JPH04185017A - Quantization error reducing device for audio signal - Google Patents

Abstract

PURPOSE:To reduce the noise on audition and to extend the dynamic range by setting the characteristic of a noise filter based on one of plural allowable noise levels. CONSTITUTION:The quantization error generated in a quantizer 11 is fed back to the input side of the quantizer 11 through a noise filter 13. In such a case, an input audio signal is divided into blocks in each prescribed period by a blocking circuit 15, and frequency analysis is performed by a frequency analyzing means 16 with respect to each block. The allowable noise level is set by a means 10 based on the energy of each frequency component in the block out of the analysis output. The allowable noise level of each frequency component of the block is set by a means 18 based on one energy of data preceding the block and data succeeding the block with respect to time out of the analysis output. The characteristic of the filter is set based on one output or the synthesized output of circuits 17 and 18.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an audio signal quantization error reduction device that reduces quantization errors that occur during quantization of audio signals.

[Summary of the invention]

The present invention is an audio signal quantization error reduction device in which the quantization error is fed back to the input side of a quantizer via a noise filter, in which an input audio signal of a block for each predetermined period is frequency-analyzed, and each a first allowable noise level based on the energy of the frequency component; and a second allowable noise level based on the energy of at least one of data temporally preceding and following the block. To provide an audio signal quantization error reduction device that can reduce auditory noise by setting characteristics.
It is something that

[Conventional technology]

Currently, digital audio devices that handle digital audio signals include, for example, so-called compact disc (CD) playback machines, so-called digital audio tape recorders (DAT), and the like. Various unified standards are defined for these digital audio devices, and for example, the bit length of digital audio signals handled by these devices is defined as 16 bits according to the unified standard. In addition, the digital audio signals in these digital audio devices are digital audio signals obtained by encoding analog audio signals (sound waveform signals) using simple quantization such as so-called PCM (pulse coding). I am using it.

[Problem to be solved by the invention]

Incidentally, in recent years, it has been desired that digital audio equipment as described above be able to produce reproduced sound that is audibly higher in quality than the reproduced sound that is actually obtained from the above-mentioned unified standard.

In order to obtain such audibly better reproduced sound, it is considered effective to reduce the noise components contained in the digital audio signals handled by these digital audio devices, for example. in this way,
The reproduced sound obtained from the digital audio signal with reduced noise components has less noise and is more audible. As the noise component reduction process of this digital audio signal, for example, there is a quantization error reduction process using so-called error feedback, which is performed during quantization of the audio signal. That is, the quantization error reduction process using error feedback includes, for example, quantization errors (quantization noise) generated by the quantizer during quantization of audio signals.

A so-called noise shaving process is known in which quantization errors are reduced by error feedback in which quantization distortion is fed back to the input side of the quantizer via a noise filter.

Here, the quantization noise caused by quantization such as the PCM encoding described above develops a flat frequency characteristic in the entire frequency band of the audio signal, and the above noise shaving process quantizes this entire frequency band. From the noise, only the noise in the audio band (audible band) is reduced. However, despite the fact that human ears have different audible sensitivities due to so-called equal loudness curves or masking effects, the conventional noise shaving processing described above does not take into account the characteristics of the human ear (auditory characteristics). has not been done. For this reason, it is difficult to say that the above-described conventional quantization error reduction process using error feedback is necessarily effective in terms of human hearing.

Therefore, the present invention has been proposed in view of the above-mentioned actual situation, and is an audio system that can effectively reduce quantization errors (quantization noise) in terms of auditory perception, taking into consideration the characteristics of the human ear. It is an object of the present invention to provide a signal quantization error reduction device.

[Means to solve the problem]

The audio signal quantization error reduction device of the present invention has been proposed to achieve the above-mentioned object, and the quantization error generated in the quantizer is passed through a noise filter to the input side of the quantizer. In the audio signal quantization error reduction device, the audio signal is fed back into blocks, the blocking means for blocking the input audio signal at predetermined intervals, and the frequency analysis means for performing frequency analysis for each block from the blocking means. , based on the energy of each frequency component in the block among the outputs of the frequency analysis means.
a first noise level setting means for setting an allowable noise level of the block; and an output of the frequency analysis means based on the energy of at least one of data preceding the block in time and data following the block in time. , a second noise level setting means for setting a second allowable noise level of each frequency component of the block, and the filter characteristics of the noise filter are set by at least one of the first and second noise level setting means. The settings are made based on one output.

[Effect]

According to the invention, a first permissible noise level based on the energy of each frequency component or corresponding to the amount of a so-called simultaneous masking effect in human auditory characteristics is based on data temporally preceding and following the block. The second allowable noise level corresponds to the amount of the so-called temporal masking effect, and since the filter characteristics of the noise filter are determined based on both of these, quantization errors (quantization noise) can be effectively reduced perceptually. Can be reduced.

[Embodiments] Hereinafter, embodiments to which the present invention is applied will be described with reference to the drawings.

FIG. 1 shows a block diagram of a schematic configuration of an audio signal quantization error reduction device according to this embodiment.

The device of this embodiment shown in FIG. 1 is configured to feed back the quantization error generated in the quantizer 11 to the input side of the quantizer 11 via the noise filter 13, and the input audio signal is a blocking circuit 15 that blocks each block for each predetermined period (for example, 1st to 10th drive); a frequency analysis circuit 16 that performs frequency analysis for each block from the blocking circuit 15; and an output of the frequency analysis circuit 16. Of the outputs of the first noise level setting circuit 17 that sets a first allowable noise level based on the energy of each frequency component in the block, and the frequency analysis circuit 16, a second noise level setting circuit 18 that sets a second allowable noise level for each frequency component of the block based on the energy of at least one of the preceding data and the temporally subsequent data; The filter characteristics of the noise filter 13 are
and the second noise level setting circuit 17, 18, or their combined output.

That is, the device of this embodiment is an adder that obtains the quantization error generated during quantization in the quantizer 11 by subtracting the input to the quantizer 11 from the output of the quantizer 1. 12, the noise filter 13 that filters and outputs the output of the adder 12, and the adder 10 that adds the output of the noise filter 13 to the input side of the quantizer 11, forming a so-called error feedback circuit. It consists of Here, the filter characteristics of the noise filter 13 are specifically determined by filter coefficients based on information on the allowable noise level (or allowable noise spectrum), which will be described later, from the first and second noise level setting circuits 17 and 18. It is determined by the filter coefficient calculated by the calculation circuit 14. Therefore, in this error feedback circuit, quantization error reduction processing (so-called noise shaving processing) is performed based on the above-mentioned allowable noise spectrum. The signal after this noise shaving process is output from the output terminal 2.

By the way, when performing quantization error reduction processing (noise shaving processing) on the audio signal using the error feedback circuit, the perceptual dynamic range can be increased by performing processing that takes into account so-called masking of the input signal spectrum. I can do it. Noise shaving that takes this masking into consideration is, for example, noise shaving according to the spectrum of an input audio signal whose signal spectrum pattern is fixed to some extent, that is, noise shaving that takes into account the so-called masking that will be described later of the input audio signal spectrum. One can mention noise shaving using an allowed noise spectrum.

Alternatively, when the spectrum of an input audio signal changes, there is noise shaving using an adaptive noise spectrum that is obtained by taking masking of the spectrum into consideration. The above masking refers to a phenomenon in which certain signals mask other signals and become inaudible due to the characteristics of human hearing.This masking effect includes simultaneous masking effect and temporal masking effect. . Due to this masking effect, even if there is noise in the masked area, this noise becomes inaudible. The simultaneous masking effect mentioned above is a small sound that occurs at the same time as a certain loud sound (
or noise) is masked by the loud sound and becomes inaudible. Also, signal components in one frequency band mask signal components in other bands,
It also includes an effect (masking effect on the frequency axis) in which the sound of signal components in other bands becomes inaudible. The above-mentioned temporal masking effect is an effect in which small sounds (noise) temporally before and after a loud sound are masked by this loud sound and become inaudible, and masking temporally behind the loud sound is called forward masking. It is called,
Masking forward in time is called packed word masking. Furthermore, in temporal masking, due to the characteristics of human hearing, the effect of forward masking is effective for a long time (for example, about 1001 sec), whereas the effect of packed word masking is effective for a short time (for example, about 5 m5 sec). It becomes. Further, the level of the masking effect (masking amount) is approximately 20 dB for forward masking and approximately 30 dB for packed word masking.

In addition, the input signal is divided into so-called critical bands using human hearing characteristics, and noise shaving is performed for each band using an allowable noise spectrum that takes masking into account as described above. For example, more audibly effective noise shaving can be performed.

This makes it possible to increase the perceptual dynamic range of reproduced sound.

For this reason, the frequency analysis circuit 16 divides the input audio signal for each block from the blocking circuit 15 into so-called critical bands using human auditory characteristics. Frequency analysis is performed for each critical band. At this time, the division and analysis at the above-mentioned critical bandwidth is performed by, for example, first converting the input audio signal into components (FFT coefficients) on the frequency axis using, for example, Fast Fourier Transform (FFT), and then! The amplitude product Am (hereinafter m-0 to 1024) of the fFFT coefficients is defined as a critical bandwidth in which the higher the frequency band, the wider the band width is, taking into account the human auditory characteristics. Indicates the band number, and groups (bands) into groups (n=0 to 24). Next, from the signals for each of these critical bands, for example, the sum of the amplitude products Am for each band (peak or average or energy sum of the amplitude products Am) can be obtained using equation (1). The so-called park spectrum (summation spectrum) SB is determined.

S B = 1010g+s Cn (Pn)' [
dB] (1) However, n = 0 to 24, and Cn
is the number of elements in the n-th band, that is, the amplitude product (number of points), and Pn is the peak value of each band. The park spectrum SB of each of the above bands is shown in FIG. 2, for example. However, in the example of FIG. 2, in order to simplify the illustration, the total number of bands in the critical band is expressed by, for example, 12 bands (B1 to B12). The frequency analysis circuit 16 performs division by the critical bandwidth as described above and frequency analysis for each band, and outputs the output information from the above-mentioned first. The signal is sent to the second noise level setting circuit 17, 18.

These first. In the second noise level setting circuit 17 and 18, the simultaneous masking (J1
(including masking on the wavenumber axis), and the allowable noise level is set in consideration of temporal masking.

These first. Information on the allowable noise level (allowable noise spectrum) from the second noise level setting circuit 17-18 is sent to the filter coefficient calculation circuit 14, and the filter coefficient calculation circuit 14 calculates the filter coefficient according to the allowable noise spectrum. The signal is output and sent to the noise filter 13 mentioned above.

Here, the blocking circuit 15 and the frequency analysis circuit 16
, first and second noise level setting circuits ・17.18
Specifically, the filter coefficient calculation circuit 14 can be realized, for example, by specific example circuits as shown in FIGS. 3 and 6 below.

In the specific example circuit shown in Fig. 3, fast Fourier transform (
FFT) circuit 111. Amplitude phase information generation circuit 112. Band division circuit 113. The sum detection circuit 114 operates as the blocking circuit 15 and frequency analysis circuit 16 in FIG. Each component from the filter circuit 76 to the synthesis circuit 80 operates as the first noise level setting circuit 17 in FIG. Block delay circuits 115, 116 and coefficient multipliers 117, 118 . The synthesis circuit 119
It operates as the second noise level setting circuit 18 in the figure. Synthesis circuit 120. The subtracter 121° ROM 122 is
It operates as the filter coefficient calculation circuit 14 shown in FIG. 1 above. The output of this ROM 122 is sent to the noise filter 13 in FIG. 1 via the output terminal 4.

That is, in this specific example circuit, each circuit serving as the first noise level setting circuit 17 in FIG. ), and also takes into account temporal masking with signals that temporally precede and follow the band of interest by each circuit as the second noise level setting circuit 18 in FIG. A second allowable noise level is set.

That is, in this specific example circuit of FIG. 3, an audio signal is supplied to the input terminal 1, and this audio signal on the time axis is transmitted to the fast Fourier transform (FFT) circuit 111. The fast Fourier transform circuit 11
1, the audio signal on the time axis is converted into a signal on the frequency axis every predetermined period (that is, blocks in the blocking circuit 15 in FIG. 1), and the real component value Re is converted into a signal on the frequency axis.
An FFT coefficient consisting of and an imaginary component value of 1m is obtained.

These FFT coefficients are transmitted to the amplitude and phase information generation circuit 112, and the amplitude and phase information generation circuit 112 obtains an amplitude value Am and a phase value from the real component value Re and the imaginary component value 1m. information is sent to the band division circuit 113.

That is, in general, human hearing is sensitive to amplitude (power) in the frequency domain, but is quite insensitive to phase. Therefore, only the amplitude value Am is extracted from the output of the amplitude phase information generation circuit 112 and I try to use it.

The above band division circuit 113 divides the audio signal of the above amplitude (IAm) into the above critical bandwidth (critical band).The above amplitude value Am for each band divided into critical bands by this band division circuit 113 is , are respectively transmitted to the summation detection circuit 114.The summation detection circuit 1
14, the energy for each band (spectral intensity in each band) is determined by the respective amplitude value Am in each band.
(the peak or average of the amplitude value Am or the total energy). The sum detection circuit 11
4 or the park spectrum SB shown in Figure 2.
It is.

Here, in order to consider the influence of the park spectrum SB on masking on the frequency axis, the park spectrum SB is convolved with a predetermined weighting function (convolution). Therefore, the output of the summation detection circuit 114, that is, each value of the park spectrum SB, is sent to the filter circuit 76 via a block delay circuit 115 that delays the output of the summation detection circuit 114 in units of blocks.

The filter circuit 76 includes, for example, a plurality of delay (z-') elements that sequentially delay input data, and a plurality of multipliers that multiply the output from each delay element by a filter coefficient (weighting function), respectively. It consists of a summation adder that takes the sum of the outputs of these multipliers. The filter circuit 76 performs convolution processing on the park spectrum SB.

Through the narrowing down process, the sum of the parts shown by the dotted line in FIG. 2 is calculated.

By the way, at the level α when calculating the masking spectrum on the frequency axis (that is, the allowable noise spectrum) of the park spectrum SB, if this level α is small, the masking spectrum (masking curve) for the signal on the frequency axis will be It will go down,
Conversely, if the level α is large, the masking spectrum will rise. Note that the level α is a level at which the masking spectrum for each critical band is obtained by performing inverse convolution processing as described later. In general, in audio signals, etc., the spectral intensity (energy) of the high frequency region
is small. Therefore, in this specific example, taking these things into consideration, the level α is increased as the energy goes to a higher frequency range. For this reason, in the circuit for the first noise level setting circuit 17 in this specific example circuit, the level α for the energy in the high frequency interval is set high.

That is, in this specific example circuit, a level α corresponding to the above-mentioned masking spectrum is calculated, and the level α is controlled so that it becomes higher as the frequency range increases. Therefore, the output of the filter circuit 76 is sent to a subtracter 78. The subtracter 78 calculates the level α in the convolved area.

Here, the subtracter 78 is supplied with a tolerance function (a function expressing the masking level) for determining the level α. The above level α can be adjusted by increasing or decreasing the tolerance function.
is under control. The allowable function is generated by the function generation circuit 77
Supplied from.

The level α corresponding to the above-mentioned masking spectrum can be determined by equation (2), where i is the number given sequentially from the low band of the critical band.

α=5-(n-ai)...(2) This (2nd
), n and a are constants, a>0, S is the intensity of the park spectrum after convolution processing, and (n-ai) in equation (2) is the tolerance function. In this specific example, n==8
8. It is assumed that a=1.

In this way, the level α is determined, and this data is transmitted to the divider 79. The divider 79 is used to perform inverse convolution calculation on the level α in the area subjected to the convolution process. Therefore, by performing this inverse convolution process, a masking spectrum can be obtained from the level α. In other words, this masking spectrum is the first allowable noise level (allowable noise spectrum) determined for each band.
becomes. Although the above deconvolution processing requires complicated calculations, in this specific example, a simplified divider 79 is used to perform the deconvolution.

Next, the masking spectrum is supplied to a subtracter 121 via a synthesis circuit 80 and a synthesis circuit 120. Here, the subtracter 121 includes the sum detection circuit 114.
, that is, the park spectrum SB from the aforementioned summation detection circuit 114 is supplied via a delay circuit 127. Therefore, by performing a subtraction operation between the masking spectrum and the park spectrum SB in the subtracter 121, as shown in FIG. It will be masked.

The output of the subtracter 121 is sent to the ROM 122. The ROM 122 stores a plurality of filter coefficients, and outputs a filter coefficient according to the output of the subtracter 121. The output of this ROM 122 is sent to the noise filter 13 in FIG. 1 via the output terminal 4. Note that the delay circuit 127 is connected to the sum detection circuit 114 in consideration of the amount of delay in each circuit before the synthesis circuit 120.
The park spectrum SB is provided to delay the park spectrum SB.

Further, during the synthesis in the synthesis circuit 80, data indicating the so-called minimum audible curve (equal loudness curve) RC, which is the human auditory characteristic as shown in FIG. 5, is supplied from the minimum audible curve generation circuit 129. and the masking spectrum MS can be synthesized. In this case, by combining the minimum audible curve RC and the masking spectrum MS, the first allowable noise level can be increased up to the shaded portion in this figure. Note that this FIG. 5 shows the critical band shown in FIG. 2 described above, and also shows the signal spectrum SS at the same time.

As described above, in each circuit corresponding to the 10th noise level setting circuit 17 of this specific example, the first allowable noise level is determined by taking into account the masking on the frequency axis of the signal for each band. can. Therefore, filter coefficients can be set based on the first allowable noise level.

In addition, in this specific example circuit, in addition to determining the filter coefficients in consideration of the above-mentioned simultaneous masking (masking on the frequency axis), the filter coefficients are determined based on the energy of temporally preceding and following signals of the band of interest (critical band). By setting the second allowable noise level of the band of interest, filter coefficients are simultaneously determined in consideration of temporal masking on the time axis. That is, in each circuit corresponding to the second noise level setting circuit 18 of this specific example, as described above, the current signal of an arbitrary band (band of interest) for which the first allowable noise level is set is set. On the other hand, the allowable noise level (second allowable noise level) of the current band of interest is set in consideration of temporal masking caused by signals preceding and following the current signal of the band of interest on the time axis. That's what I do.

For this reason, the synthesis circuit 120 is supplied with the output of the synthesis circuit 119, which is one of the circuits corresponding to the second noise level setting circuit 18, as well as the output of the synthesis circuit 80.

That is, the synthesis circuit 119 is supplied with a calculated temporal masking level based on the energy of the signals temporally before and after the current signal of the band of interest as the second allowable noise level signal. The allowable noise levels of these temporally preceding and following signals are synthesized by the synthesis circuit 119 and output.

In order to obtain such a second allowable noise level, this specific example circuit includes a block delay circuit 115 that delays the output of the summation detection circuit 114 in units of blocks (the number of blocks to be delayed may be plural); 116, coefficient multipliers 117, 118.degree. and the above-mentioned synthesis circuit 119. That is, if the time point at which the output of the block delay circuit 115 is obtained is the current time point T0, then the time point at which the output of the sum detection circuit 114 is supplied to the block delay circuit 115 is temporally after the current time point T0 (current time T0). The time point T++ is after the current time T0 (which is a future time for the current time T0), and the time point output from the block delay circuit 116 is before the current time T0 (which is a past time with respect to the current time T0). The time point is T-1.

The signal at the later time point T + +, that is, the sum detection circuit 11
The output of 4 is provided to a coefficient multiplier 118.

The coefficient multiplier 118 considers temporal masking (packed word masking) for the signal of the band of interest at the current time T0 by the signal of the band of interest at time T+I after the signal is supplied to the coefficient multiplier 118. The signal at the later time Tel is multiplied by the determined multiplication coefficient. That is, this multiplication coefficient is a coefficient that is set in consideration of the influence of the combining circuits 119 and 120. For example, if the signal at the later time point T, 1 is normalized to l, then the signal at the later time point T, I The signal at the subsequent time point Tel is multiplied by a multiplication coefficient corresponding to the level at which packed word masking by the signal acts on the signal at the current time T0. Further, the signal at the previous time point T, ie, the output of the block delay circuit 116, is supplied to the coefficient multiplier 117.

The coefficient multiplier 117 considers temporal masking (forward masking) for the signal of the band of interest at the current time T by the signal of the band of interest at the previous time T-, which is supplied with the signal to the coefficient multiplier 117. The signal at the previous time point T is multiplied by the multiplication coefficient determined as follows. That is, the multiplication coefficient is sent to the synthesis circuit 119 as described above.
．． For example, the previous time point T
-, when the signal at the previous time T, is normalized, the multiplication coefficient corresponding to the level at which forward masking by the signal at the previous time T, acts on the signal at the current time T0, becomes the signal at the previous time T-+. Multiplied. Each of these coefficient multipliers 11
The outputs obtained by combining the outputs of 7 and 118 in the synthesis circuit 19 become the second allowable noise level.

Note that the synthesis in the synthesis circuit 119 is performed by, for example, adding the outputs of the respective coefficient multipliers 117 and 118. The output of the synthesis circuit 119 obtained in this way is sent to the synthesis circuit 120.

In the synthesis circuit 120, for example, the synthesis circuit 119
A combination process is performed in which at least one of the output of the combination circuit 119 and the output of the combination circuit 80 is selected, for example, the larger one, or a combination process is performed in which the outputs of each of the combination circuits 119 and 119 are added with predetermined weighting. . Further, when performing this synthesis process by addition, for example, the second allowable noise level may be determined in consideration of the energy of the entire band and then added.

In addition, in this specific example, it is also possible to adopt a configuration in which the above-described minimum audible curve synthesis process is not performed. In this case, the configuration shown in FIG. 3 eliminates the need for the minimum audible curve generation circuit 129 and the synthesis circuit 80, and therefore the output from the subtracter 78 is deconvoluted by the divider 79 and then immediately It will be transmitted to the synthesis circuit 120.

As described above, in this specific example circuit, for the signal of the band of interest, each circuit corresponding to the first noise level setting circuit 17 in FIG. In addition to setting the permissible noise level, each circuit corresponding to the second noise level setting circuit 18 of the ttg sets a second permissible noise level that takes into account temporal masking in signals temporally preceding and following the band of interest. I'm trying to set it up. Subtractor based on these 1st and 2nd allowable noise levels! 21 is performed, and a filter coefficient is output from the ROM 122 based on the output of the subtracter 121, so that noise shaving that is good for human hearing becomes possible.

Another specific example circuit shown in FIG. 6 includes a filter bank 90 that frequency-analyzes the input audio signal supplied to the input terminal 1 in a non-block manner, and each frequency component analyzed by the filter bank 90 is For example, 10a+.

FFT (Fast Fourier Transform) circuits 20, 40 for further frequency analysis in blocks of 5ms+2.5ms, respectively.
60, further comprising a masking spectrum calculation circuit 75 for setting the first allowable noise level, and a masking spectrum calculation circuit 75 for setting the first allowable noise level;
period delay circuit 23.43 to set the allowable noise level of
．． 63. Synthesizing circuit 50°70.5ms delay (DL) circuit 51. Z5IIIs delay circuit 71. 72. 73. Selection circuit 52, combination selection circuit 74, weighted combination circuit 24.4
4.64. Furthermore, the filter bank 90 serves as at least a band dividing means for dividing the input audio signal into a plurality of frequency bands (in this specific example, into three)! one filter (e.g. the so-called QM
It has a mirror filter such as F), and performs division so that the high frequency band width becomes large.

Further, the FF7 circuits 20, 40, and 60 perform processing to reduce the block time length in which the fast Fourier transform processing is performed in the high frequency band. The output of this specific example circuit is sent to the noise filter 13 in FIG. 1 via the output terminal 4.

That is, in FIG. 6, the input terminal I is supplied with an input audio signal of DC to 22 kHz obtained by sampling at a sampling frequency fs = 44.1 kHz, for example, and the input audio signal is passed through the filter. It is supplied to bank 9o. The filter bank 90 is, for example, a QMF91.9 as shown in FIG.
2 are connected in cascade in two stages, and the input audio signal of DC to 22kHz supplied to the input terminal 1 is Q
It is supplied to MF91. The QMF91 divides the input audio signal into two at 11kHz, so the QMF91 outputs DC~1lk-211kHz~2
An output signal with a band of 2 kHz is obtained. The output signal in the band of 11 kHz to 22 kHz is sent to the FF7 circuit 60 via the terminal 93. Above DC~11
The output signal of & is sent to QMF92. The QMF9
2 divides the input signal into two at 5.5kHz, so from the QMF92, DC to 5.5kHz
, 5.5, an output signal in the band of 7 to 11 kHz is obtained. 5.5kl (z ~ 1lkl (z signal is sent to the FF7 circuit 40 through the terminal 94, DC ~ 5.5kl)
A Hz signal is sent to the FF7 circuit 20 via the terminal 95. In this manner, the filter bank 90 frequency-divides the input audio signal into three bands in a non-blocking manner, and performs the division such that the high frequency band has a large bandwidth. In the example of FIG. 7, the filter is a QMF, but a configuration using a BPF (band pass filter) is also possible.

The FFT circuit 20 to which signals in the band from DC to 5.5 kHz are supplied from the filter bank 90 divides the supplied signals into blocks every 10 drives and performs FFT on each block.
Perform processing. Further, the FFT circuit 40 to which signals in the band of 5.5kHz to 11k[z are supplied performs FFT processing in blocks every 5ms, and the FFT circuit 60 to which signals in the band from 11kHz to 22kHz are supplied performs FFT processing in blocks of every 15m5. Perform FFT processing in blocks. That is, these FF
In the T circuit 20°40.60, processing is performed to reduce the block length to which the high frequency band is subjected to FFT. In this way, in this specific example, when forming blocks in each of the FF7 circuits 20, 40, 60, the time resolution in the high frequency range is increased by performing processing to reduce the time block length in the high frequency range. In the upper and lower frequencies, the time resolution is lowered and the number of samples within the l block is increased to increase the frequency resolution. In other words, since a normal audio signal has a short stationary section in the high frequency range, it is effective to increase the time resolution in the high frequency range as described above.
Furthermore, since the frequency resolution of human hearing is generally high in the low range, it is also effective to increase the frequency resolution in the low range as described above.

Here, FIG. 8 shows the filter bank 90 and each FF 7.
3 shows time blocks of processing by circuits 20, 40, and 60; In other words, Fig. 8 shows each processing unit (block) of the above-mentioned band division, FFT, etc., and the block is specified.

p indicates the passage of time, q indicates the band, and r indicates the time block. FIG. 8 shows that in the low frequency range from DC to 5.5 kHz, one time block in each band has a time length (time resolution) of 0+ns. Also 5,5
It shows that in the middle range of kHz to 11 kHz, the length of one time block is 5 ms, and in the high range of II kHz to 22 kHz, the length of one time block is L5 ms.

The FFT coefficient data for each band obtained by the FFT processing in the FF7 circuit 20, 40, and 60 is shown in FIG. Each circuit for the second noise level setting circuit 17 and 18 controls the first noise level setting circuit which takes simultaneous masking (masking effect on the frequency axis) and temporal masking effect into consideration. A second acceptable noise level is set.

This first. The second allowable noise level is determined, for example, as follows.

The output data of the FF7 circuit 20, 40, 60 as a whole is further divided into the critical bands. To perform this band division, the output data (frequency band DC to 5.5 kHz) of the FFT circuit 20 is further divided into, for example, 20 bands on the lower frequency side of the critical band by the critical band division circuit 21. . In addition, the output data of the FFT circuit 40 (5, 5 to 1lkl
) is further divided into, for example, three bands in the middle range of the critical band by the critical band dividing circuit 4, and F
The output data (from 11 kHz to 22 kHz) of the FT circuit 60 is further divided by a critical band dividing circuit 61 into, for example, two high-frequency bands of the critical band.

The outputs of the critical band division circuits 21, 41, and 61 are respectively connected to the energy detection circuits 22. 42. Sent to 62. Each energy detection circuit 22, 42. .

62, the energy of data (spectral intensity in each band) for each time block and for each critical band in each of the FF7 circuits 20, 40.60, for example,
It is obtained by taking the sum of the respective amplitude values in each band (the peak or average of the amplitude values, or the sum of energy). The output of the energy detection circuit 22.42.62 is
This is the spectrum of the sum of each critical band (park spectrum SB, see above-mentioned FIG. 2).

Here, in order to consider the influence of the park spectrum SB on masking on the frequency axis, the park spectrum SB is convolved with a predetermined weighting function (convolution). Therefore, the energy detection circuit 22.42
．． 46, that is, each value of the park spectrum SB, is sent to the masking spectrum calculation circuit 75 as the first noise level setting circuit 17 shown in FIG. The masking spectrum calculation circuit 75 includes the third masking spectrum calculation circuit 75 described above.
The filter circuit 76 and subtracter 78 shown in the figure. Function generation circuit 77
, and the divider 79, detailed explanation will be omitted. However, the signal supplied to the second masking spectrum calculation circuit 75 is divided into three critical bands as described above.

In addition, in this specific example circuit, the masking on the frequency axis mentioned above is taken into account! In addition to setting the allowable noise level, the second allowable noise level is also set. Here, in FIG. 8, for example, when looking from block B2, the temporally preceding data is block B1.
(including previous blocks), and subsequent data becomes data of block B3 (including subsequent blocks). A second permissible noise level (masking level) for each frequency component in the block B2 is set based on data of at least one of these blocks B1 and B3 (or a plurality of blocks). Further, at this concrete level, at least one of the above at the second allowable noise level is temporally preceding data. That is, in temporal masking, the allowable noise level for block B2 is determined based on the data of the temporally preceding block B1, taking into account forward masking in which the masking effect takes a long time. Furthermore, in this specific example, the second allowable noise level is set based on the energy of the same frequency component in temporally different periods. That is, the second allowable noise level is set based on the energy of temporally preceding and following data in the same frequency band among the critical bands.

Therefore, the outputs of the energy detection circuit 22.42.62 are the period delay circuit 23.43.63 and the 5ms delay circuit 51 of the second noise level setting means, respectively.
The signal is sent to a 2.5 ms delay circuit 71.

Here, the period delay circuits 23, 43, and 63 delay the supplied data, for example, for each block period of 10 au. In addition, the period delay circuit 4
The outputs of 3 and 63 are sent to a combining circuit 50.70, respectively. The combining circuits 50 and 70 are the FF7 circuit 4.
The data of time blocks (5 mS+2..5 ms blocks) at 0.60 are combined into data of 10 m5 each. Furthermore, the 5 ms delay circuit 51
A delay is performed for each drive block, and the output of the 5 drive delay circuit 51 is sent to a selection circuit 52. The selection circuit 5
2, if the data of the supplied 5ms block is the previous block data in the 10m5 block currently being processed, it passes that data;
If the data is from a later block within a 10 m5 block preceding the block, switching is performed such that it is not allowed to pass. In other words, the currently processed 10m5
Assuming that the block is block B2 in FIG. 8, when the 5 ms block data supplied to the selection circuit 52 is block b (2°2, l) in FIG.
, 2°2), the selection is made to turn off. Above Z5
The delay circuit 71 delays each block of Z5ms, and the output of the Z5ms5ms delay 1 is sequentially sent to the 2.5ms delay circuits 72 and 73.

Each 15m5 delay circuit? The outputs of 1, 72, and 73 are sent to a combination selection circuit 74, respectively. The synthesis selection circuit 74 is turned off when the supplied 2.5 ms block data is the subsequent block b (1, 3° 4) in the preceding block B2 of the currently processed block B2, for example, Also, block b (2, 3, 1
), b (2,3°2L b (2,3,3), a switching selection is made such that it is turned on. At the same time, the synthesis selection circuit 74 selects, for example, block b (2,3,2).
When data is supplied, this block is combined with the previous block b (2, 3, 1), and block b (2, 3, 1) is combined.
3, When the data of 3) is supplied, this block and the previous 2
two blocks b (2, 3, 1), b (2, 3, 2
), when the data of block b (2,3゜4) is supplied, this block and the previous three blocks b (
2,3,l), b (2,3゜2), block b
(2, 3, 3).

The output of the period delay circuit 23 is sent to the weighted synthesis circuit 24, and the output of the synthesis circuit 501 selection circuit 52 is sent to the weighted synthesis circuit 4.
4, the output of the synthesis circuit 702 and the synthesis selection circuit 74 is sent to the weighted synthesis circuit 64.

Furthermore, data from the masking spectrum calculation circuit 75 is also supplied to each weighted synthesis circuit 24, 44, 64. Here, each weighted synthesis circuit 24.4
4.64 is for synthesizing weighting coefficients that take into account simultaneous time and spectral masking effects for supplied data.

That is, this weighting coefficient is a coefficient that is set in consideration of the masking effect. For example, if the signal of a block preceding or following the current block is normalized to l, A weighting coefficient corresponding to the level that acts on the signal of the current block based on frequency axis masking and temporal masking etc. by the signal of the following block is applied to the signal of the preceding or following block. weighted against. This makes it possible to set the allowable noise level (masking spectrum) using the frequency-axis and temporal masking effects.

Note that although the masking spectrum that takes into account the above-mentioned masking effect is obtained within the same critical band, it is also possible to take into account masking between other critical bands.

The outputs of these weighted synthesis circuits 24, 44.64 are further passed through synthesis circuits 25, 45.65 to a subtracter 27.
, 47.67. Here, these subtractors 27.
47.67 functions substantially in the same manner as the subtracter 121 shown in FIG. 3 described above, and detailed explanation thereof will be omitted. However, the bands handled by these subtracters 27°47.67 are the respective bands obtained by dividing the critical band into three, as described above.

The output of the subtracter 27, 47.67 is stored in the ROM 28.48.
．． Sent to 68. These ROM28°48.68 are also
It functions in substantially the same way as the ROM 122 in FIG. 3, and detailed explanation will be omitted. However, these ROMZ8.48
．． The bands handled by 68 are also divided into three critical bands.

For this reason, in this specific example circuit as well, the first masking method takes into account the simultaneous masking based on the energy of each critical band and the temporal masking. Second
The filter coefficients of the noise filter 13 are determined according to the allowable noise level.

The filter coefficients determined as described above are applied to the synthesis circuit 8.
5 performs full band synthesis, and is then sent from the output terminal 4 to the noise filter 13 in FIG.

Note that the delay circuits 30, 55.80 are provided in consideration of the amount of delay in the circuits after the critical band division circuit 21, 41.61, and the delay circuits 31.56.8
1 is provided in consideration of the amount of delay in each circuit after the period delay circuit 23°43.63.5ms delay circuit 51, 2.5ms delay circuit 71, and masking spectrum calculation circuit 75.

Furthermore, during the synthesis in the synthesis circuits 25, 45.65, data from the minimum audible curve generation circuit 26° 46.66, which is similar to the minimum audible curve generation circuit 129 shown in FIG. 3, is also supplied. There is. However, the data from these minimum audible curve generating circuits 26.46°66 are also data corresponding to each of the three bands of the critical band.

From the above, the audio signal quantization error reduction device of this embodiment can be used, for example, with standardized digital audio equipment (for example, so-called compact discs, etc.).
If applied to a digital audio tape recorder, etc., it becomes possible to obtain reproduced sound with a higher dynamic range perceptually than the dynamic range actually obtained from the unified standard. For example, the unified standard (the above CD
.

While maintaining the 16-bit slot word length standard (in the case of DAT) (without making any changes to the playback side and maintaining compatibility), we can change the perceptual dynamic changes in the playback sound of this audio signal. be able to raise the

〔Effect of the invention〕

The audio signal quantization error reduction device of the present invention frequency-analyzes the input digital signal of the block for each predetermined period, and determines the first permissible noise level based on the energy of each frequency component and the temporally preceding noise level of the block. By setting the filter characteristics of the noise filter based on at least one of the second allowable noise level based on the energy of at least one of the data to be processed,
It has become possible to reduce auditory noise and increase the auditory dynamic range.

Therefore, if the audio signal quantization error reduction device of the present invention is applied to, for example, digital audio equipment that has a unified standard, it will be possible to reproduce a dynamic range that is perceptually higher than the dynamic range actually obtained from the unified standard. You will be able to get sound. For example, it becomes possible to increase the perceptual dynamic range of the reproduced sound of this audio signal while maintaining the unified standard (while maintaining compatibility without making any changes to the reproduction side).

[Brief explanation of drawings]

Fig. 1 is a block diagram of a schematic configuration of an audio signal quantization error reduction device according to an embodiment of the present invention, Fig. 2 is a diagram showing a park spectrum, Fig. 3 is a block diagram showing a specific example circuit, and Fig. 4 is a masking spectrum. FIG. 5 is a diagram showing the minimum audible curve and masking spectrum combined. FIG. 6 is a block diagram showing another specific example circuit. FIG. 7 is a block circuit diagram showing an example of a filter bank. FIG. 8 is a diagram for explaining blocks in another specific example. 10.12... Adder 11... Quantizer 13... Noise filter 14... Filter coefficient calculation circuit 15...
- Blocking circuit 16... Frequency analysis circuit 17... First noise level setting circuit 18...
...Second noise level setting circuit patent applicant
Sony Corporation

Claims (1)

[Scope of Claim] An audio signal quantization error reduction device in which a quantization error generated in a quantizer is fed back to the input side of the quantizer via a noise filter, comprising: a blocking means for dividing each block into blocks; a frequency analysis means for performing frequency analysis on each block from the blocking means; and a frequency analysis means for performing frequency analysis on each block from the frequency analysis means, based on the energy of each frequency component in the block among the outputs of the frequency analysis means. a first noise level setting means for setting a first allowable noise level; and a first noise level setting means for setting a first allowable noise level; and a second noise level setting means for setting a second allowable noise level of each frequency component of the block based on the filter characteristics of the noise filter, the filter characteristics of the noise filter are set by the first and second noise level setting means. A quantization error reduction device for an audio signal, characterized in that the setting is made based on the output of at least one of the following.