JPH04104618A - Digital signal coder - Google Patents

Abstract

PURPOSE:To attain coding with less sound quality deterioration even with low bit rate compression by providing a band split means, a noise level setting means, a control means and a quantization means to the coder. CONSTITUTION:The coder is provided with a band split circuit 13 splitting an input digital voice signal into plural frequencies bands and a noise level setting means 40 calculating the effect on other band based on energy for each split band and setting an allowable noise level for each band. Moreover, a control means 41 controlling the effect on other band with the means 40 in response to the tonality for each band and a quantization circuit 24 quantizing a signal component of each band in a bit number in response to a difference with an allowable noise level controlled in response to the energy and the tonality of each band are provided. Since the signal component of each band of an input digital voice signal is quantized in a bit number in response to a difference from the allowable noise level controlled in response to the energy and the tonality of each band in this way, the deterioration in the sound quality is reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a digital signal encoding device that encodes an input digital audio signal.

[Summary of the invention]

The present invention divides an input audio signal into multiple bands, determines the degree of influence on other bands based on the energy and tonality of each band, sets an allowable noise level for each band, and sets the allowable noise level for each band. By quantizing each band component with the number of bits according to the difference from the energy of each band, or filtering the energy of each band to calculate the degree of influence on other bands, or The signal component of each band is separated into a single frequency component and a white noise component, and each is filtered and synthesized using different filter characteristics.Furthermore, the filter characteristics are varied according to an index indicating the tonality of each band. This provides a digital signal encoding device capable of encoding with little deterioration in sound quality even when compressed to a low bit rate.

(Prior art) In high-efficiency encoding of audio, voice, etc. signals,
2. Description of the Related Art There is an encoding technique using bit allocation that divides an input signal such as audio or voice into a plurality of channels on a time axis or a frequency axis and adaptively allocates the number of bits for each channel. For example, the encoding technology using the above-mentioned bit allocation for audio signals, etc.
Band division coding (sub-band coding: 5BC), which divides and encodes audio signals on the time axis into multiple frequency bands, and converts time-axis signals to signals on the frequency axis (orthogonal transformation) So-called adaptive transform coding (A
TC), or by combining the above SBC and so-called adaptive predictive coding (APC), dividing the time domain signal into bands and converting each band signal to the baseband (low band), and then performing multi-order linear predictive analysis. There are coding techniques such as so-called adaptive bit allocation (APC-AB) that performs predictive coding.

In recent years, many attempts have been made to develop high-efficiency encoding methods that take into account the so-called masking characteristics of human auditory characteristics.

The masking effect refers to a phenomenon in which a certain signal masks another signal so that it becomes inaudible, and the masking effect includes, for example, a masking effect on an audio signal on the frequency axis.

Here, in the masking effect on the audio signal on the frequency axis, for example, if there is a sine wave W of a certain frequency f, a masking spectrum (masking curve) MS indicating the masking effect due to human hearing
is as shown in Fig. 8, and this masking spectrum M
By S, the shaded area in the figure is masked. That is, even if there is noise within the masking spectrum MS, it will not be audible, so that the noise within the masking spectrum MS can be tolerated in an actual audio signal. Therefore, the allowable noise level in the case of the sine wave W8 is lower than the level indicated by j in FIG.

Also, at this time, the masking effect is highest at the frequency f6 of the sine wave W, and the masking effect decreases as the frequency moves away from the frequency f of the sine wave W.

[Invention or problem to be solved]

For this reason, when performing high-efficiency encoding, the number of bits allocated for quantization is reduced for signal components below the allowable noise level, taking into account the masking effect described above. , it becomes possible to further reduce the bit rate.

By the way, it is currently desired to further reduce the bit rate. However, if the bit rate is further reduced, the sound quality will deteriorate, which is not desirable.

The present invention was proposed in view of the above-mentioned circumstances, and an object of the present invention is to provide a digital signal encoding device that can perform encoding with less deterioration in sound quality even when compressed to a lower bit rate. This is the purpose.

[Means to solve the problem]

The digital signal encoding device of the present invention has been proposed to achieve the above-mentioned object, and includes a band dividing means for dividing an input digital signal into a plurality of frequency bands, and a frequency band for each frequency band divided by the band dividing means. Noise level setting means that calculates the degree of influence on other bands based on the energy of each band and sets an allowable noise level for each band;
A control means for controlling the degree of influence of the noise level setting means on other bands according to the tonality of each band, and a difference between the energy of each band and the allowable noise level controlled according to the tonality. quantization means for quantizing the signal components of each band with the number of bits corresponding to It has a filter means for calculating the degree of influence on the band, and further,
The noise level setting means has two filter means each having different characteristics, into which the signal components of each band are separated into a single frequency component and a white noise component. , the outputs from the respective filters of the noise level setting means are synthesized, and furthermore, the control means detects an index indicating tonality for each band, and adjusts the noise level according to the index indicating tonality. The characteristics of the filter means of the setting means are made variable.

In general, a signal component with high tonality has a large influence on other bands, that is, a masking effect, and conversely, a signal component with low tonality has a small masking effect.

[Effect]

According to the present invention, the signal components of each band of the input digital audio signal are quantized with the number of bits corresponding to the difference between the energy of each band and the allowable noise level controlled according to the tonality. Deterioration can be reduced.

〔Example〕

Embodiments to which the present invention is applied will be described below with reference to the drawings.

The digital signal encoding apparatus of this embodiment divides an input digital audio signal into a plurality of frequency bands, as in the first embodiment shown in FIG. 1 and the second embodiment shown in FIG. Based on the band division circuit 13 and the energy of each band divided by the band division circuit 13, the degree of influence on other bands (that is, the amount by which the signal of one band masks the signal of another band) is determined. Noise level setting means 40 that calculates and sets an allowable noise level for each band, and control means that controls the degree of influence of the noise level setting means 4o on other bands according to the tonality of each band. 41, and a quantization circuit 24 that quantizes the signal component of each band with the number of bits corresponding to the difference between the energy of each band and the allowable noise level controlled according to the tonality. It is. Further, the noise level setting means 4o includes filter circuits 15s and 15w as filter means for receiving the energy of each band and calculating the degree of influence on the other bands.
(FIG. 1) and a variable filter circuit 15 (FIG. 7). Furthermore, as shown in FIG. 1, the noise level setting means 40 separates the signal component for each band into a single frequency component and a white noise component by the component separation circuit 31, and separates each component into a single frequency component and a white noise component. the two input filter circuits 15s each having different characteristics;
15w, and the control means 41 is configured to synthesize the outputs from the respective filter circuits of the noise level setting means 4o using the synthesis circuit 32 (and the synthesis circuit 18). Furthermore, as shown in FIG. 7, the control means 41 detects an index indicating tonality (for example, standard deviation, etc.) for each band with a tonality index detection circuit 51, and according to the index indicating tonality, The characteristics of the variable filter circuit 15 of the noise level setting means 40 are made variable. The quantized output from the quantization circuit 24 is sent to the output terminal 2 of the digital signal encoding device of this embodiment.
will be output from.

Here, the digital signal encoding device of this embodiment shown in FIGS. 1 and 7 performs fast Fourier transform (FFT) on an input voice (audio) signal to convert a time-axis signal into a frequency-axis signal. , which performs encoding (requantization).

That is, first, in FIG. 1, an input audio signal on the time axis supplied to the input terminal l is transmitted to the fast Fourier transform circuit 11. This fast Fourier transform circuit 11
Then, the audio signal on the time axis is converted into a signal on the frequency axis at every predetermined time (unit block), and an FFT coefficient consisting of a real component value Re and an imaginary component value 1m is obtained. These FFT coefficients are transmitted to the amplitude and phase information generation circuit 12, and the amplitude and phase information generation circuit 12 obtains an amplitude value Am and a phase value from the real component value Re and the imaginary component value 1m, and the amplitude value Am information will be output. In other words, human hearing generally depends on frequency domain amplitude (
Although it is sensitive to power), it is quite insensitive to phase, so in this embodiment, the amplitude and phase information generating circuit 1 is
Only the amplitude value Am is taken out from the output of No. 2, and this is used as the input digital signal in the embodiment of the present invention.

The input digital signal with the amplitude value Am thus obtained is transmitted to the band division circuit 13.

This band division circuit 13 divides the input digital signal expressed by the amplitude value Am into so-called critical bandwidths. The critical band is one that takes human auditory characteristics (frequency analysis ability) into consideration, and is, for example, 24 bands (or 0 to 16).
The frequency band is divided into 22 and 25 bands), and the high frequency band has a wide band width. That is, human hearing has characteristics like a kind of band-pass filter, and the bands divided by each filter are called critical bands. Here, FIG. 2 shows the critical band. However, in this Figure 2, in order to simplify the illustration, the number of critical bands is reduced to 1.
It is expressed in two bands (B, ~B12).

The amplitude value Am for each band (for example, 24 bands) divided into critical bands by the band division circuit 13 is transmitted to the noise level setting means 40. This noise level setting means 40 calculates the degree of influence (masking amount) on other bands based on the energy of each band, and sets the allowable noise level for each band. It is controlled according to the level or tonality of each band.

By the way, in general, a signal component with high tonality, such as a single frequency component, for example, a sine wave component Si, has a large influence on other bands, that is, a masking effect, and -
It is known that a signal component with low tonality, such as a white noise component WN, has a small degree of influence (masking effect).

For this reason, in this embodiment, the amplitude value Am is sent to the component separation circuit 31 of the noise level setting means 40, and the component separation circuit 31 extracts the signal components of the single frequency from the signal components of each band. Sine wave component Si as a component
and a white noise component WN, and then a control means 41, which will be described later, controls the output of the noise level setting means 40 to obtain an allowable noise level according to the tonality of each band. .

In order to perform the above, the sine wave component Si and white noise component WN are detected by a sine wave component sum detection circuit 14s and a white noise component sum detection circuit 14, respectively.
transmitted to w. These sum detection circuits 14s and +4w detect the energy (spectral intensity in each band) of the sine wave component Si or white noise component WN for each band, or the sum (amplitude value) of the respective amplitude values Am in each band. It is obtained by taking the peak, average, or total energy of Am. The sum spectrum of components for each band is generally called a park spectrum, and the park spectrum SB for each band is as shown in FIG. 3, for example. That is, the sum detection circuits 14s, 14
In w, the park spectrum SB for each band of the sine wave component St or white noise component WN is determined.

Here, in order to consider the influence of the sine wave component Si or white noise component WN on masking of each park spectrum SB, each park spectrum SB is convolved with a predetermined weighting function (convolution).
. That is, the outputs of the summation detection circuits 14s and 14w (
Each value of each park spectrum SB) is a filter means to which the energy of each band is input and calculates the degree of influence on the other bands, and two filter circuits 15s each have different filter characteristics.

Sent to 15W. These filter circuits 15s.

15w are delay (z-I) elements that sequentially delay human data, as shown in FIG. 4, respectively.
-z~101. +3..., a multiplier that multiplies the output from each of these delay elements by a filter coefficient (weighting function)...102=-s~102-, s..., and a summation adder 1
04.

At this time, each of the multipliers 102. ．． ~102. . The filter coefficients in , are respectively predetermined values for the sine wave component St or the white noise component WN, and therefore the filter circuits 15s and 15W have different characteristics. Specifically, the filter coefficients in the filter circuit 15s to which signal components with high tonality are supplied are coefficients that provide filter characteristics such that the degree of influence on other bands is large, and the The coefficients in the filter circuit 15w to which the components are supplied are coefficients that provide characteristics such that the degree of influence of other bands is reduced. By multiplying the output of each delay element by these filter coefficients in each multiplier, convolution processing of the park spectrum SB is performed. For example, by this convolution process, the sum of the parts shown by the dotted line in FIG.

Thereafter, the outputs of the filter circuits 15s and 15w are sent to subtracters 16s and 16w, respectively.

The subtracters 16s and 16w each obtain a level α corresponding to a permissible noise level, which will be described later, in a region convolved for each band of the sine wave component Si or white noise component WN. Note that the level α corresponding to the above-mentioned allowable noise level is a level that becomes the allowable noise level for each critical band by performing inverse convolution processing, as will be described later. Here, the subtracters 16s and 16w are supplied with a tolerance function (a function expressing the masking level) for determining the level α. The level α is controlled by increasing or decreasing this tolerance function. The cough tolerance function is supplied from function generation circuits 29s and 29w, which will be described later.

In other words, the level α corresponding to the allowable noise level can be determined by equation (1), where i is the number given in order from the low range of the critical band.

α= 5-(n-ai)・・・・・・(1) Second (
In formula (1), n and a are constants, ago and S are the intensity of the park spectrum after convolution processing, and in formula (1), (
n-ai) becomes the tolerance function. In this embodiment, for example, n=38. a=1 or n=24. It is assumed that a=1.

The level α is determined as described above, and each data is transmitted to the dividers 17s and 17w, respectively. The dividers 17s and 17w are for inversely convoluting the levels α in the respective regions subjected to the convolution processing. Therefore, by performing these inverse convolution processes, a masking spectrum can be obtained from the level α. That is, these masking spectra become the permissible noise level for the sine wave component S1 and the white noise component WN. Note that the above deconvolution process requires complicated calculations, but in the device of this embodiment, a simplified divider 17s
, 17w is used to perform inverse convolution.

Note that, normally, the masking spectrum MS is, for example, the fifth
The result will be as shown in the figure. That is, as will be described later, the Nox vector SB is based on the masking spectrum M
Levels below the level indicated by each level of S will be masked.

Next, the outputs of these dividers 17s and 17w are sent to control means 41 and combined. In this control means 41, the signals are first synthesized in the synthesis circuit 32, and then sent to the synthesis circuit 18. The synthesis circuit 32 and the dividers 17s and 17
By combining the outputs of w, the tonality of each band is taken into consideration in the degree of influence on other bands in the noise level setting means 40. That is, by combining the allowable noise level based on the sine wave component St with high tonality and the allowable noise level based on the white noise component WN with low tonality, the degree of influence on other bands according to the tonality of each band can be reduced. You will be asked for it. The synthesis circuit 18 is also supplied with an output from a minimum audible curve generation circuit 22. Once, when the synthesis circuit 18 synthesizes, the so-called minimum audible curve (equal loudness curve) RC, which is the human auditory characteristic as shown in FIG. 6, is supplied from the minimum audible curve generation circuit 22. data and the combined output of the masking spectrum MS of the sine wave component Si and white noise component WN, which is the output of the synthesis circuit 32, that is, the output with tonality added to the allowable noise level for each band. It turns out. in this way,
By combining the above-mentioned minimum audible curve RC and masking spectrum MS, the allowable noise level can be set to, for example, the area shown by diagonal lines in Fig. 6. It becomes possible to reduce the number of allocated bits in the part shown by . Note that this FIG. 6 shows the critical band shown in FIG. 2 described above, and also shows the signal spectrum SS at the same time.

Thereafter, the output of the synthesis circuit 18 is transmitted to a subtracter 19. Here, the subtracter 19 has a total sum of amplitude values Am (the peak of the amplitude value Am or The output of a summation detection circuit 14 which takes the average or energy summation is supplied via a delay circuit 21.

Therefore, in the subtracter 19, the total energy of each band, that is, the bark spectrum YSB by the ρ signal component for each band, and the degree of influence on other bands are controlled by the control means 41 according to the tonality. A calculation is performed to calculate the difference between the output from the noise level setting means 40 and the allowable noise level (masking spectrum MS). From this, the level at which the park spectrum SB of the signal component of each band is masked is determined.

The output of the subtracter 19 is supplied to a quantization circuit 24 via a ROM 20. Here, the ROM 20 stores information on the number of allocated bits during quantization in the quantization circuit 24, and outputs information on the number of allocated bits in accordance with the output of the subtracter 19. Therefore, the quantization circuit 24 quantizes the amplitude value Am supplied via the delay circuit 23 using the number of bits assigned according to the output of the subtracter 19. In other words, in the quantization circuit 24, the energy of each band of the critical band and the output (
The components of each band are quantized using the number of bits allocated according to the level of the difference from the permissible noise level (according to the tonality). The delay circuit 21 is provided in consideration of the amount of delay in each circuit before the synthesis circuit 18, and the delay circuit 23 determines the amplitude value Am in consideration of the amount of delay in each circuit before the ROM 20. It is designed to delay.

Next, in the second embodiment of the apparatus shown in FIG. 7, the same components as those in FIG. 1 except for the noise level setting means 40 and the control means 41 are given the same reference numerals, and the explanation thereof will be omitted. In this second embodiment, the control circuit 41 detects an index indicating the tonality of each band, and the filter characteristics of the variable filter 15 of the noise level setting means 40 are varied according to the index indicating the tonality. Also in this second embodiment, it is possible to determine the allowable noise level according to the tonality of each band.

That is, in this second embodiment, the signal components for each band from the band division circuit 13 are sent to the summation detection circuit 14 that calculates the energy sum of the signal components for each band, and the signal components for each band are Tonality index detection circuit 51 that detects an index indicating tonality directly from the components
It will also be sent to

The tonality index detection circuit 51 can use, for example, the standard deviation σ, variance σ2, etc. of the signal components for each band as the tonality index. That is, for example, the standard deviation σ is calculated by the following formula. In this formula (2), Xl is a variable,
N is the total number and X is the arithmetic mean. Regarding this standard deviation σ, when the standard deviation σ is large, the tonality is low, and when the standard deviation σ is small, the tonality is high. Of course, the standard deviation is not limited to this standard deviation σ, and other values may be used.

At this time, the tonality index detection circuit 51 has a terminal 5.
Standard values are supplied starting from 0. This standard value is used to detect the tonality ratio of the signal component for each band by comparing it with the value of the standard deviation σ. That is, for example, the percentage of tonality for each band is determined from the ratio between the standard value and the value of the standard deviation σ. Therefore, for example, when the percentage of tonality is high, there will be many single frequency components such as the sine wave component in the first embodiment, and when the percentage of tonality is low, there will be many components such as white noise components. There will be many.

The output of the tonality index detection circuit 51 is sent to a filter coefficient setting circuit 52. The filter coefficient setting circuit 52
outputs a filter coefficient (according to the percentage of tonality) according to the comparison result between the tonality index and the standard value, and the output is sent from the filter coefficient setting circuit 52 or to the variable filter circuit 15. It will be done. The variable filter circuit 15 also has a configuration similar to that shown in FIG. ing. Therefore, the filter characteristics of the variable filter circuit 15 are:
The filter coefficient setting circuit 52 changes the filter coefficient in accordance with the detection result of the tonality.

The output of this variable filter 15 or the subtracter 16s in FIG.
, 16w, function generation circuit 29s 29w divider +7s
, a subtracter 16 and a function generation circuit 2 that operate in the same way as +7w.
The signal is sent to the synthesis circuit 18 via the 91 divider 17.

Furthermore 1. From the ROM 20, a synthesis circuit 18 and a delay circuit 21
Information on the number of allocated bits for quantization according to the result of subtracting the output from is output and sent to the quantization circuit 24. Thereby, in the quantization circuit 24, the signal component supplied via the delay circuit 23 is quantized using the number of bits allocated according to the output of the subtracter 19. In other words, in the quantization circuit 24, bits are allocated according to the level of the difference between the energy of each band of the critical band and the output of the synthesis circuit 18 (allowable noise level according to tonality). The components of each of the above bands are quantized by the number.

As described above, in the digital signal encoding apparatus of the first embodiment shown in FIG. 1 and the second embodiment shown in FIG. 7, the energy and tonality of each critical band are By quantizing the components of each of the above bands with the number of bits allocated according to the level of the difference from the corresponding allowable noise level, it becomes possible to reduce sound quality degradation and reduce the bit rate. .

In addition, 1. In the second embodiment, band division is performed on the output of the fast Fourier transform of the input signal on the time axis.
The time axis signal may be directly band-divided without performing this fast Fourier transform.

Furthermore, in the embodiments of the present invention described above, only the masking effect on the signal on the frequency axis is described, but the present invention also provides the masking effect on the frequency axis as well as the masking effect on the signal on the frequency axis.
For example, a so-called temporal masking effect on the time axis may also be considered. By using this temporal masking effect together with masking on the frequency axis, it becomes possible to set an allowable noise level that takes into account the degree of influence of the tonality of a signal in a certain band on other bands at other times. Higher bit compression is possible.

Here, the above-mentioned temporal masking effect is an effect in which small sounds temporally before and after a loud sound are masked by the loud sound and become inaudible. In the temporal masking effect, masking temporally behind the loud sound is called forward masking, and masking temporally forward is called packed word masking. In addition, in temporal masking, due to the characteristics of human hearing, the effect of forward masking is effective for a long time (for example, about 100 m5ec), whereas the effect of packed word masking is effective for a short time (for example, about 5 m5ec). It becomes. Further, the level of the masking effect (masking amount) is about 20 dB for forward masking and about 30 dB for packed word masking.

〔Effect of the invention〕

In the digital signal encoding device of the present invention, an input audio signal is divided into multiple bands, and based on the energy and tonality of each band, the degree of influence on other bands is determined, and the allowable noise level for each band is set. Then, by quantizing each band component with the number of bits according to the difference between this allowable noise level and the energy of each band, the energy of each band is also filtered to determine the influence on other bands. Alternatively, the signal component of each band can be separated into a single frequency component and a white noise component, each filtered with different filter characteristics, and then synthesized. Furthermore, an index indicating the tonality of each band can be calculated. By varying the filter characteristics according to the signal, it is possible to perform encoding with little deterioration in sound quality even when compressed to a low bit rate.

[Brief explanation of drawings]

FIG. 1 is a block circuit diagram showing a schematic configuration of a digital signal encoding device according to a first embodiment of the present invention, FIG. 2 is a diagram showing a critical band, FIG. 3 is a diagram showing a park spectrum, and FIG. 5 is a circuit diagram showing a filter circuit, FIG. 5 is a diagram showing a masking spectrum, FIG. 6 is a diagram in which the minimum audible curve and the masking spectrum are synthesized, and FIG. 7 is an outline of the digital signal encoding device of the second embodiment. FIG. 8, a block circuit diagram showing the configuration, is a diagram for explaining masking of signals on the frequency axis by human hearing. 13... Band division circuit 14s, 14w, 14... Sum detection circuit 15s
, 15w, 15... Filter circuit 6s, 16w
, 16...subtractor 7s, 17w, 17...
...Divider 8.32...Synthesizing circuit 9...Subtractor 0...ROM 1, 23...Delay circuit 2...Minimum Audible curve generation circuit 4...
...Quantization circuit 0...Noise level setting means L...
... Control means l ... Tonality index detection circuit 2 ...
...Filter coefficient setting circuit patent applicant Sony Corporation

Claims (4)

[Claims] (1) A band dividing means for dividing an input digital signal into a plurality of frequency bands; and a band dividing means for calculating the degree of influence on other bands based on the energy of each band divided by the band dividing means. noise level setting means for setting an allowable noise level; control means for controlling the degree of influence of the noise level setting means on other bands according to the tonality of each band; and the energy of each band and the tonality. 1. A digital signal encoding device comprising: quantization means for quantizing a signal component of each band with a number of bits corresponding to a difference from a permissible noise level controlled according to. (2) The noise level setting means includes filter means for receiving the energy of each band and calculating the degree of influence on the other bands.
). (3) The noise level setting means has two filter means each having different characteristics, into which the signal components of each band are separated into a single frequency component and a white noise component. 3. The digital signal encoding apparatus according to claim 2, wherein the control means synthesizes outputs from each of the filters of the noise level setting means. (4) The control means detects an index indicating tonality for each band, and varies the characteristics of the filter means of the noise level setting means according to the index indicating tonality. ).