CN1901042B - Audio data encoding and decoding system and method for making bit rate adjustable - Google Patents

Abstract

A method and apparatus for scalable encoding and decoding of audio data are provided. A method for scalable encoding of audio data includes encoding additional information including scale factor information corresponding to a first layer and coding model information; by referring to the coding model information, in order from symbols formed from the most significant bits (MSB) up to an order of symbols formed by least significant bits (LSBs), encoding a plurality of quantized samples corresponding to the first layer in units of symbols; Multiple layers of encoding. Following this approach, fine-grained scalability (FGS) can be provided with low complexity and good audio quality even at low layers.

Description

Method and device for scalable encoding and decoding of audio data

This application is a divisional application of a patent application entitled "Method and Device for Scalably Encoding and Decoding Audio Data" with the filing date of September 17, 2003 and the application number of 03165036.8.

technical field

The present invention relates to encoding and decoding audio data, and more particularly, to methods and apparatus for encoding audio data such that the encoded audio bitstream has scalable bit rates, and methods and apparatus for decoding audio data.

Background technique

Due to the recent development of digital signal processing technology, audio signals are usually stored as digital signals and reproduced in many cases. A digital audio storage/recovery device converts an audio signal into a pulse code modulation (PCM), that is, a digital signal, by sampling and quantizing. Through such operations, the digital audio storage/reproduction apparatus stores PCM audio data in information storage media such as compact discs (CD) and digital video discs (DVD), and reproduces the stored signals in response to user commands so that the user can listen to the audio data. Digital storage/retrieval methods greatly improve audio quality and significantly reduce degradation caused by long storage periods relative to analog methods using LP recording or magnetic tape. However, digital methods have problems with storage and transmission due to the large amount of digital data.

To solve this problem, various compression methods are used to compress digital audio signals.

In Moving Picture Experts Group (MPEG) standardized by ISO, or AC-2/AC-3 developed by Dolby, the amount of data is reduced using a sound quality model. As a result, the amount of data can be effectively reduced regardless of the characteristics of the signal. That is to say, the MPEG/audio standard or the AC-2/AV-3 method can provide almost the same audio quality as a CD at a bit rate of only 64-384Kbps, which is 1/6 to 1/6 of the bit rate of the previous digital encoding method. 1/8.

However, in these methods, an optimum state suitable for a fixed bit rate is searched for and then quantization and encoding are performed. Therefore, if the transmission bandwidth is reduced due to poor network conditions when the bit stream is transmitted through the network, disconnection may occur and appropriate services can no longer be provided to users. Furthermore, when a bitstream is expected to be transformed into a smaller-sized bitstream more suitable for a mobile device with limited storage capacity, a re-encoding process should be performed to reduce the size of the bitstream and increase the amount of calculation required.

To solve this problem, the applicant of the present invention filed Korean Patent Application No. 97-61298, Nov. 19, 1997, titled "Method for Scalable Bit-Rate Audio Encoding/Decoding Using Bit Slice Algorithm Coding (BSAC)" and device", this patent was authorized on April 17, 2000, Korean Patent No. 261253. According to the BSAC technique, a bit stream encoded with a high bit rate can be changed into a bit stream with a low bit rate, and can be restored using only a part of the bit stream. Therefore, when the network is overloaded, or when the performance of the decoder is poor, or when the user requests a low bit rate, by using only part of the bit stream, it is possible to provide a service with a certain audio quality to the user, although the bit rate decreases with the bit rate. drop, the quality will inevitably drop proportionally.

However, since the BSAC technique employs arithmetic coding, the complexity is high, and the cost increases when the BSAC technique is implemented in an actual device. In addition, since the BSAC technology uses Modified Discrete Cosine Transform (MDCT) to transform the audio signal, the audio quality in the lower layer will be seriously damaged.

Contents of the invention

The invention provides a method and apparatus for encoding and decoding audio data with scalability, by which lower complexity is provided for fine-grained scalability (FGS).

The present invention also provides a method and apparatus for scalable encoding and decoding of audio data, by which better audio quality can be provided at a lower layer even when FGS is provided.

According to an aspect of the present invention, there is provided a method for scalable coding of audio data, comprising coding additional information corresponding to a first layer, comprising scale factor information and coding model information, by referring to the coding model information to obtain from using A number of quantized samples corresponding to the first layer are sequentially coded in symbol units from the symbol formed by the most significant bit (MSB) up to the symbol formed by the least significant bit (LSB), and each time the layer is increased layer by layer The ordinal number of , the steps are repeatedly executed until the encoding of predetermined multiple layers is completed.

According to another aspect of the present invention, there is provided an encoding method, including slicing audio data so that the sliced audio data corresponds to a plurality of layers, obtaining scaling segment information and encoding segments corresponding to each of the plurality of layers Information, based on the scaling segment information and encoding segment information corresponding to the first layer, encode additional information including scaling factor information and coding model information, quantize the audio data corresponding to the first layer by referring to the scaling factor information, and obtain quantized samples , by referring to the encoding model information, the resulting plurality of quantized samples are encoded in units of symbols in order from a symbol formed using the most significant bit (MSB) to a symbol formed using the least significant bit (LSB), and then The sequence numbers of the layers are increased layer by layer each time, and the steps are repeated until the coding of a predetermined number of layers is completed.

The method may further comprise, prior to encoding the additional information, obtaining a range of bits allowed in each of the plurality of layers, wherein in the encoding of the obtained plurality of quantized samples, the number of encoded bits is counted, and if the counted bits If the number of counted bits exceeds the bit range corresponding to bits, the encoding stops, and even after the quantized samples are all encoded, if the number of counted bits is less than the bit range corresponding to bits, the bits that are still unencoded after the lower layer encoding is completed are encoded to the range allowed by the bit range.

In addition, slicing of audio data includes performing wavelet transformation of audio data, and slicing the wavelet-transformed data with reference to a cutoff frequency so that sliced data corresponds to a plurality of layers.

The encoding of audio data may include differential encoding scale factor information and encoding model information.

The encoding of the plurality of quantized samples may include Huffman coding, and the encoding of the plurality of quantized samples may include mapping the plurality of quantized samples on a bit-plane, in order from symbols formed using the most significant bits (MSB) up to Samples are coded in units of symbols within a range of bits allowed in one layer corresponding to the samples using the order of symbols formed of least significant bits (LSBs).

In the mapping of multiple quantized samples, K quantized samples are mapped on the bit plane, and in the encoding of the samples, a scalar value corresponding to a symbol formed by K-bit binary data is obtained, and by referring to the K-bit binary data , the scalar value obtained, and the scalar value corresponding to one symbol higher than the current symbol on the bit-plane perform Huffman encoding, where K is an integer.

According to another aspect of the present invention, there is provided a method for scalable decoding of audio data encoded in a layered structure, comprising decoding an additional Information, in the order from the symbol formed by the most significant bit (MSB) to the symbol formed by the least significant bit (LSB), decode the audio data in units of symbols, and obtain quantized samples by referring to the coding model information, by referring to the scale The obtained quantized samples are inversely quantized by the factor information, the inversely quantized samples are inversely transformed, and the sequence numbers of the layers are increased layer by layer each time, and the steps are repeated until the decoding of predetermined multiple layers is completed.

The decoding of additional information includes differential decoding scaling factors and coding model information.

In decoding audio data, quantized samples are obtained by Huffman decoding. In addition, the decoding of the audio data may further include, in the order from the symbol formed by the most significant bit (MSB) to the symbol formed by the least significant bit (LSB), within the range of bits allowed in the layer corresponding to the audio data in the order of Audio data is decoded in units of symbols, and quantized samples are obtained from bit planes on which the decoded symbols are arranged.

When decoding audio data, a 4*K bit plane formed of decoded symbols is obtained, and when obtaining quantized samples, K quantized samples are obtained from the 4*K bit plane, where K is an integer.

According to another aspect of the present invention, there is provided an apparatus for scalable decoding of audio data encoded in a layered structure, comprising a decapsulation unit that decodes the information containing the scale factor information corresponding to the first layer and The additional information of the encoding model information, and by referring to the encoding model information, decodes the audio data in units of symbols and obtains the quantized a sample; an inverse quantization unit that inversely quantizes the obtained quantized samples with reference to the scale factor information; and an inverse transform unit that inversely transforms the inverse quantized samples.

The unpacking unit differentially decodes the scaling factor information and the encoding model information, and outputs quantized samples through Huffman decoding. The depacking unit decodes the audio data in units of symbols within the bit range allowed in the layer corresponding to the audio data in order from the symbol formed of the most significant bit (MSB) to the symbol formed of the least significant bit (LSB) , and quantized samples are obtained from the bit-plane on which the decoded symbols are arranged. The unpacking unit obtains a 4*K bit plane formed by the decoded symbols, and then obtains K quantized samples from the 4*K bit plane, where K is an integer.

According to another aspect of the present invention, there is provided an apparatus for scalable decoding of audio data, including a transform unit for transforming audio data; a quantization unit for transforming audio corresponding to each layer when quantizing by referring to scale factor information data, and outputs quantized samples; and a packing unit that encodes additional information including scale factor information and coding model information corresponding to each layer, and by referring to the coding model information, according to the symbol formed from the most significant bit (MSB) A plurality of quantized samples from the quantization unit are coded in units of symbols up to the order of symbols formed of least significant bits (LSBs).

The encapsulation unit obtains the scaling section information and the encoding section information corresponding to each of the layers, and encodes additional information including the scaling factor information and the encoding model information based on the scaling section information and the encoding section information corresponding to each layer.

In addition, the encapsulation unit counts the number of encoded bits, and if the counted number of bits exceeds a bit range corresponding to bits, the encoding is stopped, and even after the quantized samples are all encoded, if the counted number of bits is smaller than the bit range corresponding to bits, Encode the bits that are still unencoded after the low-level encoding is completed to the range allowed by the bit range.

The transform unit performs wavelet transform on audio data.

The encapsulation unit slices the wavelet transform data by referring to a cutoff frequency so that the sliced data corresponds to a plurality of layers.

The encapsulation unit differentially encodes scaling factor information and encodes model information.

Encapsulates unit Huffman encoding quantized samples. In particular, the packing element maps a plurality of quantized samples on the bit plane, and in the layer corresponding to the symbol in the order from the symbol formed by the most significant bit (MSB) to the symbol formed by the least significant bit (LSB) Decode symbols in units of symbols within the allowed bit range.

The packing unit maps K quantized samples on the bit plane, obtains scalar values corresponding to symbols formed by K-bit binary data, and by referring to the K-bit binary data, obtains scalar values corresponding to Huffman encoding is performed on a scalar value of a symbol for the current symbol, where K is an integer.

Description of drawings

The above objects and advantages of the present invention will become clearer by describing in detail preferred embodiments of the present invention with reference to the accompanying drawings, wherein:

Fig. 1 is a block diagram of an encoding device according to a preferred embodiment of the present invention;

Fig. 2 is a block diagram of a decoding device according to a preferred embodiment of the present invention;

Fig. 3 is a structural diagram of frames forming a bit stream coded in a hierarchical structure so as to be able to control the bit rate;

FIG. 4 is a detailed diagram of the structure of additional information;

FIG. 5 is a reference diagram for explaining an encoding method according to the present invention;

Fig. 6 is a reference figure, in order to more concretely explain according to the encoding method of the present invention;

Fig. 7 is a flowchart for explaining the coding method according to the preferred embodiment of the present invention;

Fig. 8 is flow chart, is used to explain the decoding method according to the preferred embodiment of the present invention; And

Fig. 9 is a flowchart for explaining a decoding method according to another preferred embodiment of the present invention.

Detailed ways

Referring to Fig. 1, according to the present invention, the encoding device encodes audio data in a hierarchical structure so that the bit rate of the encoded bit stream can be controlled, and includes a transformation unit 11, a sound quality unit 12, a quantization unit 13, and a bit packing unit 14.

Transformation unit 11 receives pulse code modulation (PCM) audio data as a time-domain audio signal, and transforms the signal into a frequency-domain signal, referring to information on a sound quality model provided by sound quality unit 12 . When the difference between the characteristics of the human-perceivable audio signal is not very large in the time domain, in the frequency-domain audio signal obtained by transformation, the difference between the characteristics of the human-perceivable signal and the signal that cannot be perceived by humans have a big difference. Therefore, compression efficiency can be improved by differentiating the number of bits allocated to the respective frequency bands. In the embodiment of the present invention, the transformation unit 11 performs wavelet transformation. In MDCT, even slight distortions can cause degradation that can be perceived by the human ear due to unnecessary high frequency resolution in the low frequency band. However, in wavelet transform, the time/frequency resolution is more appropriate so that it can provide more stable audio quality even in low layers with low frequency bands.

The sound quality unit 12 provides sound quality model information, such as shock information, to the transformation unit 11, and combines the audio signals transformed by the transformation unit 11 into signals of appropriate sub-bands. In addition, the sound quality unit 12 calculates the masking threshold in each sub-band by using the masking effect caused by the interaction between the respective signals, and provides the threshold value to the quantization unit 13 . The masking threshold is the maximum value of a signal that cannot be perceived by humans due to the interaction between the signals. In this embodiment, the sound quality unit 12 calculates the masking threshold of the stereo component through binaural masking level depression (BMLD).

The quantization unit 13 scalarizes the audio signal in each frequency band according to the scaling factor information corresponding to the audio signal, so that the level of quantization noise in the frequency band is smaller than the masking threshold provided by the sound quality unit 12, so that people cannot perceive the noise. Next, the quantization unit 13 outputs the quantized samples. That is, by using the masking threshold calculated in the sound quality unit 12 and a noise-masking ratio (NMR) generated for each frequency band as a noise ratio, the quantization unit 13 performs quantization so that the NMR value in all frequency bands is 0 dB or less. An NMR value of 0 dB or less means that humans cannot perceive quantization noise.

The bit packing unit 14 encodes quantized samples and additional information belonging to each layer, and packs the encoded signal in a hierarchical structure. The additional information includes scaling segment information in each layer, encoding segment information, their scaling factor information, and encoding model information. The scale segment information and the code segment information may be encapsulated into header information, and then sent to the decoding device. Otherwise, the scaling segment information and the encoding segment information may be encoded and encapsulated into additional information of each layer, and then sent to the decoding device. Scale section information and coded section information may not be transmitted to the decoding device because they are pre-stored in the decoding device in some cases.

More specifically, when encoding additional information including scaling factor information and coding model information corresponding to the first layer, the bit packing unit 14 refers to the coding model information corresponding to the first layer, according to the most significant bit ( The encoding of samples and information is performed in units of symbols in the order from a symbol formed by MSB) to a symbol formed by least significant bit (LSB). Then, in the second layer, the same processing is repeatedly performed. That is, encoding is performed as the number of layers increases until encoding of a plurality of predetermined layers is completed. In this embodiment, the bit packing unit 14 differentially encodes the scale factor information and the encoding model information, and Huffman encodes the quantized samples. The hierarchical structure of the bitstream encoded according to the present invention will be explained later.

The scale segment information refers to information for performing quantization more appropriately in accordance with the frequency characteristics of the audio signal. When the frequency region is divided into a plurality of frequency bands and an appropriate scaling factor is assigned to each frequency band, the scale band information indicates the scale band corresponding to each layer. Thus, each layer belongs to at least one calibration segment. Each scaling segment has an assigned scaling factor. Also, the encoding segment information refers to information for performing encoding more appropriately according to the frequency characteristics of the audio signal. The coding segment information indicates a coding segment corresponding to each layer when a frequency region is divided into a plurality of frequency bands and an appropriate coding model is allocated to each frequency band. The scaling segment and the encoding segment are divided based on experience, and the corresponding scaling factor and encoding model are determined based on the same method.

Fig. 2 is a block diagram of a decoding apparatus according to a preferred embodiment of the present invention.

Referring to FIG. 2, the decoding device decodes the bit stream to a target layer determined by network conditions, performance of the decoding device and user's selection, so that the bit rate of the bit stream can be controlled. The decoding device includes a decapsulation unit 21 , an inverse quantization unit 22 , and an inverse transformation unit 23 .

The decapsulation unit 21 decapsulates the bitstream to the target layer, and decodes the bitstream in each layer. That is, additional information including scale factor information and coding model information corresponding to each layer is decoded, and then based on the obtained coding model information, coded quantized samples belonging to the layer are decoded and quantized samples are restored. In this embodiment, the unpacking unit 21 differentially decodes the scaling factor information and the encoding model information, and Huffman decodes the encoded quantized samples.

At the same time, the scaling section information and the coding section information are obtained from the header information of the bit stream, or by decoding the additional information in each layer. Alternatively, the decoding device may store the scaling segment information and the encoding segment information in advance. The inverse quantization unit 22 inverse quantizes and restores the quantized samples in each layer according to the scale factor information corresponding to the samples. The inverse transform unit 23 frequency/time maps the restored samples to transform the samples into PCM audio data in the time domain, and outputs it.

FIG. 3 is a structural diagram of frames forming a bit stream encoded in a hierarchical structure so that a bit rate can be controlled.

Referring to FIG. 3, a frame of a bitstream according to the present invention is encoded into a hierarchical structure by mapping quantized samples and additional information to achieve fine-grained scalability (FGS). In other words, the lower layer bitstream is included in the hierarchically structured enhancement layer bitstream. Additional information required in each layer is assigned to each layer and then encoded.

A header area for storing header information is placed in front of the bitstream, then information about layer 0 is packed after the header area, and then information belonging to layers 1-N as enhancement layers is packed in order. The layer from the header area to layer 0 information is called base layer, the layer from the header area to layer 1 information is called layer 1, and the layer from the header area to layer 2 information is called layer 2. Also, the uppermost layer represents information from the header area to layer N, that is, from the base layer to layer N as an enhancement layer. Additional information and coded audio data are stored as each layer information. For example, additional information 2 and coded quantized samples are stored as layer 2 information. Here, N is an integer greater than or equal to 1.

FIG. 4 is a detailed diagram of the structure of additional information.

Referring to FIG. 4, additional information and coded quantized samples are stored as arbitrary additional information, and in this embodiment, the additional information includes Huffman coding model information, quantization factor information, additional information about channels, and other additional information. The Huffman coding model information is index information of the Huffman coding model that should be used to encode or decode quantized samples belonging to a layer corresponding to the information. The quantization factor information indicates a quantization step size for quantizing or dequantizing audio data belonging to a layer corresponding to the information. The additional information on the channel is information on the channel such as M/S stereo. Other additional information is flag information on whether to use M/S stereo.

In the present embodiment, the bit packing unit 14 performs differential encoding on Huffman encoding model information and quantization factor information. In differential encoding, the differential value of the value of an immediately preceding band is encoded. Additional information about the channel is Huffman coded.

FIG. 5 is a reference diagram for more specifically explaining the encoding method according to the present invention.

Referring to FIG. 5, quantized samples to be encoded have a 3-layer structure. Slashed rectangles indicate spectral lines including quantized samples, solid lines indicate scaled segments, and dashed lines indicate encoded segments. Scaling segments (1), (2), (3), (4) and (5) and coding segments (1), (2), (3), (4) and (5) belong to layer 0. Scaling segments (5) and (6) and encoding segments (6), (7), (8), (9) and (10) belong to layer 1. Scaling segments (6) and (7) and coding segments (11), (12), (13), (14) and (15) belong to layer 2. Meanwhile, layer 0 is defined so that encoding is performed up to band (a), layer 1 is defined so that encoding is performed up to band (b), and layer 2 is defined so that encoding is performed up to band (c).

First, the quantized samples belonging to layer 0 are coded in the bit range of 100 using the corresponding coding model. Furthermore, as additional information of layer 0, the scaled segments (1), (2), (3), (4) and (5) and the encoded segments (1), (2), (3) belonging to layer 0, (4) and (5) are coded. When encoding quantized samples in units of symbols, the number of bits is counted. If the counted number of bits exceeds the allowable bit range, encoding of layer 0 is stopped, and layer 1 is arithmetic-encoded. Among quantized samples belonging to layer 0, uncoded quantized samples are coded next time when there is still room for the allowed number of bits in layers 0 and 1.

Next, the quantized samples belonging to layer 1 are coded using the coding segments belonging to layer 1, that is, the quantizations to be coded in coding segments (6), (7), (8), (9) and (10) The coding model of a coding segment to which the sample belongs. Furthermore, as additional information of layer 1, scaling segments (5) and (6) and encoding segments (6), (7), (8), (9) and (10) belonging to layer 1 are coded. Even after encoding all samples corresponding to layer 1, if there is still room in the allowed bit range, i.e. 100 bits, the remaining unencoded bits in layer 0 are encoded until the allowed bits, i.e. 100 bits, are counted. If the number of bits counted for encoding exceeds the allowable bit range, encoding of layer 1 is stopped, and encoding of layer 2 starts.

Finally, the quantized samples belonging to layer 2 are coded using the coding section belonging to layer 2, i.e. the quantized samples to be coded in coding sections (11), (12), (13), (14) and (15) belong to An encoding model for an encoded segment. Furthermore, as additional information of layer 2, scaling segments (6) and (7) and encoding segments (11), (12), (13), (14) and (15) belonging to layer 2 are coded. Even after encoding all samples corresponding to layer 2, if there is still room in the allowed bit range, ie 100 bits, the remaining uncoded bits in layer 0 are encoded until the allowed bits, ie 100 bits, are counted.

If all quantized samples are coded regardless of the allowed bit range of layer 0, i.e. if all quantized samples are coded, even after the number of coded bits exceeds the allowed bit range, i.e. 100 (which means the next layer, i.e. layer 1 Some bits in the allowable bit range of the layer are used to code the current layer), it is usually the case that quantized samples belonging to layer 1 cannot be coded. Therefore, in the case of scalable decoding, if decoding is performed on layers ranging up to layer 1, since all quantized samples corresponding to the range of layer 1 to a predetermined frequency band (b) are not coded, at layers lower than (b) The decoded quantized samples fluctuate in frequency, causing the "Birdy" effect, where the audio quality deteriorates.

When determining among the layers (target layers), bit ranges are allocated, taking into account the overall size of all audio data to be encoded. In this way, there is no possibility that encoding will not be performed due to a defect in the bit range in which bits to be encoded are arranged.

When decoding is performed in the reverse manner to the encoding process, the number of bits is counted in the allowed bit range. Therefore, a decoding timing point of a predetermined layer can be identified.

FIG. 6 is a reference diagram for more specifically explaining the encoding method according to the present invention.

According to the present invention, the bit packing unit 14 performs encoding on quantized samples corresponding to each layer by bit-plain encoding and Huffman encoding. A number of quantized samples are mapped on a bit-plane for subsequent representation in binary form and are coded in order from symbols formed by MSBs up to symbols formed by LSBs within the allowed bit range of each layer. Important information on the bit plane is encoded first, while less important information is encoded later. By doing so, the bit rate and frequency band corresponding to each layer are fixed in the encoding process, making it possible to reduce distortion called "Birdy effect".

FIG. 6 illustrates an encoding example in the case where the number of bits of a symbol including MSB is 4 or less. When quantized samples 9, 2, 4, and 0 are mapped on the bit-plane, they are represented in binary form, ie, 1001b, 0010b, 0100b, and 0000b, respectively. That is, in this embodiment, the size of a coding block as a coding unit on a bit plane is 4*4.

The symbol msb formed from the MSB is "1001b", the symbol msb-1 formed from the next MSB is "0010b", the symbol msb-2 formed from the next MSB is "0100b", and the symbol msb-3 formed from the LSB is "1000b".

Huffman model information for Huffman coding, i.e. codebook index is shown in Table 1:

Table 1

Additional Information Significance Huffman model 0 0 0 1 1 1 2 1 2 3 2 3 4 4 2 5 6 5 3 7 8 9

6 3 10 11 12 7 4 13 14 15 16 8 4 17 18 19 20 9 5 * 10 6 *

11 7 * 12 8 * 13 9 * 14 10 * 15 11 * 16 12 * 17 13 * 18 14 * * * *

According to Table 1, there are even two models for the same effective level (msb in this example). This is because the two models were produced for quantized samples that exhibit different distributions.

The process for encoding according to the example of FIG. 6 of Table 1 will now be explained in more detail.

In the case where the number of bits of a symbol is 4 or less, Huffman coding according to the present invention is shown in Equation 1:

Huffman code value=Huffman codebook [codebook index] [higher bit plane] [symbol]...(1)

That is, Huffman coding uses 3 input variables, including codebook index, higher plane, and sign. The codebook index represents the value obtained from Table 1, and the higher bit-plane represents the symbol on the bit-plane immediately above the symbol currently desired to be coded. symbol represents the symbol currently expected to be encoded.

Since the msb of the Huffman model is 4 in the example in Figure 6, choose 13-16 or 17-20. If the encoded additional information is 8,

The codebook index of the symbol formed by msb bits is 16,

The codebook index of the symbol formed by msb-1 bits is 15,

The codebook index of the symbol formed by msb-2 bits is 14, and

A codebook index of a symbol formed of msb-3 bits is 13.

Meanwhile, since the symbol formed by receiving msb bits does not have the data of the higher bit plane, if the value of the higher bit plane is 0, encoding is performed with the code of the Huffman codebook [16][0b][1000b]. Since the higher bit plane of the symbol formed by msb-1 bits is 1000b, encoding is performed with the codes of the Huffman codebook [15][1000b][0010b]. Since the higher bit plane of the symbol formed by msb-2 bits is 0010b, encoding is performed with the codes of the Huffman codebook [14][0010b][0100b]. Since the higher bit plane of the symbol formed by msb-3 bits is 0100b, encoding is performed with the codes of the Huffman codebook [13][0100b][1000b].

The bit packing unit 14 counts the number of encoded bits, compares this count with the number of bits allowed to be used in the layer, and stops encoding if the count is greater than the allowed number. When space is allowed in the next layer, the remaining bits that were not coded are coded and put into the next layer. After the quantized samples allocated to the corresponding layer are all coded, if there is still room in the number of bits allowed in the layer, ie, if there is room in the layer, the quantized samples in the lower layer that are still uncoded after the coding is completed are coded.

Meanwhile, if the number of bits of the symbol formed by msb is greater than or equal to 5, the position on the current bit plane is used to determine the Huffman code value. In other words, if the significance is greater than or equal to 5, there is only a small statistical difference in the data on each bit plane, and the data is Huffman coded using the same Huffman model. That is, there is a Huffman pattern for each bit-plane.

If the validity is greater than or equal to 5, that is to say, the number of bits of the symbol is greater than or equal to 5, the Huffman coding of the present invention satisfies formula 2:

Huffman code = 20+bp1 ...2

where 'bp1' represents the index of the bit-plane expected to be currently coded, and is an integer greater than or equal to 1. , as listed in Table 2, the constant 20 is a value added to indicate that the index starts from 21, because the last index of the Huffman model (corresponding to the additional number 8) is 20. Therefore, the additional information for a coded segment simply indicates availability. In Table 2, the Huffman model is determined by the index of the bit-plane expected to be currently coded.

Table 2

Additional Information Effectiveness Huffman model 9 5 21-25 10 6 21-26 11 7 21-27 12 8 21-28 13 9 21-29 14 10 21-30 15 11 21-31 16 12 21-32 17 13 21-33 18 14 21-34 19 15 21-35

For the quantization factor information and the Huffman model information in the additional information, DPCM is performed on the coded section corresponding to the information. When the quantization factor information is encoded, 8 bits are used to represent the initial DPCM value in the header information of the frame. The initial value of DPCM for Huffman model information is set to 0.

The differences between the encoding methods of the BSAC technique according to the present invention and the prior art are listed below. First, in the BSAC technique, encoding is performed in units of bits, whereas encoding is performed in units of symbols in the present invention. Second, in the BSAC technique, arithmetic coding is used, while Huffman coding is used in the present invention. Arithmetic coding provides high compression gains at the expense of increased complexity and cost. Therefore, in the present invention, data is not coded in units of bits, but is coded by Huffman in units of symbols in order to reduce complexity and cost.

In order to control the bit rate, that is to say to provide scalability, the bit stream corresponding to one frame is cut off, taking into account the number of bits allowed in each layer, so that only a small amount of data is utilized but still decodable. For example, if only a bit stream corresponding to 48 kbps is desired to be decoded, only 1048 bits of the bit stream are used, enabling decoding audio data corresponding to 48 kbps to be obtained.

An encoding and decoding method according to the present invention based on the above structure will now be explained.

The encoding device reads the PCM audio data, stores the data in a memory (not shown), and obtains the masking threshold and additional information from the stored PCM audio data through sound quality modeling. Since PCM audio data is a time-domain signal, the PCM audio data is wavelet-transformed into a frequency-domain signal. Next, the encoding device obtains quantized samples by quantizing the wavelet-transformed signal according to the quantization segment information and the quantization factor information. Quantized samples are encoded and packed by bit-slice encoding, symbol-unit based encoding, and Huffman encoding, as described above.

Fig. 7 is a flowchart for explaining an encoding method according to a preferred embodiment of the present invention.

Referring to Fig. 7, the process of encoding and packing quantized samples by the bit packing unit 14 of the encoding device will now be explained.

First, the bit packing unit 14 extracts information corresponding to each layer according to the supplied target bit rate and additional information. This processing is performed in steps 701 to 703 . More specifically, in step 701, the cut-off frequency used as the basis for the cut-off of each layer is obtained, in step 702, the quantized segment information and the encoded segment information corresponding to each layer are obtained, and in step 703, a bit range is allocated, within the range , bits that should be coded can be coded in each layer.

Next, in step 704 , the layer index is set to the base layer, and additional information (including quantization section information and coding section information) is encoded in step 705 .

Next, the quantized samples corresponding to the base layer are mapped on the bit plane, and coded in units of 4*4 blocks according to the symbol formed by msb bits at step 706 . In step 707, the number of encoded bits is counted, and if the count exceeds the bit range of the current layer, the encoding of the current layer is stopped, and encoding is started at the next layer. If the number of bits counted in step 707 does not exceed the bit range, then in step 709, the process returns to step 705 to process the next layer. Since the base layer has no lower layers, step 708 is not performed, but step 708 is performed for the layers following the base layer. Through the above steps, all ranges of layers up to the target layer are encoded.

Step 706, that is, the step for encoding quantized samples is as follows:

1. Quantized samples corresponding to one layer are grouped in units of N samples and mapped on a bit plane.

2. Perform Huffman encoding according to the symbols formed by the msb bits of the mapped binary data.

Sub-step 2 can be explained in detail as follows:

2.1 A scalar value (curVal) corresponding to the symbol desired to be encoded is obtained.

2.2 A Huffman code is obtained corresponding to a scalar value (upperVal) corresponding to a symbol in a higher bit-plane, that is, a symbol at a higher position in the bitstream than the symbol expected to be currently coded .

For the quantization factor information and the Huffman model information in the additional information, DPCM is performed on the coded section corresponding to the information. When quantization factor information is encoded, the initial value of DPCM is represented by 8 bits in header information of a frame. The initial value of DPCM for Huffman model information is set to 0.

Fig. 8 is a flowchart for explaining a decoding method according to a preferred embodiment of the present invention.

Referring to FIG. 8, a decoding device receives a bitstream formed of audio data encoded in a hierarchical structure, and decodes header information in each frame. Next, in step 801, additional information including scale factor information and coding model information corresponding to the first layer is decoded. In step 802, referring to the encoding model information, quantized samples are obtained by decoding the bit stream in units of symbols in the order from symbols formed of MSB bits to symbols formed of LSB bits. In step 803, the obtained quantized samples are inversely quantized by referring to the scale factor information, and in step 804, the inversely quantized samples are inversely transformed. As the ordinal number of the layer is increased layer by layer each time, steps 801-804 are repeatedly executed until the coding of a predetermined number of layers is completed.

Fig. 9 is a flowchart for explaining a decoding method according to another preferred embodiment of the present invention.

Referring to FIG. 9, a bit stream formed of audio data encoded in a hierarchical structure is received, and in step 901, a cutoff frequency corresponding to each layer is decoded according to header information in each frame. In step 902, through decoding, the quantized section information and coding section information corresponding to each layer are identified according to the header information. At step 903, the allowed usage bit ranges for each layer are identified. At step 904, the layer index is set to the base layer. Step 905 decodes the additional information on the base layer, and in step 906 quantized samples are obtained by decoding the bitstream symbol-wise into the allowed bit range for each layer in the order from the symbol formed by the MSB bits up to the symbol formed by the LSB bits . In step 907, it is checked whether the current layer is the last one. As the number of layers increases one by one, steps 905 and 906 are repeatedly executed on each layer until a predetermined target layer is reached. In steps 901-903, the decoding device may have the cutoff frequency, quantized section information, coded section information and bit range in advance instead of obtaining these information according to the header information stored in each frame of the received bitstream. In this case, by reading the stored information, the decoding means obtains the information.

As described above, according to the present invention, by encoding bits in units of symbols after bit slicing is performed, scalability whereby the bit rate can be controlled in a top-down manner is provided so that the computational load of the encoding device is not much larger than that without providing the scalability. stretchable device. That is, according to the present invention, there is provided a method and apparatus for encoding and decoding audio data with scalability, wherein the complexity is low while providing FGS and good audio quality even at low layers.

In addition, compared to the MPEG-4 audio BSAC technology using arithmetic coding, the codec device of the present invention using Huffman coding reduces the amount of computation for encapsulation/decapsulation processing down to one-eighth of the BSAC technology . Even when bit packing according to the present invention is performed to provide FGS, the overhead is small so that the coding gain is the same as when no scalability is provided.

Furthermore, since the apparatus according to the present invention has a layered structure, the process of regenerating the bit stream to enable the server side to control the bit rate is simple, and thus the complexity of the apparatus for transform coding is low.

When sending audio streams over the network, the transmission bit rate can be controlled according to user choice or network conditions so that uninterrupted service can be provided.

When an audio stream is stored in an information storage medium having a limited capacity, the size of a file can be arbitrarily controlled and stored. If the bit rate becomes lower, the frequency band is restricted. Thus, the complexity of the filter, which is the most complex device in the codec, is greatly reduced, and inversely proportional to the bit rate, the actual complexity of the coder device is reduced.

By using wavelet transform, the time/frequency domain resolution is higher than that of the state-of-the-art MDCT-based coding, so that better audio quality is provided even at lower layers.

Claims (23)

1. method that is used for the telescopically coding audio data comprises:

Coding comprises corresponding to the scaling factor information of ground floor and the additional information of encoding model information;

By reference encoding model information,, be a plurality of quantized samples of unit Huffman encoding corresponding to ground floor with the symbol according to from the symbol that forms by highest significant position (MSB) order up to the symbol that forms by least significant bit (LSB) (LSB); With

Along with each ordinal number that successively increases layer, repeat described step, up to the coding of finishing predetermined a plurality of layers.

2. coding method comprises:

The section voice data makes the voice data of cutting into slices corresponding to a plurality of layers;

Acquisition is corresponding to a plurality of layers each calibration segment information and coding section information;

Based on calibration segment information and the coding section information corresponding to ground floor, coding comprises the additional information of scaling factor information and encoding model information;

By the voice data of reference scaling factor information quantization, obtain quantized samples corresponding to ground floor;

By reference encoding model information,, be that a plurality of quantized samples that the Huffman sign indicating number is obtained are compiled by unit with the symbol according to from the symbol that forms by highest significant position (MSB) order up to the symbol that forms by least significant bit (LSB) (LSB); With

Along with each ordinal number that successively increases layer, repeat described step, up to the coding of finishing a plurality of layers.

3. according to the method for claim 2, further comprise, before the additional information of coding,

The bit range that acquisition is allowed in each of a plurality of layers, wherein when a plurality of quantized samples that coding is obtained, the number of the bit of coding is counted, if and the bit of counting outnumber bit range corresponding to this bit, coding stops, even and after quantized samples is encoded entirely, if the number of bit of counting less than bit range corresponding to this bit, still uncoded bit is encoded into the scope of bit range permission after the low layer coding is done.

4. 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, the data of section wavelet transform make the data of cutting into slices corresponding to a plurality of layers.

5. according to the process of claim 1 wherein that the coding of voice data comprises differential coding scaling factor information and encoding model information.

6. according to the process of claim 1 wherein that the coding of a plurality of quantized samples comprises:

The a plurality of quantized samples of mapping on bit-planes; With

According to from the symbol that forms by the MSB bit order, be the unit encoding sample with the symbol in the bit range that in layer, is allowed corresponding to sample up to the symbol that forms by the LSB bit.

7. according to the method for claim 6, wherein when a plurality of quantized samples of mapping, K quantized samples is mapped on the bit-planes, and in the coding of sample, obtain scalar value, and pass through with reference to K bit-binary data corresponding to the symbol that forms by K bit-binary data, the scalar value that obtains, with corresponding to the scalar value that is higher than a symbol of current symbol on the bit-planes, carry out Huffman encoding, wherein K is an integer.

8. method that is used for telescopically decoding with the voice data of hierarchy coding comprises:

Decoding comprises corresponding to the scaling factor information of ground floor and the additional information of encoding model information;

By reference encoding model information, according to from the symbol that forms by the MSB bit order, be unit Huffman decoding voice data, and obtain quantized samples with the symbol up to the symbol that forms by the LSB bit;

The quantized samples that is obtained by reference scaling factor information inverse quantization;

The sample of this inverse quantization of reciprocal transformation; With

Along with each ordinal number that successively increases layer, repeat described step, up to the decoding of finishing predetermined a plurality of layers.

9. method according to Claim 8, wherein the decoding of additional information comprises differential decoding scaling factor information and encoding model information.

10. method according to Claim 8, wherein the decoding of voice data further comprises:

According to from the symbol that forms by the MSB bit order, be the unit decoding audio data with the symbol in the bit range that in layer, is allowed corresponding to voice data up to the symbol that forms by the LSB bit; With

Arranged the bit-planes of decoding symbols to obtain quantized samples from it.

11., wherein when decoding audio data, obtain the 4*K bit-planes that forms by decoding symbols, and from the quantized samples that obtains, from K quantized samples of 4*K bit-planes acquisition, wherein K is an integer according to the method for claim 10.

12. a device that is used for the telescopically decoding with the voice data of hierarchy coding comprises:

The deblocking unit, its decoding comprises corresponding to the scaling factor information of ground floor and the additional information of encoding model information, and pass through with reference to encoding model information, according to from the symbol that forms by the MSB bit order up to the symbol that forms by the LSB bit, with the symbol is unit Huffman decoding voice data, and obtains quantized samples;

The inverse quantization unit, its quantized samples by being obtained with reference to scaling factor information inverse quantization; With

The reciprocal transformation unit, this inverse quantization sample of its reciprocal transformation.

13. according to the device of claim 12, wherein deblocking unit differential decoding scaling factor information and encoding model information.

14. device according to claim 12, wherein the deblocking unit is according to from the symbol that formed by the MSB bit order up to the symbol that is formed by the LSB bit, in corresponding to the layer of voice data, be the unit decoding audio data with the symbol in the bit range that allowed, arranged the bit-planes of decoding symbols to obtain quantized samples from it.

15. according to the device of claim 12, wherein the deblocking unit obtains the 4*K bit-planes that formed by decoding symbols, and from K quantized samples of 4*K bit-planes acquisition, wherein K is an integer.

16. a device that is used for the telescopically coding audio data comprises:

The converter unit of converting audio frequency data;

Quantifying unit, it quantizes the converting audio frequency data corresponding to every layer by with reference to scaling factor information, and the output quantized samples; With

Encapsulation unit, its coding comprises corresponding to the scaling factor information of ground floor and the additional information of encoding model information, by reference encoding model information, according to from the symbol that forms by highest significant position (MSB) order, be a plurality of quantized samples of unit Huffman encoding from quantifying unit with the symbol up to the symbol that forms by least significant bit (LSB) (LSB).

17. device according to claim 16, wherein encapsulation unit obtains each calibration segment information and the coding section information corresponding to a plurality of layers, and based on the additional information that comprises scaling factor information and encoding model information corresponding to every layer calibration segment information and coding section information coding.

18. device according to claim 16, wherein encapsulation unit is counted the number of the bit of coding, if and the bit of counting outnumber bit range corresponding to this bit, coding stops, even and after quantized samples is encoded entirely, if the number of the bit of counting is less than the bit range corresponding to this bit, still uncoded bit is encoded into the scope that bit range allows after the coding of low layer is finished.

19. according to the device of claim 16, wherein converter unit is carried out wavelet transform to voice data.

20. according to the device of claim 16, wherein encapsulation unit makes the data of cutting into slices corresponding to a plurality of layers by the data with reference to cutoff frequency section wavelet transform.

21. according to the device of claim 16, wherein encapsulation unit differential coding scaling factor information and encoding model information.

22. device according to claim 16, wherein encapsulation unit shines upon a plurality of quantized samples on bit-planes, and according to from the symbol that forms by the MSB bit order, be the unit encoding sample with the symbol in the bit range that in layer, is allowed corresponding to sample up to the symbol that forms by the LSB bit.

23. device according to claim 16, wherein encapsulation unit shines upon K quantized samples on bit-planes, acquisition is corresponding to the scalar value of the symbol that is formed by K bit-binary data, and pass through with reference to K bit-binary data, the scalar value that obtains, with carry out Huffman encoding corresponding to the scalar value that is higher than a symbol of current symbol on the bit-planes, wherein K is an integer.

Images