JP2827410B2 - Efficient coding method for digital data - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital data high-efficiency encoding method for encoding input digital data by so-called high-efficiency encoding.

[Summary of the Invention]

The present invention relates to a high-efficiency encoding method for digital data in which a block composed of a plurality of samples is formed for each of a plurality of bands obtained by dividing input digital data, and orthogonal transform is performed for each block of each band to obtain coefficient data. At least the blocks before orthogonal transform of the lowest band are temporally divided into a plurality of small blocks to detect energy, and the gain of data for each block of the lowest band is controlled based on the energy. By doing so, it is possible to provide a high-efficiency digital data encoding apparatus capable of reducing noise generated by orthogonal transformation or the like, and in particular, reducing noise due to transient input.

[Conventional technology]

There are various high-efficiency coding methods for signals such as audio and voice. For example, band division coding (sub-band coding) for dividing an audio signal or the like on a time axis into a plurality of frequency bands and coding them. : SBC), so-called transform coding in which a signal on the time axis is converted into a signal on the frequency axis (orthogonal transform), divided into a plurality of frequency bands, and encoded for each band. Further, a high-efficiency coding method combining the above-described band division coding and transform coding is also considered.In this case, for example, after performing band division by the band division coding, The signal for each band is orthogonally transformed into a signal on the frequency axis, and the orthogonally transformed signal for each band is encoded for each band. Here, as the above-described orthogonal transform, for example, an orthogonal transform that converts an input audio signal into blocks in a predetermined unit time and performs a fast Fourier transform (FFT) for each block to convert a time axis into a frequency axis is used. is there. Further, when band-dividing the orthogonally transformed data on the frequency axis, band division may be performed in consideration of, for example, human auditory characteristics. That is, an audio signal may be divided into a plurality of (for example, 25 bands) bands with a bandwidth that is generally higher in a higher band called a critical band (critical band). Further, at this time, when encoding data for each band, encoding of the FFT coefficient data by predetermined bit distribution for each band or adaptive bit allocation (bit allocation) for each band. Is performed. For example,
In the coding by the bit allocation, the FFT coefficient data for each band obtained by the FFT processing for each block is coded with an adaptive number of allocated bits.

By the way, there is generally called a masking effect in human auditory characteristics of sound, and the masking effect includes a temporal masking effect, a same-time masking effect, and the like. The same-time masking effect is an effect in which a small sound (or noise) generated at the same time as a certain loud sound is masked by the loud sound and becomes inaudible, and the temporal masking effect is a time of a loud sound. The effect is that small sounds (noise) before and after the target are masked by the loud sounds and become inaudible. In this temporal masking effect, the temporally backward masking of the loud sound is called forward masking, and the temporally forward masking is called backward masking. Further, in temporal masking, the effect of forward masking is effective for a long time (for example, about 100 msec), whereas the effect of backward masking is for a short time (for example, about 5 msec) due to human auditory characteristics. ing. Further, the level of the masking effect (masking amount) is
Forward masking is about 20 dB, and backward masking is about 30 dB.

[Problems to be solved by the invention]

Here, when the audio signal in the predetermined unit time block is subjected to the fast Fourier transform at the time of encoding as described above, the inverse fast Fourier transform (IF
FT) is performed. In a signal obtained by such encoding and decoding, noise generated by the FFT and IFFT usually appears in the entire block. For this reason, for example, when a transient change occurs in the block subjected to the FFT / IFFT, for example, as shown in FIG. In the case where a signal C whose level sharply increases, such as a signal due to sound, enters and the transient change of the signal in the block becomes large, the FFT,
Noise due to the IFFT processing also appears in the non-signal portion U. That is, as shown in FIG. 6, a noise component caused by the large-level signal portion C also appears in the non-signal portion U. Therefore, when this signal is reproduced, noise in the portion where the signal was originally absent becomes noticeable.

Of the noise generated by processing such a block having a transient change by FFT, IFFT, etc., the noise temporally behind the large-level signal portion C is a long-time forward noise as shown in FIG. Since it is masked by the effect of masking FM, it is less likely to be heard. However, the temporally preceding noise of the large-level signal section C is backward masking.
Because the effect of BM is short, it is easy to catch the ear. That is, noise at a time before the time when the effect of the backward masking BM acts is noticeable.

As a countermeasure in the case where the effect of the backward masking BM as described above cannot be expected, for example, the unit time block length in which the fast Fourier transform processing is performed is set to a time range (for example, 5 msec) in which the backward masking BM works. It can be shortened. That is, it is conceivable to increase the time resolution (shortening the block length) in the high-efficiency encoding processing until the time when the effect of the backward masking BM by the large-level signal portion C effectively works.

However, shortening the unit time block length to be Fourier-transformed as described above means nothing less than reducing the number of samples in the block, so that the frequency resolution by the Fourier transform is reduced. However,
Frequency analysis ability (frequency resolution) in human hearing
Is generally not so high in the high range but high in the low range. Therefore, since the frequency resolution in the low frequency band needs to be ensured, the unit time block length cannot be actually shortened as described above. That is,
It is not preferable to increase the time resolution in the low band.

In general, a low-frequency signal has a long stationary section and a high-frequency signal has a short duration. Therefore, it is effective to increase the time resolution in a high frequency (shortening the block length).

Therefore, the present invention has been proposed in view of the above-described circumstances, and it is possible to obtain a high time resolution in a high frequency band, and further to obtain a high frequency resolution in a low frequency band, and It is an object of the present invention to provide a high-efficiency encoding method for digital data in which noise caused by a large-level signal portion in a block can be reduced in a low frequency range where the resolution cannot be increased.

[Means for solving the problem]

The high-efficiency encoding method for digital data of the present invention has been proposed to achieve the above-mentioned object, and divides input digital data into a plurality of bands and comprises a plurality of samples for each divided band. A high-efficiency encoding method for digital data in which blocks are formed and coefficient data is obtained by performing orthogonal transformation for each block of each band,
At least the block before orthogonal exchange of the lowest band is divided into a plurality of small blocks in time, and the energy of each small block is detected. Based on the detected energy, the small block of the lowest band is detected. The gain of each data is controlled. Here, in the energy detection of the small block, it is possible to detect at least a transient change in data of the block before the orthogonal transform of the lowest band. In this case, for example, in the gain control, the energy detection is performed. Based on the output, control is performed to reduce the gain of a small block having a transient change.
Further, when forming the block, it is possible to increase the block length (increase the time resolution) in the high band and increase the number of samples in one block (increase the frequency resolution) in the low band. Furthermore, coefficient data for each band to be quantized is stored in each band (for example, 25
Band).

[Action]

According to the present invention, by performing gain control for each small block according to the energy detected for each small block in at least the lowest band, noise due to orthogonal transformation of data in the block is reduced. Like that.

〔Example〕

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

The digital data high-efficiency encoding apparatus according to one embodiment to which the digital data high-efficiency encoding method of the present invention is applied, for example, converts input digital data such as audio or voice into the above-mentioned high-efficiency encoding band division. Encoding (SB
C) or the like. That is, in the high-efficiency encoding apparatus of the present embodiment, as shown in FIG. 1, a so-called mirror filter QMF (quadrature mirror filter) is used.
r) The input digital data supplied via the input terminal 30 is divided into a plurality of bands by 41 and 42 so that the higher the band, the wider the bandwidth, and a fast Fourier transform (FFT) circuit 4
3, 44, 45, a block composed of a plurality of samples is formed for each of the divided bands, and an orthogonal transform (for example, a time axis is converted to a frequency axis) by a fast Fourier transform is performed for each of these blocks. Coefficient data (FFT coefficient data) is obtained. At this time, when forming the block, the time resolution is increased by increasing the block length in the high frequency range, and the frequency resolution is increased by increasing the number of samples in one block in the low frequency range. In this embodiment, the coefficient data of each band to be quantized is coefficient data of each band (for example, 25 bands) in a so-called critical band (critical band).
The FFT coefficient data is quantized by the quantization processing unit 58 with an adaptive number of allocated bits, and then output from the output terminal 31.

Here, in the apparatus of the present embodiment, at least the block before orthogonal transformation of the lowest band is divided into a plurality of small blocks, and the energy detection circuit 50 (or the peak level is detected) for detecting the energy of each small block. Circuit) and a gain control circuit 60 as a gain control means for controlling the gain of the small block, and the gain of the gain control circuit 60 is controlled by the output of the energy detection circuit 50. . For example, when there is no signal, that is, when the energy value is 0 (for example, 0 dB)
Is controlled so as to lower the gain of the small block by an amount corresponding to the magnitude of the energy in the small block. Alternatively, when the energy in the small block is larger than the reference energy value with reference to an arbitrary energy value, the gain of the small block is increased by an amount corresponding to the difference between the reference energy value and the energy value in the small block. On the contrary, when it is smaller than the reference energy value, the control is performed in a direction to increase the gain. That is, for example, as shown in (a) of FIG.
Is divided into a plurality of small blocks Bs, and the energy detection circuit 50 detects the energy of each small block Bs. At this time, for example, when 0 dB at the time of no signal is used as a reference, as shown in FIG. 2B, each small block Bs is controlled by the gain control circuit 60 in accordance with the detection output of the energy detection circuit 50. The gain of data is controlled every time. By performing such gain control, it is possible to reduce the level of noise generated by orthogonal transformation of a signal having a large energy. When the energy detection is performed in small block units as described above, the data on which gain control is performed is delayed by the time required for detection in one small block unit.

Further, the energy detection circuit 50 may be configured to detect a transient change in data of a block before orthogonal transformation of at least the lowest band. in this case,
For example, the energy detection circuit 50 sequentially detects the energy of the small block and calculates the energy difference or ratio between the adjacent small block and the small block.
When this difference or ratio exceeds a predetermined value (ie, when there is a transient change), the gain control circuit 60
The gain of the small block having the transient change is controlled. For example, by performing control such that the gain is reduced only for small blocks having a transient change,
In other small blocks, a predetermined reference gain is set without performing gain control. That is, when the large-level signal section C as shown in FIG. 5 exists in the block B, the gain of only a small block having the large-level signal section C in the block B is reduced. To control. As described above, by controlling the gain of a small block having a transient change, it becomes possible to reduce the level of noise caused by the orthogonal transformation of the large-level signal portion C and the like. Further, when only small blocks having such a transient change are detected, in addition to the method of sequentially detecting and comparing the energy of the small blocks as described above, for example, the energy of all the small blocks in the block may be detected. After calculating, it is also possible to detect only a small block having a transient change and reduce the gain of this small block. By doing so, it becomes possible to detect only the small blocks for which the temporal masking cannot be effectively used, and by controlling the gain of the detected small blocks to be reduced, As described above, noise caused by orthogonal transformation or the like can be reduced. In this case, a delay of one block unit is performed.

Furthermore, by considering the above-described masking (temporal masking), it may be possible to control whether or not to perform gain control for each small block. That is, when a signal in an adjacent small block is masked by a signal in a certain small block, it is not particularly necessary to control the gain of the adjacent small block.

As described above, in the embodiment of the present invention, by performing gain control for each small block or gain control for only small blocks having a transient change, noise generated by orthogonal transform in a low band or the like is performed. Can be reduced, and noise generated due to the orthogonal transformation of the large-level signal portion C and the like can be reduced.

Further, in the present embodiment, when forming the above-mentioned block, the block length is increased in a high frequency range (time resolution is increased).
However, in the low frequency range, the number of samples in one block can be increased (frequency resolution can be increased).

That is, returning to FIG. 1 again, for example,
Audio digital data (0 to 24 kHz) sampled at a sampling frequency fs of 48 kHz is supplied, and the digital data is roughly divided into three bands (QMF41, 42) so that the higher the band, the wider the band becomes. 0 to
6kHz, 6kHz-12kHz, 12kHz-24kHz). Q above
In the MF 41, the digital data of 0 to 24 kHz is divided into two to obtain two outputs of 12 kHz to 24 kHz and 0 to 12 kHz. The outputs of 12 kHz to 24 kHz are output to the fast Fourier transform circuit 43.
The output of 0 to 12 kHz is sent to QMF42. The output of 0 to 12 kHz sent to QMF42 is further divided into two by QMF42 and
Two outputs of 12kHz and 0-6kHz are obtained. These outputs are sent to fast Fourier transform circuits 44 and 45, respectively.

In each of the fast Fourier transform circuits 43, 44, and 45, one block is constituted by a plurality of samples of the supplied data of each band, and the Fourier transform process is performed for each block to obtain FFT coefficient data. At this time, the fast Fourier transform circuit 43 forms one block with 64 samples,
The FFT coefficient data is obtained for each block. As a result, the time resolution in the band of 12 kHz to 24 kHz is a high time resolution of about 2.67 msec. The fast Fourier transform circuit 44 obtains FFT coefficient data in 64 samples per block. As a result, the time resolution at 6 kHz to 12 kHz is about 5.3 msec. Further, in the fast Fourier transform circuit 45, since the FFT coefficient data is obtained by 128 samples per block, the time resolution at 0 to 6 kHz is about 10.67 m.
sec.

As described above, in this embodiment, the high frequency range (12 kHz to 24 k
Hz) and the mid-range (6kHz to 12kHz) time resolution is 2.67mse
c and 5.3 msec, the large-level signal section C
Even if noise as shown in FIG.
In the high and middle ranges, the backward masking by the large-level signal portion C in the block can be effectively used (the effect time is about 5 msec). Also,
In the device of the present embodiment, since the frequency resolution in the low frequency band needs to be secured,
0.67 msec, and even with a time resolution of 10.67 msec, it is possible to cope with noise caused by the large-level signal portion C in the block in the low frequency band. That is, as described above, the block before the orthogonal transform of the low band is divided into a plurality of small blocks, and the energy of each small block is detected (or the small block having a transient change portion is detected). By controlling the gain of each small block (or a small block having a transient change portion) in accordance with the detection output, noise generated by orthogonal transformation of the large-level signal portion C in the block in the low-frequency range, and the like. (Noise reduction).

Here, FIG. 3 shows the resolution in the frequency domain and the time domain in this embodiment. FIG. 3 shows one unit of processing such as the above-described band division and fast Fourier transform,
A block is specified by two parameters m and n in (m, n). m indicates a band number, and n indicates a time number. In FIG. 3, one block in each band has a time length (time resolution) of 10.67 msec in the low band of 0 to 6 kHz. Also, 6KHz ~ 12kHz
In the middle band, the time length of one block is 5.3 msec, and in the high band of 12 kHz to 24 kHz, the time length of one block is 2.67.
msec.

As described above, in the present embodiment, the resolution on the frequency axis and the resolution on the time axis required from the auditory sense are simultaneously satisfied. The frequency resolution is increased by increasing the number, and in the high frequency range (12 kHz to 24 kHz), the bandwidth is widened and the time resolution is also increased. The time resolution is also increased in the middle band (6 kHz to 12 kHz. The detection and gain control processing by the energy detection circuit 50 and the gain control circuit 60 are performed only in the low band. The number of malfunctions is smaller than when performing control.

Further, in the present embodiment, at the time of quantization in the quantization processing unit 58, quantization is performed with an adaptively allocated number of bits in consideration of masking based on human auditory characteristics. Conversion circuit output (FFT coefficient data)
Is also associated with each of the critical bands (for example, 25 bands) based on the human auditory characteristics. That is, the output of the fast Fourier transform circuit 43 corresponds to the two bands of the higher critical band B24 and bandon B25, and the output of the fast Fourier transform circuit 44 corresponds to the three bands B21 to B23 and the high speed band. The output of the Fourier transform circuit 45 is made to correspond to 20 bands of the lower bands B1 to B20 of the critical band.

In the case where the above-described gain control of the present embodiment is performed, gain control information is also transmitted to the decoder side together with the quantized FFT coefficient data, and a decoding process based on the gain control information is performed. Become like

FIG. 4 shows a state of band division on the frequency axis in the above-described embodiment.

In FIG. 4, bands B1 to B20 in the low band (0 to 6 kHz)
The number of coefficient data in is, for example, one for each of bands B1 to B8,
Two bands B9 to B11, three bands B12 and B13, four bands B14 to B16, six bands B17 and B18, nine bands B19, and eleven bands B20. Mid frequency (6kHz-12
kHz), the number of coefficient data in bands B21 to B23 is, for example,
The number of bands B21 is 7, the number of bands B22 is 11, and the number of bands B23 is 14. Also, bands B24 and B25 for high frequency (12kHz ~ 24kHz)
Is 16 for example.

〔The invention's effect〕

In the high-efficiency encoding method for digital data according to the present invention, the gain of a small block is controlled based on an energy detection output of each small block in which at least the block before the orthogonal transform in the lowest band is divided into small blocks. By doing so, it is possible to reduce noise generated by orthogonal transformation or the like, and in particular, it is possible to reduce noise due to transient input.

In addition, a high time resolution can be obtained in a high frequency range, and a high frequency resolution can be obtained in a low frequency range where the time resolution cannot be increased.

[Brief description of the drawings]

FIG. 1 is a block circuit diagram showing a schematic configuration of a high-efficiency digital data encoding apparatus according to an embodiment of the present invention. FIG. 2 is a view for explaining small blocks whose gain is controlled by the apparatus according to the embodiment of the present invention. FIG. 3 is a diagram showing resolutions in a frequency domain and a time domain, FIG. 4 is a diagram showing a state of band division, FIG. 5 is a diagram for explaining data before fast Fourier transform in which a transient change exists, FIG. 6 is a diagram for explaining generation of noise after fast Fourier transform and inverse fast Fourier transform.
The figure is a diagram for explaining temporal masking. 41, 42 QMF 43 to 45 Fast Fourier transform circuit 50 Energy detection circuit 58 Quantization processing unit 60 Gain control circuit

Claims (1)

(57) [Claims] 1. A digital system in which input digital data is divided into a plurality of bands, a block including a plurality of samples is formed for each of the divided bands, and orthogonal transform is performed for each block of each band to obtain coefficient data. A high-efficiency coding method for data, wherein at least a block before orthogonal transform of the lowest band is divided into a plurality of small blocks, and energy of each small block is detected. A high-efficiency encoding method for digital data, wherein a gain of data for each small block in the lowest band is controlled based on the following.