WO2009066869A1

WO2009066869A1 - Frequency band determining method for quantization noise shaping and transient noise shaping method using the same

Info

Publication number: WO2009066869A1
Application number: PCT/KR2008/005918
Authority: WO
Inventors: Taejin Lee; Minje Kim; Seungkwon Beack; Dae-Young Jang; Kyeongok Kang; Jeong-Il Seo; Jinwoo Hong; Hochong Park; Rin-Chul Kim; Jeong-Geun Kim; Youngcheol Park
Original assignee: Electronics And Telecommunications Research Institute; Kwangwoon University Industry-Academic Collaboration Foundation
Priority date: 2007-11-21
Filing date: 2008-10-09
Publication date: 2009-05-28

Abstract

A frequency band determining method for quantization noise shaping includes checking whether audio signals obtained from low-pass filtering are transient, determining a predetermined frequency band to be applied as a frequency band to be applied for quantization noise shaping when the audio signals are not transient, and determining an extended frequency band extended more than the predetermined frequency band to be applied as a frequency band to be applied when the audio signals are transient.

Description

FREQUENCY BAND DETERMINING METHOD FOR QUANTIZATION NOISE SHAPING AND TRANSIENT NOISE SHAPING METHOD USING THE SAME

Technical Field

[1] The present invention relates to a frequency band determining method for quantization noise shaping and a transient noise shaping method using the same. More particularly, the method shapes the noise using a long block and reduces pre-echo and musical noise by figuring out whether an applied frequency band is a general frequency band and an extended frequency band based on whether audio signals are transient.

[2] This work was supported by the IT R&D program for MIC/IITA [2007-S-005-01,

"Development of Richmedia Broadcasting Technologies through Advanced Audio and Video Codec Technologies"].

[3]

Background Art

[4] In High Efficiency Advanced Audio Coding (HE-AAC) technology, Temporal Noise

Shaping (TNS) algorithm, which is one of algorithms for shaping quantization noise, is used to effectively represent transient signals. Accordingly, the TNS algorithm reduces pre-echo.

[5] However, despite the use of TNS algorithm, the pre-echo and the musical noise frequently occur at low bit rates.

[6] To acquire perceptually transparent coding sound quality in the HE-ACC audio coding technology, the quantization noise should not exceed a masking threshold value. However, in a perceptional coding technology using a frequency signal analysis method, the quantization noise is coded and then widely spread in the time domain. Thus, at low bit rates, it is hard to satisfy a condition that the quantization noise does not exceed the masking threshold value in the time domain.

[7] For instance, in the AAC audio coding technology generally using 1,024 Modified

Discrete Cosine Transform (MDCT) coefficients, quantization noise with a 48kMz sampling rate is distributed over 40 ms. This distribution may cause audible artifacts when the signals are transient. At this time, the quantization noise can be percep- tionally detected before the transient signals are generated. This quantization noise is called pre-echo phenomenon.

[8] In the TNS algorithm which is devised to effectively deal with the pre-echo phenomenon, shape of the quantization noise widely spread in the time domain is adjusted to have a masking effect.

[9] The TNS algorithm uses a Linear Predictive Coding (LPC) based on a duality between the time domain and a frequency domain. Table 1 explains an optimal coding method for tone signals and transient signals in consideration of the duality.

[10] Table 1 [Table 1] [Table ]

[H] [12] That is, in the frequency domain, the optional coding method for the tone signals with a certain frequency is a direct coding method using a frequency coefficient coding. In the time domain, the optimal coding method for the tone signals with a certain frequency is a predictive coding method using an LPC coding.

[13] On the contrary, when the duality is considered, in the frequency domain, the optimal coding method for the transient signals is the predictive coding method using the frequency coefficient predictive coding. In the time domain, the optional coding method for the transient signals is the direct coding method using the time sample coding.

[14] The TNS algorithm is applied based on the predictive coding method in the frequency domain.

[15] Table 2 shows the frequency band applying the TNS algorithm. [16] Table 2 [Table 2] [Table ]

[17]

[18] Herein, the TNS frequency range (band) is classified according to block length into a long block and a short block. The frequency band applying the TNS algorithm is over 1,275 Hz for the long block, and over 2,750 Hz for the short block.

[19] That is, in the long block, the TNS algorithm is applied to a frequency range from a frequency band of 1,800 Hz to a boundary frequency where a Spectrum Band Replication (SBR) starts. On the other hand, in the short block, the TNS algorithm is applied to a frequency range from a frequency band of 2,750 Hz to the boundary frequency where the SBR begins. In a band lower than the above frequency bands, the pre-echo occurs frequently.

[20] To reduce the pre-echo more, a block switching is performed. The block switching indicates the method of replacing a long window having one-frame length with a short window having 1/8 frame length. The block switching between the long block and the short block is for perceptionally improving the pre-echo by applying the quantization noise effect only in the short block.

[21] However, when the signals are stable and the bit rate is low, the short window may cause an opposite effect. Since the bit is insufficient in the low bit rate, frequency components lost in each short block are shown as spectral holes. The spectral holes are discontinuously connected on a time axis in a corresponding frame to cause the musical noise. That is, in the low bit rate with insufficient bit, when the long block is used instead of the short block, the pre-echo occurs. Also, in the low bit rate with insufficient bit, when the short block is overly used, the musical noise occurs.

[22]

Disclosure of Invention Technical Problem

[23] An embodiment of the present invention is directed to providing a frequency band determining method for quantization noise shaping and transient noise shaping method using the same.

[24] In conventional Temporal Noise Shaping (TNS) technology, when the bit rate is low, pre-echo and musical noise may not be prevented at transient blocks. An object of the present invention is to solve the above problems. [25] Thus, this invention provides a method for determining frequency range for quantization noise shaping by using a long block according to an applied frequency band is classified into a general frequency band and an extension frequency band according to whether audio signals are transient to effectively reduce the pre-echo and the musical noise, and transient noise shaping method using the same.

[26] The objects of the present invention are not limited to the above-mentioned ones.

Other objects and advantages of the present invention can be understood by the following description, and become apparent with reference to the embodiments of the present invention. Also, it is obvious to those skilled in the art of the present invention that the objects and advantages of the present invention can be realized by the means as claimed and combinations thereof.

[27]

Technical Solution

[28] In accordance with an aspect of the present invention, there is provided a frequency band determining method for quantization noise shaping including checking whether audio signals obtained from low-pass filtering are transient, determining a predetermined frequency band to be applied as a frequency band to be applied for quantization noise shaping when the audio signals are not transient, and determining an extended frequency band extended more than the predetermined frequency band to be applied as a frequency band to be applied when the audio signals are transient.

Advantageous Effects

[29] This invention shapes quantization noise of audio signals using a long block according to frequency band for applying a Temporal Noise Shaping (TNS) algorithm and classifies the frequency band into a general frequency band and an extension frequency band according to whether the audio signals are transient. Thus, pre-echo and musical noise can be easily reduced.

[30] Thus, in this invention, sound quality is better than a quantization noise shaping method using the conventional TNS algorithm. The method of the present invention using the long block reduces the pre-echo more effectively than the typical method. Furthermore, it is possible to provide almost the same performance that can be obtained in the method using the short block by.

[31] Thus, in this invention, the short block is not overly used to thereby reduce the musical noise.

[32]

Brief Description of the Drawings

[33] Fig. 1 is a block view showing a Temporal Noise Shaping (TNS) processing apparatus in accordance with an embodiment of the present invention. [34] Figs. 2 and 3 show the pre-echo according to the transient index.

[35] Fig. 4 is a flowchart describing a method for shaping quantization noise in a low frequency band by using a long block in accordance with an embodiment of the present invention.

[36]

Best Mode for Carrying Out the Invention

[37] The advantages, features and aspects of the invention will become apparent from the following description of the embodiments with reference to the accompanying drawings, which is set forth hereinafter. Therefore, those skilled in the field of this art of the present invention can embody the technological concept and scope of the invention easily. In addition, if it is considered that detailed description on a related art may obscure a point of the present invention, the detailed description will not be provided herein. The preferred embodiments of the present invention will be described in detail hereinafter with reference to the attached drawings.

[38]

Fig. 1 is a block view showing a Temporal Noise Shaping (TNS) processing apparatus in accordance with an embodiment of the present invention.

[39] Referring to Fig. 1, a TNS processing apparatus 100 includes a TNS determiner 110 and a TNS coder 120.

[40] The TNS processing apparatus 100 properly reshapes the quantization noise in a time domain in a filter bank window to make the noise imperceptible. Hereinafter, the TNS processing apparatus 100 in a general HE-ACC coding apparatus is described.

[41] The TNS determiner 110 determines whether a TNS process is applied or not.

[42] To be specific, the TNS determiner 110 multiplies weight as shown in Eq. 1 to calculate Linear Predictive Coding (LPC) of pre-calculated Modified Discrete Cosine Transform (MDCT) spectrum.

[43]

[44] X_w (k) =X(k)-wfac(k) Eq. 1

[45]

[46] where

, k and n respectively represent MDCT coefficient unit and scale factor unit. [47] That is, the Eq. 1 normalizes into energy of a corresponding scale band. The MDCT spectrum range is applied to a predetermined range. Accordingly, the TNS determiner 110 determines a frequency range (band) for applying the LPC. [48] The TNS determiner 110 applies a smoothing filter to the normalized spectrum. This is for the LPC analysis. The smooth filtering indicates filtering down in a frequency band range from an LPC interruption frequency to an LPC operation frequency through the process as shown in Eq. 2.

[49]

wf ac (k) = 2

[51] Eq. 2

[52] where

, k and n respectively represent MDCT coefficient unit and scale factor unit. [53] On the other hand, the TNS determiner 110 performs filtering up in a frequency band range from the LPC operation frequency to cut-off frequency. Eq. 3 shows the filtering up process. [54]

wf ac (k) = 2

[56] Eq.3

[57]

[58] where

, k and n respectively represent MDCT coefficient unit and scale factor unit.

[59] The TNS determiner 110 calculates an auto-correlation function and the LPC using a

Levinson-Durbin algorithm. The TNS determiner 110 acquires a Partial Autocorrelation Coefficient (PARCOR) and calculates prediction gain based on the calculation result using the Levinson-Durbin algorithm.

[60] When the calculated prediction gain exceeds a threshold value, the TNS determiner

110 determines that the LPC should be performed on the spectrum and a TNS algorithm should be applied to the current window. [61] The TNS coder 120 carries out a quantization simulation in order of high to low

PARCOR coefficient to determine the TNS order as a first coefficient not smaller than the threshold value, e.g., 0.1. This is to use only effective TNS PARCOR coefficient.

[62] The TNS coder 120 passes through an LPC filter with the determined order and coefficient and applies the TNS algorithm to the MDCT spectrum coefficient to perform coding. The ACC coding is carried out using the applied MDCT spectrum coefficient.

[63] This invention extends and applies the TNS algorithm down to low frequency as low as 100 Hz. Herein, since the TNS algorithm is extensively applied, the pre-echo is decreased. However, the tone components of the frequency applying the TNS algorithm, i.e., the low frequency, may be distorted.

[64] Thus, this invention simultaneously uses the general TNS algorithm and the extension TNS algorithm. That is, this invention determines whether it applies the general TNS algorithm or the extension TNS algorithm, and then performs the TNS algorithm based on the determination result. A reference for the determination is extension range of the extended low frequency.

[65] Hereinafter, conditions for determining that the TNS algorithm is applied or the extension TNS is applied are described.

[66] First, the TNS determiner 110 determines if the general TNS algorithm can be applied or not. When the block switching result only in the low frequency band is transient, the TNS determiner 110 applies the extension TNS algorithm.

[67] Second, the TNS determiner 110 applies the extension TNS algorithm when the prediction gain in the frequency band extensively applying the TNS algorithm up to 100 Hz exceeds the threshold value and the transient signals with increased energy are located between fourth and seventh frames among eight frames. On the other hand, the TNS determiner 110 applies the general TNS algorithm when the transient signals with decreased energy are located in between zeroth to third frames among eight frames.

[68] The transient index of the transient signals 0 to 7 indicates the transient index determined by the block switching between the short block and long block. Each block indicates each point where the corresponding frame is divided into eight. This transient index is used for the effective coding for the HE-ACC and referred when eight short blocks are bound up into four groups for applying the short blocks. The reference value for the TNS algorithm is the transient index of the signals passed and filtered through the low frequency.

[69] In the above described second condition, the effect of the pre-echo is considered and the degree of effect is influenced to the extension TNS application. Herein, time for pre-echo is the same as the corresponding window length.

[70] That is, as the transient portion with increased energy is located in a front portion of the corresponding frame, the pre-echo occurs in a narrower range. On the other hand, as the transient portion with increased energy is located in an end portion of the corresponding frame, the pre-echo occurs in a wider range.

[71] Figs. 2 and 3 show the pre-echo according to the transient index.

[72] Referring to Fig. 2, a first transient index 101 indicates a transient signal in an end portion of the frame. Herein, since first pre-echo 102 is located in the end portion of the frame, the first pre-echo 102 occurs in larger range.

[73] Referring to Fig. 3, a second transient index 103 indicates a transient signal in a front portion of the frame. Herein, when the second transient index 103 is in the end portion of the frame, the second pre-echo 104 affects more than the first pre-echo 102 shown in Fig. 2. Thus, the TNS determiner 110 determines that the extension TNS algorithm is applied.

[74] Fig. 4 is a flowchart describing a method for shaping quantization noise in a low frequency band by using a long block in accordance with an embodiment of the present invention.

[75] A TNS determiner 110 calculates prediction gain of audio signals by using a long block in step S302. That is, the TNS determiner 110 calculates the auto-correlation function and LPC by using a Levinson-Durbin algorithm and acquires a PARCOR based on the calculation result, and calculates prediction gain.

[76] The TNS determiner 110 determines whether the calculated prediction gain exceeds the threshold value in step S304.

[77] The TNS determiner 110 separately performs low-pass filtering by using the low- pass filter to determine frequency components in an extension band. This is for using only the long block. An example of the low pass filter function is shown in Eq. 4.

[78]

^[79] 0.0000285 (z³ + 3z² + 3z + 1)

H ( z ) = _z ³ _ 2.9i78 z³ + 2.8433 z - 0.9253

Eq. 4

[80]

[81] where H(z) indicates the low-pass filter function. Various low-pass filer functions can be applied to the H(z). There is no big difference in a low-pass filtering ability of various low-pass filer functions. The TNS determiner 110 uses the low-pass filter to acquire the signal in a low frequency band under 1 kHz.

[82] Based on the comparison result in step S304, when the prediction gain exceeds the threshold value, the TNS determiner 110 checks the signals passed and filtered through the low frequency are transient in step S306. That is, the TNS determiner 110 determines the frequency band applied for shaping the quantization noise according to the check result in step S306. On the other hand, when the prediction gain does not exceed the threshold value, the TNS determiner 110 calculates the prediction gain of the extension band and checks the prediction gain of the extension band exceeds the threshold value in step S314. Herein, it is checked that the signals passed and filtered through the low frequency pass are transient by using the block switching algorithm in the AAC apparatus.

[83] Based on the check result in step S306, when the signals passed and filtered through the low frequency are transient, the TNS determiner 110 does not re-adjust a masking threshold value 308. On the other hand, when the signals passed and filtered through the low frequency are not transient, the TNS determiner 110 determines that the general TNS algorithm in applied in step 312.

[84] When the masking threshold value is not re-adjusted, in the general TNS algorithm, the making threshold value is lowered in the frequency band not applying the TNS algorithm to effectively use the bit. On the other hand, in this invention using the extension TNS algorithm, since the entire frequency bands apply the TNS algorithm, a re-adjustment of the masking threshold is not required.

[85] The TNS determiner 110 does not re-adjust the masking threshold value and extends the frequency band applying the TNS algorithm down to the frequency as low as 100 Hz in step S310. The TNS coder 120 re-calculates the coefficient based on the TNS algorithm extensively applied to the low frequency band, and then performs the TNS coding.

[86] Based on the check result in step S314, when the prediction gain of the extension band exceeds the threshold value, the TNS determiner 110 analyzes kind and index of transient signal passed and filtered through the low frequency and checks the effect of the pre-echo exceeds a reference value in step S316. That is, the TNS determiner 110 determines if the quantization noise is shaped or not based on the analysis result in step 316. For instance, the TNS determiner 110 determines that the effect of the pre-echo exceeds the reference value and applies the TNS algorithm when the kind and the index of the transient with increased energy is located in the end portion of the corresponding frame or when the kind and the index of the transient with decreased energy is located in the front portion of the corresponding frame.

[87] On the other hand, based on the check result in step S314, when the prediction gain of the extension band does not exceed the threshold value, the TNS determiner 110 does not apply the TNS in step S318.

[88] Based on the check result in step S316, when the effect of the pre-echo is big, the

TNS determiner 110 does not re-adjust the masking threshold value in step S308. On the other hand, when the effect of the pre-echo is small, the TNS determiner 110 does not apply the TNS algorithm in step S318. [89] As described above, the technology of the present invention can be realized as a program. A code and a code segment forming the program can be easily inferred from a computer programmer of the related field. Also, the realized program is stored in a computer-readable recording medium, i.e., information storing media, and is read and operated by the computer, thereby realizing the method of the present invention. The recording medium includes all types of recording media which can be read by the computer.

[90] The present application contains subject matter related to Korean Patent Application

Nos. 2007-0119413 and 2008-0048837, filed in the Korean Intellectual Property Office on November 21, 2007, and May 26, 2008, respectively, the entire contents of which are incorporated herein by reference.

[91] While the present invention has been described with respect to certain preferred embodiments, it will be apparent to those skilled in the art that various changes and modifications may be made without departing from the scope of the invention as defined in the following claims.

Claims

[1] A frequency band determining method for quantization noise shaping, comprising: checking whether audio signals obtained from low-pass filtering are transient; determining a predetermined frequency band to be applied as a frequency band to be applied for quantization noise shaping when the audio signals are not transient; and determining an extended frequency band extended more than the predetermined frequency band to be applied as a frequency band to be applied when the audio signals are transient. [2] The method of claim 1, wherein the predetermined frequency band to be applied is a certain TNS frequency band applied to a general TNS algorithm. [3] The method of claim 2, wherein the extended frequency band is extended down to a lower frequency band than a certain TNS frequency band. [4] A method for quantization noise shaping, the method comprising: comparing a prediction gain of audio signals calculated using a long block with a threshold value; determining a frequency to be applied for quantization noise shaping by checking whether the audio signals obtained from low-pass filtering are transient when the prediction gain exceeds the threshold value; and performing quantization noise shaping onto the audio signals according to the predetermined frequency band. [5] The method of claim 4, wherein said determining a frequency to be applied further includes: checking whether the audio signals obtained from low-pass filtering are transient; determining a predetermined frequency band as the frequency band to be applied for quantization noise shaping when the audio signals are not transient; and determining an extended frequency band extended more than the predetermined frequency band to be applied as a frequency band to be applied when the audio signals are transient. [6] The method of claim 5, further comprising: determining whether to perform the quantization noise shaping or not by checking whether a prediction gain of the extended frequency band exceeds a threshold value, when the prediction gain does not exceed the threshold value. [7] The method of claim 6, wherein whether to perform the quantization noise shaping is determined by checking transient kind and transient index of the audio signals when the prediction gain of the extended gain of the extended frequency band exceeds the threshold value. [8] The method of claim 7, wherein said determining the quantization noise shaping is performed when the transient kind and the transient index of the audio signals have increased energy and are located in an end portion of a corresponding frame. [9] The method of claim 7, wherein said determining the quantization noise shaping is performed when transient kind and transient index of the audio signals have decreased energy and are located in front portion of a corresponding frame. [10] The method of claim 7, wherein the quantization noise shaping does not re-adjust a masking threshold value according to the extended frequency band.