JP2007178684A - Multi-channel audio decoding device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that, in a conventional multi-channel audio decoding technology using spatial information, high arithmetic cost and high memory requirement are caused by processing an analysis and synthesis filter bank with a complex coefficient, and when a real number arithmetic filter bank is used, computational complexity is reduced but sound quality is deteriorated by aliasing. <P>SOLUTION: Sound quality deterioration by the effect of the aliasing in a specified sub-band, which arises when the complex filter bank of the conventional multi-channel decoding technology using the spatial information is changed to a real number type, is solved firstly by determining an area where the aliasing having great influence by using a reflective coefficient at first, and secondly by equalizing change of coefficient of scaling or mixing by using an equalizing tool. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention provides a decoding apparatus that can perform processing with low power consumption and a small memory capacity in a low bit rate multi-channel acoustic codec (for example, Non-Patent Document 1) using conventional spatial information. The present invention is not limited to the following, but can be applied to low-bit-rate applications such as broadcasting, home theater systems, in-vehicle audio systems, and electronic game systems.

  In recent years, a new multi-channel acoustic coding / decoding technique called a spatial audio codec (spatial acoustic codec) has been developed (Non-patent Document 1). This can compress and encode the presence of multi-channel with a very small amount of information. For example, the AAC system, which is a multi-channel codec that is already used as an audio system of Japanese digital television, is 5.1ch. While a bit rate of 512 kbps or 384 kbps is required, a spatial codec compresses and encodes a 5.1 channel multi-channel signal at a very low bit rate of 128 kbps, 64 kbps, and 48 kbps. Can do.

  FIG. 1 is a diagram for explaining the basic principle of conventional spatial acoustic coding and decoding, represented by Non-Patent Document 1, taking a stereo input signal (2ch signal) as an example. Here, L indicates a left channel signal and R indicates a right channel signal. In the encoding processing unit, the L and R signals, which are input sound signals, are processed in units of frames at predetermined time intervals, and in the downmix unit (100), for example, downmixing is performed according to an equation of M = (L + R) / 2. A signal M is generated. The spatial parameter detection module (102) calculates a plurality of spatial parameters for each spectrum band from the L, R, and M signals. The acoustic encoder (104) encodes the downmix signal M using an encoding method such as MP3 or AAC, and generates a compressed encoded sequence. Further, in the multiplexer MUX (106), the spatial parameter information and the encoded sequence of the M signal are multiplexed to generate a bit stream.

  Spatial parameter information detected by the spatial parameter detection module (102) includes an inter-channel level difference (hereinafter referred to as ILD) indicating a level / intensity difference between two signal channels and a similarity (coherence) between the two channels. Inter Channel Correlation (hereinafter referred to as ICC) indicating (correlation degree). In general, the ILD controls sound balance / localization, and the ICC controls sound width / diffusivity. These are both spatial parameters that help the listener compose an auditory scene in the head. These spatial parameters are calculated for each parameter band when the normal acoustic spectrum is divided into groups consisting of a plurality of “parameter bands”.

  In the decoding process, first, the demultiplexer DEMUX (108) separates the input bit stream into spatial parameter information and a coded sequence of the downmix signal M. The separated M encoded sequence is decoded by an acoustic decoder (110) (for example, an AAC decoder, an MP3 decoder, etc.), and the downmix signal M is restored. The stereo signal synthesis module (112) separates the decoded downmix signal M and the spatial parameters into two-channel signals, and restores the original stereo signals (L signal and R signal).

In the above example, one downmix signal and a spatial parameter are extracted from two input signals in the encoder, and the downmix signal is separated into two signals from the spatial parameter and the downmix signal in the decoder. However, an audio signal having more than 2 channels (for example, 6 signals constituting a 5.1 channel sound source) is compressed into a 1-channel or 2-channel downmix signal at the time of encoding processing, and 5.1 channels at the time of decoding processing. The signal can be restored to 6 channels. FIG. 2 shows an example in the case of 6 channels. In each channel separation module (200 to 204), the process of separating one intermediate downmix signal into two intermediate downmix signals is converted into a single signal for each of the six channels. Repeat until separated. Here, L _f , R _f , L _s , R _s , C, and LFE are a left front speaker signal, a right front speaker signal, a left rear speaker signal, a right rear speaker signal, a front center signal, and a low frequency, respectively. Corresponds to the signal.

FIG. 3 is a block diagram illustrating the principle of the channel separation module (300) in the case of 2-channel input. The input downmix signal M is processed by an all-pass filter (301) to generate an uncorrelated signal _Mrev . Next, in the module 303, the two signals M and _Mrev are combined with a mixing coefficient _Hij and separated into two signal outputs L and R by the following equation (Equation 1).

  The mixing coefficient used here is calculated from the spatial parameters ILD and ICC in the block (302) so as to maintain the degree of correlation between the separated signals and the directivity of the separated signals.

  FIG. 4 is a block diagram illustrating main modules of a conventional spatial acoustic decoder. The downmix signal encoded sequence separated by the demultiplexer is decoded into a time domain acoustic signal by the acoustic decoder (400). Next, it is converted into a plurality of subband signals by the analysis filter bank (401). This analysis filter bank is composed of, for example, a QMF filter bank and a Nyquist filter bank, and is divided into a plurality of subbands by the QMF filter bank, and then the spectrum resolution of the low frequency subband is reduced. In order to enhance it, the low frequency subband is further divided by a Nyquist filter bank.

Next, a process of generating a channel-separated y vector signal from the output vector x signal of the analysis filter bank (401) will be described by taking 5.1ch in FIG. 2 as an example. The purpose of the pre-matrix module (402) is to generate an intermediate signal that can be used by each channel separation module (200-204) of FIG. 2 to generate an uncorrelated signal. The pre-matrix module calculates a vector R _{1 of} scaling factors of the ILD spatial parameters of the synthesized signals M ₁ to M ₄ from the ILD spatial parameters that scale the energy level of the input downmix signal M.

In this example, M ₁ , M ₂ , M ₃ and M ₄ are respectively
_{M 1 = L f + R f} + C + LFE
M ₂ = L _f + R _f
M ₃ = C + LFE
M ₄ = L _s + R _s
It is.

The decorrelation module (403) performs an all-pass filter process on v (n, sb) and generates a decorrelation signal w by the following equation. Here, M _{i, rev} is obtained by subjecting M _i to decorrelation processing.

The post-matrix module (404) calculates a matrix R _{2 of} mixing coefficients that mixes M and M _{i, rev} to derive individual signals. Referring to the example of FIG.
L _f = H _{11, A} * M ₂ + H _{12, A} * M _{2, rev}
M ₂ = H _{11, D} * M ₁ + H _{12, D} * M _{1, rev}
M ₁ = H _{11, E} * M + H _{12, E} * M _rev
It becomes.

Here, H _{ij, A} is a mixing coefficient H _ij in the channel separation module CS_A (200) or the like. The above three formulas can be combined into one vector multiplication formula as shown in the following (Equation 4).

Equations similar to the above are R _f , L _s,. . . It is possible to derive the _LFE by calculating the R2 _{, Rf} , R2 _{, Ls} ... R2 _{, LFE} vector for deriving the _LFE . Therefore, the vector y can be expressed as (Equation 5) below.

R ₂ , which is a matrix consisting of multiple sets of mixing coefficients from the channel separation module (200-204), generates M, M _rev , M _{2, rev,.} . . It seems that M _{4, rev} is linearly combined.

Both R ₁ and R ₂ can be specified by r indicating a row, c indicating a column, n indicating a time, and sb indicating a subband.

  Finally, each separated signal is converted into a time domain signal by the synthesis filter bank (405) to obtain a multi-channel output signal. Here, when the analysis filter bank includes a QMF analysis filter bank and a Nyquist analysis filter bank, the synthesis filter bank (405) includes a synthesis QMF filter bank and a synthesis Nyquist filter bank.

An object of the present invention is to reduce the required memory capacity and power consumption while maintaining high sound quality in the spatial acoustic decoder having the above-described configuration.
J. et al. Herre, et al, “The Reference Model Architecture for MPEG Spatial Audio Coding”, 118th AES Convention, Barcelona.

  However, the spatial acoustic decoder described in the prior art is realized by a filter bank with complex coefficients, and decoding processing is executed in the complex domain, so that it requires a lot of calculation costs and memory capacity. By using a filter bank of real coefficients instead of complex coefficients, it is possible to greatly reduce the amount of calculation, but in this case, there is a problem that sound quality deterioration due to the effect of aliasing occurs as described below. is there.

  The complex coefficient analysis filter bank outputs a signal of each subband when the spectral region is divided into a plurality of subbands. In the prototype filter used in an actual filter bank, aliasing occurs because frequency response regions overlap between subbands. If the analysis filter bank output signal is not modified, or if the same amount of modification is made for all subbands, the aliasing elements that flow to the surrounding subbands in the synthesis filter bank will be eliminated. In addition, the filter bank is designed.

  However, if the signal is modified, the complex filter bank alleviates the aliasing problem by introducing redundancy through its “oversampling” property, but the real coefficients to reduce the computation and memory load. When the filter bank is used, the signal changes from “oversample” to “critical sample”. In other words, when the signal band is independently scaled by the mixing matrices R1 and R2, it is possible to clearly hear the deterioration of sound quality due to the influence of aliasing. In fact, the effect of aliasing is particularly noticeable around the subband region having a strong tonal component of the signal spectrum.

  In order to solve the above problems, a multi-channel audio decoding apparatus according to the present invention generates an analysis filter bank of real coefficients for generating a plurality of subbands from an input time signal sequence, and generates an uncorrelated signal corresponding to the subband signals. A non-correlated module having an all-pass filter with a real coefficient, a channel expansion module for converting the subband signal into a multi-channel subband signal, a reflection coefficient calculation module for calculating a reflection coefficient from the subband signal, and the reflection The sub-bands with strong tonal components that are highly likely to cause aliasing are identified using coefficients, and the output signal of the channel expansion module is equalized using the reflection coefficients to further suppress the aliasing. Equalizing module and real number coefficient synthesis Characterized in that it is composed of Irutabanku.

  The present invention provides a real number analysis filter bank for converting a time domain signal into a plurality of subband signals, a non-correlation module having a real type all-pass filter for generating a non-correlated signal corresponding to the subband signal, It consists of a channel expansion module that converts a subband signal into a multichannel subband signal, and a synthesis filter bank for real number conversion that converts the multichannel subband signal converted by the channel expansion module into a time domain signal. A multi-channel audio decoding device, further comprising: a reflection coefficient calculation module for calculating a reflection coefficient from each of the subband signals; and identifying a subband in which a strong tonal component exists using the reflection coefficient to perform aliasing. In order to suppress the influence, the reflection coefficient is used to Characterized by comprising an equalizing module for adjusting the output signal of the tunnel expansion module provides a multi-channel audio decoding apparatus.

  The present invention also provides an analysis filter bank for real arithmetic that converts a signal in the time domain into a plurality of subband signals, an uncorrelated module having a real type all-pass filter that generates an uncorrelated signal corresponding to the subband signals, and A channel expansion module for converting the subband signal into a multichannel subband signal, and a synthesis filter bank for real number operation for converting the multichannel subband signal converted by the channel expansion module into a time domain signal A multi-channel audio decoding device, wherein a reflection coefficient calculation module that calculates a reflection coefficient from each subband signal, a tonality calculation module that calculates a tonality from the subband signal, and the reflection coefficient Supports with strong tonal components Identify bands, in order to suppress the influence of aliasing, said characterized in that it comprises an equalizing module for adjusting the output signal of the channel expansion module, provides a multi-channel audio decoding apparatus using the tonality.

  In one embodiment of the present invention, the reflection coefficient takes a value close to +1 or −1 when a single frequency component exists between two consecutive subbands, and uses (Equation 6): The sign is a value determined by even and odd of the continuous subbands.

  In a further embodiment of the present invention, the sign of the reflection coefficient is obtained by referring to a table.

  In a further embodiment of the present invention, when equalizing the output signal of the channel expansion module, the average value of the reflection coefficients if the average absolute value of the reflection coefficients of adjacent subbands is equal to or greater than a first threshold value. Is output corresponding to each subband, and if the average of the absolute values of the reflection coefficients of adjacent subbands is smaller than the second threshold, the reflection coefficient is output corresponding to each subband and adjacent to each other. If the average of the absolute values of the reflection coefficients of the subbands is smaller than the first threshold and greater than or equal to the second threshold, the reflection coefficient and the average value of the reflection coefficients are set as outputs corresponding to the respective subbands. Characterize.

  In another embodiment of the present invention, when equalizing the output signal of the channel expansion module, if the average value of the tonalities of adjacent subbands is greater than or equal to the first threshold value, the average value of the tonality is set for each sub-band. When the average value of the tonalities of adjacent subbands is smaller than the second threshold value, the tonalities are output corresponding to the respective subbands, and the average value of the tonalities of adjacent subbands. Is smaller than the first threshold and greater than or equal to the second threshold, the tonality and the average value of the tonality are output corresponding to the respective subbands.

  In another embodiment of the present invention, the analysis filter bank includes a real coefficient QMF analysis filter bank that converts a time domain signal into a plurality of subband signals, and a real coefficient Nyquist analysis that extends the resolution of the subband signals. The filter bank is composed of a Nyquist synthesis filter bank with real coefficients and a QMF synthesis filter bank with real coefficients.

  In another embodiment of the present invention, the equalizing process is applied to only a part of the frequency band of the audio signal.

  In another embodiment of the present invention, subbands sharing a spatial parameter are grouped as one parameter band, and equalizing processing is performed for each parameter band.

  The present invention also includes a step of converting a time domain signal into a plurality of subband signals by an analysis filter operation of a real number coefficient, a step of generating an uncorrelated signal corresponding to the subband signal by a real type all-pass filter, Multi-channel audio comprising: converting a sub-band signal into a multi-channel sub-band signal; and converting the converted multi-channel sub-band signal into a time-domain signal by a real coefficient synthesis filter operation. In the decoding method, a step of calculating a reflection coefficient from each subband signal, and a subband in which a strong tonal component exists is specified using the calculated reflection coefficient, and the influence of aliasing is suppressed. For the converted multi-channel using the reflection coefficient Characterized by comprising a step of equalizing the subband output signals to provide a multi-channel audio decoding method.

  The present invention also includes a step of converting a time domain signal into a plurality of subband signals by an analysis filter operation of a real number coefficient, a step of generating an uncorrelated signal corresponding to the subband signal by a real type all-pass filter, Multi-channel audio comprising: converting a sub-band signal into a multi-channel sub-band signal; and converting the converted multi-channel sub-band signal into a time-domain signal by a real coefficient synthesis filter operation. A decoding method further comprising: calculating a reflection coefficient from each subband signal; calculating a tonality from the subband signal; and identifying a subband in which a strong tonal component exists using the reflection coefficient In order to suppress the effects of aliasing, Wherein the differential band output signal of the converted multi-channel and a step of equalizing was characterized by, providing a multi-channel audio decoding method using Nariti.

  The present invention also provides a program for causing a computer to execute the multi-channel audio decoding method described above.

  The present invention also provides an information recording medium on which the above program is recorded.

  According to the present invention, it is possible to greatly reduce the amount of calculation of the conventional spatial acoustic decoding technique without changing the structure of the bit stream, and the sound quality due to aliasing distortion, which has been a problem when using a real coefficient filter, is achieved. It is possible to realize a multi-channel audio decoding device that suppresses deterioration and achieves both low computation and high sound quality.

  In the following description, the acoustic space encoding technique shown in Non-Patent Document 1 is referred to in large numbers and directly, but the present invention is not limited to the specific technique.

  The present invention will be described using the following embodiments and drawings, but is not intended to be limited thereto.

(Embodiment 1)
FIG. 5 is a block diagram of a decoder for explaining the first embodiment of the present invention. The analysis filter bank (501) includes a QMF filter bank and a Nyquist filter bank for converting the downmix output signal decoded by the audio decoder (500) into a plurality of subband signals. Here, the QMF filter bank modulation coefficient M used in the conventional technique is a complex coefficient of the following (Equation 7). In the present invention, the real modulation coefficient M of the following (Equation 8) is used.

  The spatial acoustic decoder shown in Non-Patent Document 1 has two decoder modes: a high sound quality (HQ) mode and a low computational complexity (LC) mode. In the HQ mode, the seven subbands in the low band are divided into 16, 8, 8, 4, 4, 4, and 4 hybrid subbands by the Nyquist filter bank (Nyq). As shown in the fourth column of Table 1, the Nyquist filter used in the LC mode includes two types of filters, a type A filter using complex coefficients and a type B filter using real coefficients. Are subdivided into 8, 2, 2 hybrid subbands. The number of corresponding hybrid subbands is written in the fourth column, and the subscripts of the hybrid subbands are written in the fifth column.

  In the Type A Nyquist filter bank used in the prior art, the following complex modulation coefficient (Equation 9) is used. In the present invention, the following real modulation coefficient (Equation 10) is used.

  The Type B filter uses a real coefficient Nyquist filter bank of (Equation 11) below.

  The number of non-overlapping hybrid subbands is one half of the total number of subbands divided by the real type A Nyquist filter. The sixth column of Table 1 shows the number of hybrid subbands when using real coefficients, and the seventh column shows the case where hybrid subband subscripts and complex coefficients when using real coefficients are used. The subscript correspondence of the hybrid subband is written.

Next, a method for suppressing the influence of aliasing that occurs when the above-described real coefficient filter bank is used will be described.
In the reflection coefficient calculation module (506), a hybrid subband in which a high tonal component exists is specified by using the reflection coefficient ref (sb) calculated by the following equation (Equation 12).

  Here, the value of the reflection coefficient ref (sb) takes a value between −1 and 1.

  Table 2 shows codes sref (sb) of ref (sb) of two adjacent subbands when a tonal component is present between two adjacent hybrid subbands in the LC mode. For example, if a tonal component is present between the second and third hybrid subbands (when sb <7, sb = even in Table 2), the codes of ref (2) and ref (3) from Table 2 Are both positive (+). The closer to the region where the frequency responses of the two hybrid subbands overlap, the greater the value of | ref (sb) |, which is the absolute value of ref (sb), and the risk of aliasing Becomes higher. A similar relationship is shown in Table 3 for the HQ mode.

  In the equalizing process, when it is determined from the values of ref (sb) and ref (sb + 1) that the tonal component exists between sb and sb + 1, the tonal The stronger the component, the closer these two values are adjusted.

FIG. 7 is a flowchart for explaining the details of the equalizing process in the equalizing module (507). Here, v (sb) represents a scaling factor R ₁ (sb) and a mixing factor R ₂ (sb) in the hybrid subband sb. In the module (507), v (sb) = R ₁ (sb) at the time point r row c column n, and v (sb) = R ₂ (sb) at the time point r row c column n in the module (508). It becomes.

  First, the subband index sb is initialized to 0 by step (700). Step (701) checks if all subbands have been processed. If the processing of all the subbands is completed, the equalizing process ends.

  In step (702), ref0 and ref1 are calculated for each subband sb. These are values obtained by multiplying the reflection coefficients ref (sb) and ref (sb + 1) of the subbands sb and sb + 1 by their polarities sref (sb) and sref (sb + 1) described in Table 2 or Table 3, respectively. Here, the average value of ref0 and ref1 and the average value of v (sb) and v (sb + 1) are calculated and stored as ave_ref and ave_v, respectively.

  When an arbitrary real number is multiplied by the polarity (+1 or -1) of the real number, a positive value is obtained. Therefore, when a single tonal component exists between sb and sb + 1, ref0 and ref1 are positive. Value. In step (703), it is confirmed whether ref0 and ref1 are both positive values. If not, the subband sb is incremented in step (708), and the same processing is repeated for the next subband.

  In step (704), ave_ref for each subband is compared with a second threshold TH2 (where TH2> TH1), and if the value of ave_ref is greater, the average tonality of sb and sb + 1 is very high and the area It is judged that the influence of ging is large and the maximum equalizing process is necessary. In such a case, the average value ave_v of the two subbands is output for the spatial parameter in step (705).

  In step (706), ave_ref is compared with the first threshold value TH1 (where TH1 <TH2). If the value of Ave_ref is smaller, it means that the tonality of sb and sb + 1 is so low that the influence of aliasing can be ignored. Therefore, in this case, the equalizing process is not performed.

  For all subbands having ave_ref that is between the first threshold value TH1 and the second threshold value TH2, the spatial parameter of sb and sb + 1 is adjusted to a linearly interpolated value by the processing of step (707). Is done. In this interpolation process, v (sb) and v (sb + 1) are close to their original values if ave_ref is close to the first threshold TH1, and v (sb) and v are appropriate if ave_ref is close to the second threshold TH2. (Sb + 1) is close to the average value of 0.5 * (v (sb) + v (sb + 1)).

The processing from step (700) to step (708) is performed for all rows r, columns c, and times n of R ₁ and R ₂ .

  In order to reduce the amount of spatial parameters, we divide the (n, sb) plane into parameter bands that share spatial parameters in a certain range of sb and n regions, and the knowledge that consecutive subbands have the same spatial parameters By utilizing this, it is possible to speed up the above embodiment. Only the subbands whose spatial parameters have changed as a result of the switching of the parameter bands are subjected to equalizing processing. In this case, for example, the conditional expression in step (703) may be modified as follows.

  (Ref0> 0 && ref1> 0) && (PARAM_BAND (sb)! = PARAM_BAND (sb + 1))

  In the decorrelation module (503), the non-integer delay coefficient is removed, and processing is performed using a real lattice coefficient as shown in the following expression (Expression 14) instead of the complex lattice coefficient of (Expression 13).

  As described above, since the phenomenon of the echo density of the output signal when the non-integer delay coefficient is removed is small, it is possible to realize a significant reduction in the calculation amount and the memory capacity while maintaining high sound quality.

  Finally, in the synthesis filter bank (505) composed of the Nyquist synthesis filter bank and the QMF synthesis filter bank, the complex modulation of (Expression 15) used in the QMF synthesis filter bank of the synthesis filter bank (405) described in the conventional example. Instead of the coefficients, synthesis filter processing is performed using the following real number modulation coefficients of (Expression 16), and a time-domain decoded multi-channel signal is output.

(Embodiment 2)
FIG. 6 is a block diagram of a decoder for explaining the second embodiment of the present invention. The difference from FIG. 5 which is the block diagram of the first embodiment is that the reflection coefficient calculation module (506) is replaced with the reflection coefficient / tonality calculation module (606), and the operations of the equalizing modules (607, 608) are partially different. That is, the operation of the other modules is the same as that in FIG. The subband in which the tonal component is present is specified by the reflection coefficient described in the first embodiment, and another tonality measurement unit is used to evaluate the necessity level of equalization.

For example, the tonality T _g (sb) in the sb subband of the g-th frame is calculated as an average value of coherence weighted with energy as shown in the following equation (Equation 17).

  Here, (Equation 18) below shows the signal power in the two frames g and g-1, and (Equation 19) below shows the coherence between the frames.

  The following (Equation 20) takes a value between 0 and 1, with 0 indicating no tonality and 1 indicating a very high tonality. As shown in the following equation (Equation 21), the smaller value of the tonalities of the two target frames is set as the final tonality.

  FIG. 8 is a flowchart for explaining a method of equalizing processing using tonality. The difference from the equalizing process using the reflection coefficient described in the first embodiment (FIG. 7) is that the subband in which the tonal component exists is determined using the reflection coefficient in (803), and then the step (802). Ave_T that is the average tonality of subband sb and subband sb + 1 calculated in step 805 is used to perform threshold determination in step (805) and step (806), and equalization processing is performed using ave_T in step (807). It is. The contents of the processing in the other steps are the same as those in FIG.

(Embodiment 3)
As a third embodiment of the present invention, a case where the equalizing process of the first and second embodiments is applied to only a part of the frequency spectrum will be described with reference to FIG.

  FIG. 9 shows a state where the frequency spectrum is divided by subbands SB_START and SB_STOP. For example, the equalizing process described in the first and second embodiments is performed only in the region B in FIG. Specifically, in the 'A' region corresponding to the low frequency part on the frequency spectrum, complex processing is left in order to hardly generate aliasing, and in the 'B' region, the description will be given in the first and second embodiments. Perform equalization by the method. In the 'C' region, other conventional equalizing is performed. That is, this embodiment can coexist with other conventional equalizing methods.

  Actually, the equalizing process to the part of the frequency spectrum region is performed by, for example, initializing sb with a numerical value SB_START in steps (700) and (800) in the flowcharts of FIGS. ) And step (801) can be realized by replacing the termination condition with sb == SB_STOP-1.

  Note that these processes are not limited to the case of FIG. 9, and if there is a band of a frequency spectrum that is not likely to cause aliasing, it is configured not to perform equalization for the band. Is also possible.

(Embodiment 4)
As an embodiment 4 of the present invention, an equalizing processing method considering a case where two adjacent tonal components exist in one QMF subband will be described. In Table 2 and Table 3, this applies when sb is larger than 7 in the LC mode and when sb is larger than 23 in the HQ mode.

  If sb is an even number and a tonal component exists between sb and sb + 1, as is clear from Tables 2 and 3, both ref (sb) and ref (sb + 1) are negative values. Similarly, if sb is an even number and a tonal component exists between sb + 1 and sb, ref (sb + 1) and ref (sb + 2) are both positive values.

  However, when two tonal components exist simultaneously in one QMF subband, the sign of ref (sb + 1) depends on the energy of the corresponding tonal component, and the tonal component on the high frequency side has more energy than the tonal component on the low frequency side. If it is large, ref (sb + 1) becomes a positive value, and if the energy of the tonal component on the low frequency side is large, ref (sb + 1) becomes a negative value. Therefore, in this case, the equalizing amount cannot be determined using ref (sb + 1).

  Hereinafter, the equalizing method according to the fourth embodiment will be described with reference to the flowchart of FIG. First, in step (1000), sb is initialized with START_SB. Here, START_SB is 8 in the LC mode, and START_SB is 24 in the HQ mode.

  In step (1002), the reflection coefficients ref (sb), ref (sb + 1), and ref (sb + 2) of three consecutive subbands starting from sb are calculated. When sb is an even number, the sign is negative, that is, sign (sb) is set. When -1 is -1 and sb is an odd number, sign (sb) is set to 1, and values obtained by multiplying each reflection coefficient by sign (sb) are obtained and recorded as ref0, ref1, and ref2, respectively.

  In step (1003), it is checked whether the signs of ref0 and ref2 are different. If they are different, it is estimated that two tonal components exist in one QMF subband. In this case, since ref1 is affected by the high frequency tonal component, ref0 is used as ave_ref in step (1004). On the other hand, when the codes of ref0 and ref2 are equal, it is estimated that one tonal component exists in one QMF subband, and the processing from step (1005) is the same as that described in the first embodiment. Processing is performed.

  In the first, second, third, and fourth embodiments of the present invention, description will be made using an acoustic space decoder composed of a pre-matrix module and a post-matrix module as shown in FIGS. However, it is also possible to use an integrated matrixing module in which a prematrixing module and a postmatrixing module are integrated to generate R1. FIG. 11 is a block diagram of an acoustic space decoder that performs equalizing processing using reflection coefficients when FIG. 5 is configured by an integrated matrix module. Here, as the process of the equalizing module (1106) for equalizing R1, the equalizing method described in the first or fourth embodiment of the present invention can be applied. Similarly, FIG. 6 can also be configured by an integrated matrix module, and the equalizing process can be performed by the same method as that described in the second or fourth embodiment. Further, in any of these cases, when approximated to u (n.sb) = x (n.sb), the reflection coefficient needs to be calculated only once, and the amount of calculation can be further reduced.

  The multi-channel audio decoding device of the present invention is a decoding device that can perform processing with low power consumption and a small memory capacity, including application of low bit rate such as broadcasting, home theater system, in-vehicle audio system, and Applicable to electronic game systems.

It is a figure which shows the basic principle of a spatial acoustic coding decoding process. It is a figure which shows the multistage channel separation method which isolate | separates a downmix signal into each signal. It is a figure which shows the principle of channel separation. It is a block diagram which shows the conventional spatial acoustic decoding technique. It is a decoder block diagram of Embodiment 1 of the present invention. It is a decoder block diagram of Embodiment 2 of the present invention. It is a flowchart of the equalizing process of Embodiment 1 of this invention. It is a flowchart of the equalizing process of Embodiment 2 of this invention. It is a figure for demonstrating the equalizing process of Embodiment 3 of this invention. It is a flowchart of the equalizing process of Embodiment 4 of this invention. It is a block diagram of a decoder in the case of applying integrated matrix processing.

Explanation of symbols

500 audio decoder 501 analysis filter bank 503 decorrelation module 506 reflection coefficient calculation module 507 equalizing module

Claims (20)

A real number analysis filter bank for converting a signal in the time domain into a plurality of subband signals, a non-correlation module having a real type all-pass filter for generating a decorrelation signal corresponding to the subband signal, and the subband signal Multi-channel audio comprising a channel expansion module for converting to a multi-channel sub-band signal, and a synthesis filter bank for real number operation for converting the multi-channel sub-band signal converted by the channel expansion module to a signal in the time domain A decoding device,
Further, a reflection coefficient calculation module that calculates a reflection coefficient from each subband signal, and a subband in which a strong tonal component exists is specified using the reflection coefficient, and the reflection coefficient is set to suppress the influence of aliasing. A multi-channel audio decoding apparatus, comprising: an equalizing module for adjusting an output signal of the channel expansion module. A real number analysis filter bank for converting a signal in the time domain into a plurality of subband signals, a non-correlation module having a real type all-pass filter for generating a decorrelation signal corresponding to the subband signal, and the subband signal A channel expansion module that converts multi-channel subband signals;
A multi-channel audio decoding device comprising a real-valued synthesis filter bank for converting a multi-channel subband signal converted by the channel expansion module into a time-domain signal,
Further, a reflection coefficient calculation module that calculates a reflection coefficient from each of the subband signals, a tonality calculation module that calculates a tonality from the subband signal, and a subband in which a strong tonal component exists is specified using the reflection coefficient. A multi-channel audio decoding apparatus comprising: an equalizing module that adjusts an output signal of the channel expansion module using the tonality in order to suppress the influence of aliasing. The reflection coefficient takes a value close to +1 or −1 when a single frequency component exists between two consecutive subbands, and uses the equation (Equation 1), and the sign is the continuous subband. The multi-channel audio decoding device according to claim 1, wherein the multi-channel audio decoding device is a value determined by even-oddness of.

  4. The multi-channel audio decoding apparatus according to claim 3, wherein the sign of the reflection coefficient is obtained by referring to a table.   When equalizing the output signal of the channel expansion module, if the average of the absolute values of the reflection coefficients of adjacent subbands is equal to or greater than the first threshold, the average value of the reflection coefficients is output corresponding to each subband. When the average of the absolute values of the reflection coefficients of adjacent subbands is smaller than the second threshold, the reflection coefficient is output corresponding to each subband, and the absolute value of the reflection coefficient of the adjacent subbands The average value of the reflection coefficient and the average value of the reflection coefficient is output as an output corresponding to each subband when the average is smaller than the first threshold and greater than or equal to the second threshold. Multi-channel audio device.   When equalizing the output signal of the channel expansion module, if the average value of the tonalities of adjacent subbands is equal to or greater than the first threshold, the average value of the tonality is set as an output corresponding to each subband and adjacent to each other. If the average value of the tonalities of the subbands is smaller than the second threshold value, the tonality is output corresponding to each subband, and the average value of the tonality of adjacent subbands is smaller than the first threshold value and the second threshold value. The multi-channel audio decoding apparatus according to claim 2, wherein when the threshold value is equal to or greater than the threshold value, the tonality and an average value of the tonality are output corresponding to each subband.   The analysis filter bank includes a real coefficient QMF analysis filter bank that converts a time domain signal into a plurality of subband signals, and a real coefficient Nyquist analysis filter bank that extends the resolution of the subband signal. 7. The multi-channel audio decoding apparatus according to claim 1, comprising a real coefficient Nyquist synthesis filter bank and a real coefficient QMF synthesis filter bank.   The multi-channel audio decoding apparatus according to any one of claims 1 to 7, wherein the equalizing process is applied only to a part of a frequency band of the audio signal.   The multi-channel audio decoding device according to any one of claims 1 to 8, wherein subbands sharing a spatial parameter are grouped as one parameter band, and equalization processing is performed for each parameter band. . Converting a time-domain signal into a plurality of subband signals by an analysis filter operation of a real coefficient; generating an uncorrelated signal corresponding to the subband signal by a real-type all-pass filter; A multi-channel audio decoding method comprising: a step of converting into a channel sub-band signal; and a step of converting the converted multi-channel sub-band signal into a time-domain signal by a synthesis filter operation of a real coefficient. ,
Further, a step of calculating a reflection coefficient from each of the subband signals, a subband in which a strong tonal component exists is identified using the calculated reflection coefficient, and the reflection coefficient is suppressed in order to suppress the influence of aliasing. And a method of equalizing the converted multi-channel subband output signal using the multi-channel audio decoding method. Converting a time-domain signal into a plurality of subband signals by an analysis filter operation of a real coefficient; generating an uncorrelated signal corresponding to the subband signal by a real-type all-pass filter; A multi-channel audio decoding method comprising: a step of converting into a channel sub-band signal; and a step of converting the converted multi-channel sub-band signal into a time-domain signal by a synthesis filter operation of a real coefficient. ,
Further, a step of calculating a reflection coefficient from each subband signal, a step of calculating a tonality from the subband signal, a subband in which a strong tonal component is present is identified using the reflection coefficient, and the influence of aliasing And a step of equalizing the converted multi-channel differential band output signal using the tonality. The reflection coefficient takes a value close to +1 or −1 when a single frequency component exists between two consecutive subbands, and uses the equation (Equation 2), and the sign is the continuous subband. The multi-channel audio decoding method according to claim 10 or 11, wherein the multi-channel audio decoding method is a value determined by an even / odd number.

  13. The multi-channel audio decoding method according to claim 12, wherein the sign of the reflection coefficient is obtained by referring to a table.   When equalizing the converted multi-channel subband output signal, if the average of the absolute values of the reflection coefficients of adjacent subbands is greater than or equal to the first threshold value, the average value of the reflection coefficients is used for each subband. When the average of the absolute values of the reflection coefficients of adjacent subbands is smaller than the second threshold, the reflection coefficient is output corresponding to each subband, and the reflection of adjacent subbands is reflected. The average of the absolute value of the coefficient is smaller than the first threshold value and greater than or equal to the second threshold value, and the reflection coefficient and the average value of the reflection coefficient are output as corresponding to each subband. Item 11. The multichannel audio method according to Item 10.   When equalizing the converted multi-channel subband output signal, if the average value of the tonalities of adjacent subbands is equal to or greater than the first threshold, the average value of the tonality is output corresponding to each subband. If the average value of the tonality of adjacent subbands is smaller than the second threshold, the tonality is output corresponding to each subband, and the average value of the tonality of adjacent subbands is the first threshold. 12. The multi-channel audio decoding method according to claim 11, wherein the tonality and the average value of the tonality are output corresponding to respective subbands when smaller than a second threshold value.   The analysis filter operation is composed of a real coefficient QMF analysis filter operation that converts a signal in the time domain into a plurality of subband signals and a real coefficient Nyquist analysis filter operation that extends the resolution of the subband signal. 16. The multi-channel audio decoding method according to claim 10, comprising: a Nyquist synthesis filter operation for real coefficients and a QMF synthesis filter operation for real coefficients.   The multi-channel audio decoding method according to any one of claims 10 to 16, wherein the equalizing process is applied only to a part of the frequency band of the audio signal.   The multi-channel audio decoding method according to any one of claims 10 to 17, wherein subbands sharing a spatial parameter are grouped as one parameter band, and equalization processing is performed for each parameter band. .   A program for causing a computer to execute the multi-channel audio decoding method according to any one of claims 10 to 18.   An information recording medium on which the program according to claim 19 is recorded.

Images