XQ1 that obtains, XQ2, XQ3, the value of symbol S[XQ1 of XQ4], S[XQ2], S[XQ3], S[XQ4] ..., and obtain their absolute value, promptly | XQ1|, | XQ2|, | XQ3|, | XQ4| ..., make all quantized samples all be on the occasion of.
Usually, as is known, because the information mistake that 1 bit error under the low bit causes is greater than caused information mistake under the higher bit, so the importance of MSB (significance) is just greater than LSB.But, under the conventional coding technology situation, do not consider importance when encoding.Therefore, can be used iff some leading bit stream, then a large amount of be included in untapped tail bits stream in the information not too important information of comparing can be included in the precedence bits stream.Therefore, in the present invention, the absolute value that is used for the quantized samples of each section is that unit is separated with some bits.When | XQ1| is 5 Bit datas and is unit separated the time with some bits from MSB to LSB, provide XO1,5, XO1,4, XO1,3, XO1,2, XO1,1, | XQ2| provides XO2, and 5, XO2,4, XO2,3, XO2,2, XO2,1, | XQ3| provides XO3, and 5, XO3,4, XO3,3, XO3,2, XO3,1, and | XQ4| provides XO4, and 5, XO4,4, XO4,3, XO4,2, XO4,1 (?).
Then, MSB is collected according to the order from the low frequency to the high frequency and cuts to form vector with predetermined unit, and vector is encoded then.Here, C[A1 ..., Ak] and expression A1 ..., the arithmetic coding of the vector that Ak constitutes, K is an integer more than or equal to 1.Because arithmetic coding is that unit carries out with some bits, even radix point level Bit Allocation in Discrete, for example 0.001 bit is also allowed in 1 bit is encoded, and available thus lesser amt bit is encoded to bulk information, and this shows the code efficiency height.The Arithmetic coding is the alternative form of lossless coding, and it requires at least one bit of each symbol, and its code efficiency is more much lower than arithmetic coding (arithmetic coding).
Value of symbol coding or early than or be later than the absolute value of quantized samples.At first may cause at first less (importance) information being encoded to the value of symbol coding.In the present embodiment, the coding of value of symbol after.The reasons are as follows.That is, the order from MSB to LSB, if only in 1 bits of encoded that occurs before at first arriving, it is zero that quantized samples is considered to, its value of symbol is nonsensical.For example, suppose that quantized samples is 5 bit binary number 00011, have only 3 significant bits the highest to be encoded among 5, when decoding was carried out, they were reconfigured as 00000.Therefore, in this case, even quantized samples has a value of symbol, value of symbol information also is insignificant.But, if 4 significant bits the highest are encoded, quantized samples is considered to 00010, and is reconstructed into 00010.In this case, value of symbol just has great importance.Therefore, when the quantized samples that under the situation about at first arriving in 1 (rather than 0) MSB is formed is encoded, relevant corresponding quantized samples be on the occasion of or the information of negative value, promptly value of symbol is prior at first being encoded to the coding of other value.The vector 1010 that MSB constitutes at first is encoded, and determines then whether the coding of value of symbol needs.From as 1010 of the MSB of 4 quantized samples group, can determine in the first and the 3rd quantized samples it is 1 rather than 0 at first to be encoded.Therefore, the value of symbol of the first and the 3rd quantized samples is encoded.Then, inferior significant bit group is encoded.Use same mode, the value of symbol of 1 quantized samples that at first arrives is encoded.Use same mode, encode up to LSB.
The decoding of the bit stream of Xing Chenging is to carry out with the reverse order of encoding like this.The stem of the frame of formation bit stream is decoded, and then, the additional information of basic unit is decoded.More specifically, the additional information of basic unit, promptly encoding model information and scaling factor information are decoded.According to resulting encoding model information and scaling factor information, the quantized samples of coding is decoded successively from MSB to LSB and from the low frequency to the high frequency.
Explain in detail by way of example now.
0100
0000
0000
0001
In above-mentioned bit-planes, 0 probability that occurs is greater than 1 probability that occurs, and coding can carry out with less bit number.Under the situation as the Arithmetic coding of the possibility of lossless coding, in order to encode to 0, at least one bit needs.Use arithmetic coding to allow only to encode less than 1 bit number, realize at least 1 coding thus by distributing, that is, and binary coding.Therefore, meticulousr hierarchy can form.
The encoding model of arithmetic coding can selecting at preceding vector according to the vector of higher level's vector sum present encoding.Following Example illustrates the arithmetic coding of bit-planes of the present invention:
0110
0011
0101
1000
Wherein the bit section is performed, and for example, per 4 bits link to each other according to the order from MSB to LSB, obtain primary vector 0110, secondary vector 0011, the three vector the 0101 and the 4th vector 1000.At vector by in the arithmetic coding by turn, secondary vector to be used for the 3rd 1 encoding model of encoding be according to uppermost vector, promptly the position 00 of the front of primary vector 0110 and current vector of just encoding is selected.Select encoding model to mean that selection will be used as 0 probability of occurrence of 0/1 arithmetic coding information.
For the control bit rate, promptly allow the bit number that is used to every layer in order to use scalability, to consider, be cut off corresponding to the bit stream of a frame, thereby only utilize a spot of data just can decode.For example, at maximum bit rate is that 96kbps and bit stream size are under the situation of 2096 bits, if only have and wish decodedly corresponding to the bit stream of 48kbps, then only there are 1048 bits of bit stream to be used, thus can be obtained corresponding to the decoding audio data of 48kbps.
Explain Code And Decode method of the present invention now according to said structure.
Code device reads the pcm audio data, storage data (not shown) in storer, and by tonequality modeling acquisition shielding thresholding and additional information from the pcm audio data of storage.Because the pcm audio data are time-domain signals, the pcm audio data are become frequency-region signal by wavelet transform.Then, code device obtains quantized samples according to the signal that quantizes segment information and quantizing factor throughput beggar wave conversion.As mentioned above, quantized samples is cut into slices by bit, is that unit encapsulates by arithmetic coding and with hierarchy with the K bit.Here, K be one more than or equal to 1 integer.
Fig. 7 is a process flow diagram of explaining the coding method of most preferred embodiment of the present invention.
In the present embodiment, code device is determined destination layer according to a given target bit rate and additional information, and the generation information corresponding with every layer.In other words, obtained as cutoff frequency, and according to the data that will be encoded corresponding to every layer quantification segment information and coded message by the basis of each layer, allow that the bit range of coding is assigned to every layer in every layer.
With reference to figure 7, the wavelet transform of pcm audio data is carried out in step 701.In step 702, quantize according to the quantization step of the coding section under the wavelet transform data.In step 703, when anterior layer is set to basic unit.In step 704, comprise corresponding to the scaling factor information of basic unit and the additional information of arithmetic coding model information being encoded.
Then, in step 705, a plurality of quantized samples corresponding to basic unit are mapped to bit-planes by the bit section, and are that unit carries out arithmetic coding (K 〉=1) with the K bit, and order is from msb to lsb and from the low frequency to the high frequency.Number of coded bits is counted, if in step 706, this number surpasses the bit range when anterior layer, and then the coding in anterior layer is stopped, and the coding in one deck under beginning.If the bit number of being counted surpasses the bit range in 706 steps, this process turns back to step 705 in step 707 through step 708 and 709, to handle down one deck.Because basic unit does not have lower layer, step 707 is not carried out, and still, if the bit number of being counted is no more than the bit range of step 706, then step 707 is carried out.By above-mentioned steps, till all layers that reach destination layer all are encoded.
Fig. 8 is a process flow diagram of explaining the coding/decoding method of most preferred embodiment of the present invention.
With reference to figure 8, decoding device receives the bit stream that is formed by the voice data with the hierarchy coding, and the header message in each frame is decoded.In step 801, decode then to comprising corresponding to the scaling factor information of ground floor and the additional information of arithmetic coding model information.With reference to the arithmetic coding model information, in step 802, quantized samples is by being that unit carries out arithmetic decoding to bit stream and obtains with the K bit from MSB to LSB.Here, K is an integer more than or equal to 1.In step 803, by reference scaling factor information, the quantized samples that is obtained is by inverse quantization.In step 804, the inverse quantization sample is by anti-wavelet transform.Along with each ordinal number by every layer of 1 increase, step 801 to 804 repeats, till nearly the decoding of intended target layer is finished.
Fig. 9 is a process flow diagram of explaining another best coding/decoding method of implementing of the present invention.
Referring to Fig. 9, the bit stream that reception is formed by the voice data with the hierarchy coding, according to the cutoff frequency of the decoding of the header message in every frame corresponding to every layer, from header message identification quantification segment information and the coding section information corresponding to every layer, and every layer permission uses bit range to be identified by decoding.
In step 901, layer index is set to basic unit.In step 902, the additional information of relevant basic unit is decoded, in step 903, by being to obtain quantized samples in the unit bit range of successively bit stream decoding being allowed in every layer with the K bit from MSB to LSB and from the low frequency to the high frequency.Here, K is an integer more than or equal to 1.In step 904, check whether one deck is last one deck.In step 905, along with the number of plies increases one by one, repeated execution of steps 902 and 903 reaches up to predetermined destination layer in each layer.Alternatively, decoding device can have cutoff frequency in advance, quantize at least one in segment information, coding section information and the bit range, rather than obtains these information in the header message of storing from each frame of the bit stream that received.
According to above-mentioned the present invention, the method and apparatus that voice data is carried out scalable coding/decoding is provided, its complicacy is lower, and the scalability (FGS) of fine granulation is provided simultaneously.In other words, according to the present invention, more high-quality FGS can provide with more a spot of data by using arithmetic coding, and better audio quality also can provide by utilizing wavelet transform to form the frequency resolution that is similar to people's ear transport function, even in low layer.
Claims (20)
1, a kind of voice data is carried out the method for scalable coding, comprising:
Additional information is encoded, and additional information comprises corresponding to the scaling factor information of ground floor and encoding model information;
Is that unit pair with ground floor corresponding a plurality of quantized samples carry out arithmetic coding according to the order from MSB to LSB and from the low frequency to the high frequency with the K bit by the reference encoder model information, and wherein K is an integer more than or equal to 1;
Add 1 and repeat each step by the ordinal number that makes layer at every turn, be done up to the coding of multilayer.
2, a kind of coding method comprises:
To the voice data section, make that the voice data of section is corresponding with a plurality of layers;
Obtain calibration segment information and the coding section information corresponding with each layer of multilayer;
According to calibration segment information and coding section information the additional information that comprises scaling factor information and encoding model information is encoded corresponding to ground floor;
Quantize to obtain quantized samples by reference scaling factor information pair and ground floor corresponding audio data;
By the reference encoder model information, be that unit carries out arithmetic coding to a plurality of quantized samples that obtained according to the order from MSB to LSB and from the low frequency to the high frequency with the K bit, wherein K is an integer more than or equal to 1; With
Add 1 and repeat these steps by the ordinal number that makes layer at every turn, be done up to the coding of multilayer.
3, according to the method for claim 2, wherein in arithmetic coding, the reference encoder model information, the a plurality of quantized samples that obtained are by arithmetic coding, make that MSB is that unit is at first by arithmetic coding according to the order from the low frequency to the high frequency with the K bit, then time highest significant position according to the order from the low frequency to the high frequency with the K bit be unit by arithmetic coding, last LSB with the K bit be unit according to the order from the low frequency to the high frequency by arithmetic coding.
4, according to the method for claim 2, also comprise, before additional information is encoded, available bit range in each layer of acquisition multilayer, wherein in the arithmetic coding of a plurality of quantized samples that obtained, number of coded bits is counted, if the bit number of being counted surpasses the bit range corresponding to these bits, coding just stops, even if the bit number of being counted after quantized samples all is encoded is less than the bit range corresponding to these bits, the bit that is encoded not yet after the coding of low layer is finished is encoded into the scope that bit range is allowed.
5, according to the method for claim 2, wherein the section of voice data comprises:
Carry out the wavelet transform of voice data; With
By the reference cutoff frequency wavelet transform data are cut into slices, thereby make slice of data corresponding with a plurality of layers.
6, according to the method for claim 2, wherein the coding of additional information comprises the differential coding to scaling factor information and encoding model information.
7, according to the method for claim 2, wherein arithmetic coding comprises:
A plurality of quantized samples are mapped on the bit-planes; With
Is that unit carries out arithmetic coding to sample with the K bit according to the order from MSB to LSB and from the low frequency to the high frequency in the bit range of being allowed in corresponding to the layer of sample.
8, a kind of method that voice data with the hierarchy coding is carried out scalable decoding comprises:
Decode to comprising corresponding to the scaling factor information of ground floor and the additional information of encoding model information;
By the reference encoder model information, be that unit carries out arithmetic decoding according to the order from MSB to LSB and from the low frequency to the high frequency to voice data with the K bit, and obtain quantized samples, wherein K be one more than or equal to 1 integer;
By reference scaling factor information the quantized samples that is obtained is carried out inverse quantization;
The inverse quantization sample is carried out anti-wavelet transform; With
Add 1 by the ordinal number that makes layer at every turn, repeat these steps, finish up to a plurality of layers predetermined decoding.
9, method according to Claim 8, wherein the decoding that additional information is carried out comprises the differential decoding to scaling factor information and encoding model information.
10, method according to Claim 8, wherein the arithmetic decoding of voice data comprises:
According to the order from MSB to LSB and from the low frequency to the high frequency voice data is decoded;
Obtain quantized samples from the bit of decoding.
11, according to the method for claim 10, wherein carry out the decoding of voice data, make at first MSB with the K bit be unit according to the order from the low frequency to the high frequency by arithmetic decoding, then time highest significant position with the K bit be unit according to the order from the low frequency to the high frequency by arithmetic decoding, last LSB is that unit is decoded according to the order from the low frequency to the high frequency with the K bit.
12, a kind of device that voice data with the hierarchy coding is carried out scalable decoding comprises:
The deblocking unit, it is decoded to comprising corresponding to the scaling factor information of ground floor and the additional information of encoding model information, and by the reference encoder model information, with the K bit is that unit decodes to voice data according to the order from MSB to LSB and from the low frequency to the high frequency, and the acquisition quantized samples, wherein K be one more than or equal to 1 integer;
Inverse quantization unit, it carries out inverse quantization by reference scaling factor information to the quantized samples that is obtained; With
The inverse transformation unit, it carries out anti-wavelet transform to the inverse quantization sample.
13, according to the device of claim 12, wherein the deblocking unit carries out differential decoding to scaling factor information and encoding model information.
14, according to the device of claim 12, wherein the deblocking unit to be decoding to voice data in the K bit bit range that to be unit allowed in respective layer according to the order from MSB to LSB and from the low frequency to the high frequency, and obtains quantized samples from the bit-planes of arranging decoded bits thereon.
15, according to the device of claim 14, wherein the deblocking unit carries out arithmetic decoding, make MSB at first with the K bit be unit according to the order from the low frequency to the high frequency by arithmetic decoding, then time highest significant position with the K bit be unit according to the order from the low frequency to the high frequency by arithmetic decoding, last LSB with the K bit be unit according to the order from the low frequency to the high frequency by arithmetic decoding.
16, a kind of voice data is carried out the device of scalable coding, comprising:
Converter unit, it carries out wavelet transform to voice data;
Quantifying unit, it quantizes by the reference scaling factor information pair wavelet transform voice data corresponding with every layer, and the output quantized samples; With
Encapsulation unit, it is encoded to comprising corresponding to every layer the scaling factor information and the additional information of encoding model information, and by the reference encoder model information, with the K bit is that unit is according to the order from MSB to LSB and from the low frequency to the high frequency, a plurality of quantized samples from quantifying unit are carried out arithmetic coding, wherein K be one more than or equal to 1 integer.
17, according to the device of claim 16, wherein encapsulation unit obtains calibration segment information and the coding section information corresponding to each layer of a plurality of layers, and according to calibration segment information and coding section information corresponding to every layer the additional information that comprises scaling factor information and encoding model information is encoded.
18, according to the device of claim 17, wherein encapsulation unit is counted number of coded bits, if the bit number of being counted surpasses the bit range corresponding to this bit, just stop coding, even if the scope that the bits of encoded that is encoded not yet after the bit number of being counted after quantized samples all is encoded less than the bit range corresponding to this bit, is then finished coding in the low layer is allowed to this bit range.
19, according to the device of claim 16, wherein encapsulation unit carries out differential coding to scaling factor information and encoding model information.
20, according to the device of claim 16, wherein encapsulation unit is mapped to a bit-planes to a plurality of quantized samples, and in the K bit bit range that to be unit allowed according to the order from MSB to LSB and from the low frequency to the high frequency sample is carried out arithmetic coding in the layer corresponding with sample.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR102003000005O | 2003-01-02 | ||
| KR10-2003-0000050 | 2003-01-02 | ||
| KR10-2003-0000050A KR100528327B1 (en) | 2003-01-02 | 2003-01-02 | Method and apparatus for encoding/decoding audio data with scalability |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1527282A true CN1527282A (en) | 2004-09-08 |
| CN100555413C CN100555413C (en) | 2009-10-28 |
Family
ID=34805966
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB031650376A Expired - Fee Related CN100555413C (en) | 2003-01-02 | 2003-09-17 | The method and apparatus of telescopically audio data encoding and decoding |
Country Status (2)
| Country | Link |
|---|---|
| KR (1) | KR100528327B1 (en) |
| CN (1) | CN100555413C (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101828221B (en) * | 2007-08-16 | 2012-11-07 | 韩国电子通信研究院 | Apparatus and method for encoding and decoding enhancement layer |
| CN101283403B (en) * | 2005-10-05 | 2013-03-27 | Lg电子株式会社 | Method and apparatus for signal processing and encoding and decoding method, and apparatus thereof |
| CN103282959A (en) * | 2010-10-25 | 2013-09-04 | 沃伊斯亚吉公司 | Coding generic audio signals at low bitrates and low delay |
| CN101796579B (en) * | 2007-07-06 | 2014-12-10 | 法国电信公司 | Hierarchical coding of digital audio signals |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100813259B1 (en) * | 2005-07-13 | 2008-03-13 | 삼성전자주식회사 | Method and apparatus for encoding/decoding input signal |
| KR100793287B1 (en) * | 2006-01-26 | 2008-01-10 | 주식회사 코아로직 | Apparatus and method for decoding audio data with scalability |
| KR100827458B1 (en) * | 2006-07-21 | 2008-05-06 | 엘지전자 주식회사 | Method for audio signal coding |
-
2003
- 2003-01-02 KR KR10-2003-0000050A patent/KR100528327B1/en not_active IP Right Cessation
- 2003-09-17 CN CNB031650376A patent/CN100555413C/en not_active Expired - Fee Related
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101283403B (en) * | 2005-10-05 | 2013-03-27 | Lg电子株式会社 | Method and apparatus for signal processing and encoding and decoding method, and apparatus thereof |
| CN101796579B (en) * | 2007-07-06 | 2014-12-10 | 法国电信公司 | Hierarchical coding of digital audio signals |
| CN101828221B (en) * | 2007-08-16 | 2012-11-07 | 韩国电子通信研究院 | Apparatus and method for encoding and decoding enhancement layer |
| CN103282959A (en) * | 2010-10-25 | 2013-09-04 | 沃伊斯亚吉公司 | Coding generic audio signals at low bitrates and low delay |
| US9015038B2 (en) | 2010-10-25 | 2015-04-21 | Voiceage Corporation | Coding generic audio signals at low bitrates and low delay |
| CN103282959B (en) * | 2010-10-25 | 2015-06-03 | 沃伊斯亚吉公司 | Coding generic audio signals at low bitrates and low delay |
Also Published As
| Publication number | Publication date |
|---|---|
| KR100528327B1 (en) | 2005-11-15 |
| CN100555413C (en) | 2009-10-28 |
| KR20040062261A (en) | 2004-07-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1525436B (en) | Method and apparatus for encoding/decoding audio data with scalability | |
| CN1527306A (en) | Method and apparatus for coding and/or decoding digital data using bandwidth expansion technology | |
| CN1244904C (en) | Audio coding | |
| CN1110145C (en) | Scalable audio coding/decoding method and apparatus | |
| EP1617413A2 (en) | Multichannel audio data encoding/decoding method and apparatus | |
| EP1960999B1 (en) | Method and apparatus encoding an audio signal | |
| CN1926610A (en) | Synthesizing a mono audio signal based on an encoded multi-channel audio signal | |
| CN1765153A (en) | Coding of main and side signal representing a multichannel signal | |
| CN1878001A (en) | Apparatus and method of encoding audio data and apparatus and method of decoding encoded audio data | |
| CN1623185A (en) | Efficient improvement in scalable audio coding | |
| CN100350749C (en) | Lossless encoding/decoding in transmission system | |
| CN1822508A (en) | Method and apparatus for encoding and decoding digital signals | |
| CN1252678C (en) | Compressible stereo audio frequency encoding/decoding method and device | |
| KR101015497B1 (en) | Method and apparatus for encoding/decoding digital data | |
| CN1527282A (en) | Method and apparatus for expansible coding and decoding code audio-frequency data | |
| CN101105940A (en) | Audio frequency encoding and decoding quantification method, reverse conversion method and audio frequency encoding and decoding device | |
| CN1485849A (en) | Digital audio encoder and its decoding method | |
| CN1273955C (en) | Method and device for coding and/or decoding audip frequency data using bandwidth expanding technology | |
| CN1276406C (en) | Method and apparatus for encoding/decoding audio data with scalability | |
| CN1138254C (en) | Audio signal comprssing coding/decoding method based on wavelet conversion | |
| CN1290078C (en) | Method and device for coding and/or devoding audio frequency data using bandwidth expanding technology | |
| CN1378198A (en) | Position matching method and device for voice synthesizing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20091028 Termination date: 20140917 |
|
| EXPY | Termination of patent right or utility model |