KR100940532B1 - Low bitrate decoding method and apparatus - Google Patents

Abstract

Provided are a low bit rate encoding / decoding method and apparatus. A low bit rate decoding method includes decoding a bit stream; Dequantizing a quantized spectral component with a value other than '0' included in the decoded bitstream; Dequantizing a noise level, which is additional information included in the bitstream; Generating a noise component for a spectral component quantized to '0' using the dequantized noise level; Combining the dequantized spectral component and the generated noise component.

Description

Low bitrate decoding method and apparatus

The present invention relates to a method and apparatus for encoding / decoding, and more particularly, to a method and apparatus for low bit rate encoding / decoding that provides high quality sound by compressing data efficiently at a low bit rate.

The waveform containing the information is an analog signal that is continuous in time and continuous in time. Therefore, A / D (Analog-to-Digital) conversion is required to represent the waveform as a discrete signal. In order to perform the A / D conversion, two processes are required. One is a sampling process for changing the discrete signal of a continuous signal in time, and the other is an amplitude quantization process for limiting the number of possible amplitudes to a finite value. That is, quantization of amplitude is the process of converting input amplitude x (n) to y (n), an element of a finite set of possible amplitudes at time n.

The storage / restoration method of audio signal is also converted into PCM (Pulse Code Modulation) data, which is a digital signal through sampling and quantization through the development of digital signal processing technology. After storing signals on recording / storage media such as Audio Tape, users can replay the stored signals when they need them. Compared with analog methods such as LP (Long-Play Record) and Tape, the digital storage / restore method overcomes the deterioration due to the improvement of sound quality and the storage period. Showed a problem.

To solve this problem, methods such as DPCM (Differential Pulse Code Modulaton) or ADPCM (Adaptive Differential Pulse Code Modulation) have been developed to compress digital voice signals. Efforts have been made to reduce the amount of data in digital voice signals using the above methods, but the efficiency varies greatly depending on the type of signal. The psychoacoustic model of humans is used in the MPEG / audio (Moving Pictures Expert Group) technique, which was recently standardized by the International Standard Organization (ISO), or the AC-2 / AC-3 technique developed by Dolby. We used a method to reduce the amount of data using. This method greatly contributed to reducing the amount of data efficiently regardless of the signal characteristics.

Conventional audio signal compression techniques such as MPEG-1 / audio, MPEG-2 / audio or AC-2 / AC-3 combine signals in the time domain into blocks of a certain size and convert them into signals in the frequency domain. The transformed signal is then scalar quantized using a human psychoacoustic model. This quantization technique is simple but not optimal even if the input samples are statistically independent. Of course, if the input sample is statistically dependent, it is even insufficient. Because of this problem, coding is performed including lossless coding such as entropy coding or some kind of adaptive quantization. Therefore, rather than simply storing only PCM data, the process is considerably more complicated, and the bitstream is composed of additional information for compressing a signal as well as quantized PCM data.

The MPEG / audio standard or AC-2 / AC-3 offers almost the same sound quality as compact discs, with bit rates of 64 kbps to 384 kbps, reduced by 1/6 to 1/8 compared to conventional digital encoding. do. For this reason, the MPEG / audio standard is expected to play an important role in the storage and transmission of audio signals in multimedia systems such as digital audio broadcasting (DAB), internet phones, and audio on demand.

However, such an encoding technique lacks available bits for encoding only by signal quantization at low bit rates of 32 kbps or less. Therefore, there is a need for a technique capable of efficiently compressing an audio signal and restoring it close to the original sound even at a low bit rate.

SUMMARY OF THE INVENTION The present invention has been made in an effort to provide a low bit rate encoding / decoding method and apparatus capable of compressing data efficiently and restoring near an original sound.

According to an aspect of the present invention, there is provided a low bit rate encoding method comprising: (a) converting an input audio signal in a time domain into an audio signal in a frequency domain; (b) extracting an important spectrum component from the audio signal in the frequency domain and quantizing the important spectrum component; (c) extracting residual spectrum components excluding the important spectrum components from the audio signal of the frequency domain, and calculating and quantizing noise levels of the residual spectrum components; And (d) losslessly encoding the quantized significant spectral components and the noise level to output a bitstream. Before the step (b), (e) receiving the audio signal of the time domain further comprising the step of modeling according to the human auditory characteristics and calculating the coded bit allocation information, the steps (b) and (c) In this case, the number of bits may be allocated according to the encoding bit allocation information to quantize the significant spectrum component and the noise level. In addition, the residual spectrum may be set to 0 as the peripheral spectral component of the critical spectrum component. Step (c) comprises: (c1) extracting residual spectrum components excluding the important spectrum components from the audio signal in the frequency domain; (c2) calculating a noise level according to a noise size by dividing the residual spectrum component into predetermined subbands; And (c3) quantizing the calculated noise level. The noise magnitude can be calculated using linear prediction analysis.

According to an aspect of the present invention, there is provided a low bit rate encoding apparatus comprising: a signal converter converting an input audio signal in a time domain into an audio signal in a frequency domain; An important spectrum component processing unit which extracts an important spectrum component from the audio signal of the frequency domain and quantizes the important spectrum component; A noise component processing unit for extracting residual spectrum components excluding the important spectrum components from the audio signal in the frequency domain, and calculating and quantizing noise levels of the residual spectrum components; And a lossless encoding unit for losslessly encoding the quantized significant spectral components and the noise level to output a bitstream. The low bit rate encoding apparatus may further include a psychoacoustic model unit configured to receive audio signals in the time domain according to human auditory characteristics and calculate encoding bit allocation information. The important spectrum component processing unit and the noise component processing unit may include: The number of bits may be allocated according to bit allocation information to quantize the critical spectrum component and the noise level, respectively. The noise component processing unit comprises: a residual spectrum component extracting unit extracting a residual spectrum component except the important spectrum component from the audio signal in the frequency domain; A noise level calculator for dividing the residual spectrum components into predetermined subbands and calculating a noise level according to a noise level; And a noise level quantizer for quantizing the calculated noise level.

According to another aspect of the present invention, there is provided a low bit rate decoding method comprising: (a) outputting a decoded audio signal by lossless decoding on an input bit stream; (b) dequantizing important spectral components which are quantized data in the decoded audio signal; (c) inversely quantizing a noise level which is additional information in the decoded audio signal to generate a noise component; (d) outputting an audio signal in a frequency domain by combining the dequantized significant spectral components and noise components; And (e) converting the audio signal in the frequency domain into an audio signal in the time domain. Step (c) includes: (c1) inverse quantization of a noise level which is additional information in the decoded audio signal; And (c2) generating a noise component from the dequantized noise level by excluding a predetermined region around the important spectral component.

The low bit rate decoding apparatus for achieving the technical problem, lossless decoding unit for outputting the decoded audio signal by lossless decoding on the input bit stream; An important spectral component inverse quantizer for inversely quantizing an important spectrum component which is quantized data in the decoded audio signal; A noise component processor for generating a noise component by inversely quantizing a noise level as additional information among the decoded audio signals; A spectral component combiner for outputting an audio signal in a frequency domain by combining the dequantized important spectral components and noise components; And a signal generator for generating an audio signal in the time domain from the audio signal in the frequency domain. The noise component processing unit comprises: a noise level inverse quantization unit for inversely quantizing a noise level that is additional information in the decoded audio signal; And a noise component generation unit for generating a noise component except for a predetermined region around the important spectral component from the dequantized noise level.

According to the present invention, according to the low bit rate encoding / decoding method and apparatus, by separating and encoding important spectral components and noise components from an audio signal, data can be efficiently compressed and restored close to the original sound.

Hereinafter, a low bit rate encoding / decoding method and apparatus according to the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram showing a configuration of an embodiment of a low bit rate audio encoding apparatus according to the present invention. The signal converter 100, the psychoacoustic model unit 110, the important spectrum portion processor 120, and the noise component are shown in FIG. It comprises a processing unit 130 and a lossless encoding unit 140.

The signal converter 100 converts an audio signal in a time domain into an audio signal in a frequency domain. Such a time / frequency transform can be converted into a Modified Discrete Cosine Transform (MDCT). In addition, the signal conversion unit 110 is divided into subbands for a predetermined frequency component.

The psychoacoustic model unit 110 calculates coded bit allocation information for each subband divided by the signal converter 100 to remove perceptual redundancy due to human auditory characteristics. The psychoacoustic model unit 110 uses a human auditory characteristic to omit detailed information of low sensitivity and to change the bit allocation for each frequency to reduce the code amount, preferably psychoacoustic (psychoacoutic) Calculate bit allocation information in consideration of aspects. The psychoacoustic model unit 110 outputs the coded bit allocation information calculated in this way to the important spectrum component processing unit 120 and the noise component processing unit 130.

The important spectrum component processor 120 extracts and quantizes important spectrum components from the frequency domain audio signal from the signal converter 100, and includes an important spectrum component extractor 121 and an important spectrum component quantizer 122. . The important spectrum component extractor 121 extracts the spectral components determined to be important for each predetermined spectral section. The important spectrum component quantization unit 122 quantizes the extracted important spectrum component at a bit rate according to coding bit allocation information input from the psychoacoustic model unit 110.

The noise component processor 130 extracts a residual spectrum component except the important spectrum component from the audio signal in the frequency domain, and calculates and quantizes a noise level of the residual spectrum component. The noise component processor 140 will be described later in more detail as a core block of the present invention.

The lossless encoding unit 140 receives a quantized audio signal from the important spectrum component processing unit 120 and the noise component processing unit 130 and losslessly encodes the bitstream. At this time, lossless coding such as Huffman coding and arithmetic coding can be used to efficiently compress and encode.

2 is a block diagram illustrating the configuration of the noise component processor 130 shown in FIG. 1, wherein the noise component processor 130 includes a residual spectrum component extractor 200, a noise level calculator 210, and a noise level. A quantization unit 220 is included.

Referring to FIGS. 1 and 2, the residual spectrum component extractor 200 obtains a difference between the original spectrum signal from the signal converter 100 and the important spectrum component signal from the important spectrum component extractor 121. Extract the ingredients. The noise level calculator 210 divides the residual spectrum component into predetermined subbands and calculates a noise level according to the noise size. The noise level quantization unit 220 quantizes the calculated noise level at a bit rate according to encoding bit allocation information input from the psychoacoustic model unit 110.

3 is a flowchart illustrating an operation of a low bit rate encoding apparatus according to the present invention, and will be described with reference to the components of FIG. 1.

1 and 3, in step S300, the signal converter 100 converts an audio signal in a time domain into an audio signal in a frequency domain. Such a time / frequency transform can be converted into a Modified Discrete Cosine Transform (MDCT). In addition, the signal conversion unit 110 is divided into subbands for a predetermined frequency component. The MDCT spectrum X of the audio signal in the frequency domain is shown in Fig. 5A.

In operation S310, the psychoacoustic model unit 110 calculates encoding bit allocation information for each subband divided by the signal conversion unit 100 to remove perceptual redundancy due to human auditory characteristics. Since the psychoacoustic model unit 110 calculates the bit allocation information in consideration of the psychoacoustic aspect, a frequency with high human hearing sensitivity may allocate a lot of bits, and a frequency other than the psychoacoustic model may allocate a bit. .

In operation S320, the important spectrum component processor 120 extracts and quantizes an important spectrum component from the frequency domain audio signal from the signal converter 100. A spectrum Y obtained by extracting only important spectrum components from the MDCT spectrum of FIG. 5A is illustrated in FIG. 5B. At this time, the peripheral spectral region of the important spectrum component is set to zero. Table 1 shows the size (nAround) of the surrounding spectral region set to zero.

In operation S330, the noise component processor 130 extracts a residual spectrum component except for the important spectrum component from the audio signal in the frequency domain, and calculates and quantizes a noise level of the residual spectrum component. This step S330 will be described in more detail later as the core of the present invention.

In operation S340, the lossless encoding unit 140 receives the quantized audio signal from the important spectrum component processing unit 120 and the noise component processing unit 140 and losslessly encodes the bitstream. At this time, lossless coding such as Huffman coding and arithmetic coding can be used to efficiently compress and encode. The output bitstream is composed of quantized data, which is an important spectrum component, and additional information, which is a noise level.

4 is a flowchart illustrating the operation S330 in more detail, and will be described with reference to the components of FIGS. 1 and 2.

1, 2, and 4, in operation S400, the residual spectrum component extracting unit 200 may include a signal between the one-spectrum signal from the signal converter 100 and the important spectrum signal from the important spectrum component extractor 121. Find the difference and extract the residual spectrum components. The remaining spectrum Z extracted from the one spectrum X of FIG. 5A except for the important spectrum Y of FIG. 5B is illustrated in FIG. 5C.

In operation S410, the noise level calculator 210 divides the residual spectrum component into predetermined subbands and calculates a noise level according to the noise size.

The noise magnitude is calculated by performing a linear prediction analysis. Linear predictive analysis is performed using the well-known autocorrelation method. The covariance method and the Durbin's method may be used. Such methods are well known in the art and will not be described here. Through linear prediction, the encoder predicts how much noise is present in the current frame. If the noise component is strong, the noise level is transmitted as it is. If the noise component is small and the tone component is strong, the noise level is relatively reduced. In addition, in the case of a small window, the noise changes abruptly, so in this case, the noise size is additionally reduced.

The noise level can be calculated by the following equation.

Where Energy is the energy for each subband, nCountFreq is the number of nonzero spectral components in that subband, dNoise is the noise size calculated for that subband, and α is a perceptual It means the weight. α decreases the noise level for transient noise (e.g., 0.3) (usually in this case using a short window to convert the data), and for continuous noise (e.g. white noise). The noise level is set high (for example, 0.7) (usually in this case the data is converted using a long window).

In operation S420, the noise level quantization unit 220 quantizes the calculated noise level at a bit rate according to encoding bit allocation information input from the psychoacoustic model unit 110.

6 is a block diagram illustrating a configuration of a low bit rate decoding apparatus according to the present invention, and includes a lossless decoding unit 600, an important spectral component dequantization unit 610, a noise level processing unit 620, and a spectral component combining unit 630. ) And a signal generator 640.

The lossless decoding unit 600 outputs the decoded audio signal with respect to the received bitstream to the critical spectrum component inverse quantizer 610 and the noise level processor 620. That is, the lossless decoder 600 extracts quantized data and additional information from a bitstream having a hierarchical structure.

An important spectrum component inverse quantizer 610 inversely quantizes an important spectrum component in the decoded audio signal.

The noise level processor 620 generates a noise component by inversely quantizing the noise level in the decoded audio signal, and includes a noise level inverse quantizer 621 and a noise component generator 622. The noise level inverse quantizer 621 inversely quantizes the noise level in the decoded audio signal, and the noise component generator 622 excludes a predetermined region around the important spectrum component from the dequantized noise level. Produce ingredients.

The spectral component combiner 630 combines the dequantized significant spectral components and noise components to output an audio signal in a frequency domain.

The signal generator 640 generates an audio signal in the time domain from the audio signal in the frequency domain.

FIG. 7 is a flowchart illustrating an operation of a low bit rate decoding apparatus according to the present invention, and will be described with reference to the components of FIG. 6.

6 and 7, in step S700, the lossless decoder 600 performs an inverse process of the lossless encoder 140 on the received coded bitstream, and as a result, the significant spectrum component is included in the decoded audio signal. The dequantizer 610 and the noise level processor 620 output the result. That is, the lossless decoding unit 600 extracts quantized data and additional information from a bit stream having a hierarchical structure. Lossless decoding can be decoded by the arithmetic decoding method or by the Hoffman decoding method.

In operation S710, the important spectrum component inverse quantization unit 610 inversely quantizes an important spectrum component that is quantized data in the decoded audio signal.

In operation S720, the noise level processor 620 inversely quantizes the noise level of the additional information in the decoded audio signal to generate a noise component. The noise level dequantization unit 621 dequantizes the noise level in the decoded audio signal, and the noise component generation unit 622 excludes the surrounding spectral region of the important spectral component from the dequantized noise level. Create

In operation S730, the spectral component combiner 630 combines the dequantized significant spectral components and noise components to output an audio signal in a frequency domain. In FIG. 5D, a signal spectrum combining important spectral components and noise components is shown. As can be seen from FIG. 5D, the noise component is significantly reduced compared to the one-spectrum signal of FIG. 5A.

In operation S740, the signal generator 640 generates an audio signal in the time domain from the audio signal in the frequency domain.

The present invention can be embodied as code that can be read by a computer (including all devices having an information processing function) in a computer-readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording devices include ROM, RAM, CD-ROM, magnetic tape, floppy disks, optical data storage devices, and the like.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

1 is a block diagram showing a configuration of a low bit rate encoding apparatus according to the present invention.

FIG. 2 is a block diagram showing a more detailed configuration of a noise component processing section in FIG. 1; FIG.

3 is a flowchart illustrating the operation of a low bit rate encoding apparatus according to the present invention.

4 is a flowchart illustrating step S330 in more detail in FIG.

5A to 5D are diagrams showing an example of signal change according to frequency signal processing of the low bit rate audio encoding / decoding apparatus according to the present invention.

6 is a block diagram showing the configuration of a low bit rate decoding apparatus according to the present invention;

7 is a flowchart illustrating the operation of a low bit rate decoding apparatus according to the present invention.

Claims (5)

Decoding the bitstream; Dequantizing a quantized spectral component with a value other than '0' included in the decoded bitstream; Dequantizing a noise level, which is additional information included in the bitstream; Generating a noise component for a spectral component quantized to '0' using the dequantized noise level; And combining the dequantized spectral component and the generated noise component. The method of claim 1, wherein the noise component generation step And generating the noise component from the dequantized noise level except for a predetermined region around the spectral component quantized to a non-zero value. An important spectral component inverse quantizer for decoding a bitstream and inverse quantizing a quantized spectral component with a value other than '0' included in the decoded bitstream; A noise component processor for inversely quantizing a noise level, which is additional information included in the bitstream, and generating a noise component for a spectral component quantized to '0' using the dequantized noise level; And And a spectral component combiner for combining the dequantized spectral component and the noise component to output an audio signal in a frequency domain. The method of claim 3, wherein the noise component processing unit A noise level inverse quantizer for inversely quantizing a noise level which is additional information included in the bitstream; And And a noise component generator for generating the noise component except for a predetermined region around the spectral component quantized to a value other than '0'. Decoding the bitstream; Dequantizing a quantized spectral component with a value other than '0' included in the decoded bitstream; Dequantizing a noise level, which is additional information included in the bitstream; Generating a noise component for a spectral component quantized to '0' using the dequantized noise level; And a program for executing a low bit rate audio signal decoding method comprising combining the dequantized spectral component and the generated noise component.

Images