JPS59214346A - Subband encoding method and its encoding decoder - Google Patents

Abstract

PURPOSE:To simplify a device and shorten transmission delay by inputting and dividing a sound signal into plural subbands, and providing encoding means which input signals of the respective subbands and perform adaptive quantization for quantizing the signals while updating step size successively. CONSTITUTION:A signal which is inputted from a terminal 1 and sampled is divided into N subbands through filters 11, 12-1N and sampled according to divided band width. The band-divided signals are inputted to adaptive quantizers 21, 22-2N. One of those signals, i.e. signal 100 is quantized by the adaptive quan tizer 21. The output of a gain control circuit 210 is applied to a quantizing circuit 211 and encoded into a signal 103 as a number l of a quantized level, and the signal is outputted. Outputs of adaptive reverse quantizers 61, 62...6N are inputted to frequency converting filters 81, 82...8N and shifted in frequency to their original frequency bands. The outputs of the respective frequency dconvertig filters 81, 8N are summed up by an adder 52, whose output is sent out as a reproduced signal from a terminal 4.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an improvement in a voice encoding method for efficiently encoding a voice signal by dividing it into a plurality of nine bands.

Adaptive quantization is a method for efficiently quantizing audio signals. Simply put, an AGO (automatic gain adjustment circuit) is placed in front of the quantizer to maintain a constant average amplitude (an automatic gain adjustment circuit) is placed in front of the quantizer, and after the quantization result is decoded (multiplied by a gain opposite to the AGC here) In this way, even if a signal with large level fluctuations is input, the level at the input of the quantizer will always be almost constant, so distortion due to quantization can be reduced even with a small number of bits. As another implementation method of the quantization method, even if the quantization interval (step size) of the quantizer is all sequentially moved, an effect equivalent to the case where OkoAGO is used as described above can be obtained.

Now, the case where one signal is quantized has been described above, but the signal is often divided into a plurality of bands and quantized in each band, as in the case of subband coding.

In the case of 7.1', if there is a level difference between a plurality of bands, quantization with less distortion can be realized by allocating the number of quantization bits according to the level difference. This encoding method is described in Mr. Ota's paper ("Study of subband encoder configuration method", Institute of Electronics and Communication Engineers communication system study group materials, O8'
79-22゜1979), so a detailed explanation will be omitted, but the conclusion is that when encoding, it is necessary to transmit information on how many bits are encrypted to each subband on the decoding side. This method of changing bit allocation according to the signal number is called adaptive bit allocation.According to the above paper, adaptive bit allocation improves transmission quality, or instead increases hardware becomes complicated. Therefore,
It is said that adaptive bit allocation is not necessarily a good idea.

A conventional method of subband encoding using adaptive bit allocation ζ will be explained in detail using figures.

FIG. 1 shows an example of a conventional subband encoder, and is a block diagram when the subband encoder is divided into two bands.

The left half of FIG. 1 is an encoder, and the right half is a decoder.

In the encoder, an audio signal sampled at 8 KHz is input from terminal 1, and only the low frequency is extracted by filter JO and output at a sampling frequency of 4 KHz. Cough, filter J5 extracts only the high frequency range! 1. Single frequency 4
Output at KHz. The bit divider and half circuit 44 calculates the power of the two signals and determines the quantization step size and number of quantization bits of the quantizer 20 and the quantizer 30, respectively. Here, it is assumed that the first number of bits allocated to both signals is constant. The algorithm for determining the quantization step size and the number of quantization bits can be found in the paper mentioned above. 3. The demultiplexer 50 inputs the received signal and divides the information regarding the bit allocation (based on this into the first inverse quantizer 60, second inverse quantizer 70, and f. do.

Koi 1 et al)] I! The quantizer 60.70 operates inversely to the encoder's quantizer 20.30. In other words, the original signal is reproduced from the quantized signal. The output of the inverse quantizer 60 is input to a frequency conversion filter 80. The frequency conversion filter 80 interpolates the 4KHz sampled signal output from the inverse quantizer 60, samples it at 13K, l-1z, and outputs it as a low frequency component. Also, the frequency gI
The ψ filter 90 inputs the output of the inverse quantizer 70, and while interpolating it, shifts the frequency to a high frequency range and outputs it as a wood-based high frequency component. The double calculator 51 adds the outputs of these two frequency conversion filters 80 and 90 and outputs the result from the terminal 4 as a reproduced audio signal. Using this embodiment, two bands are equal; quantization distortion is reduced compared to increasing the number of quantization bits to 11. If the power of the low-frequency signal is higher than that of the high-frequency signal, the number of bits applied to the low frequency band will increase, and the 8N ratio (signal to quantization distortion) on the low frequency side will increase. ratio) increases. On the other hand, the number of bits allocated to the high frequency band decreases, and the S/N ratio on the low frequency side decreases, but if the signal is small to begin with, the quality deterioration will be minimal. The above explanation is probably one-size-fits-all, but it is theoretically explained in the paper by Mr. Ota mentioned above.

As stated above, using adaptive bit allocation can reduce the effects of quantization distortion by using adaptive bit allocation, but there is also the problem that information regarding bit allocation must be transmitted, and that the device becomes multiple There is a point.

Furthermore, in the bit allocation circuit 44, it is necessary to find the average level of the signal, so in the quantizer 20.30, a delay of one must be applied to match the timing with the bit allocation circuit 44. . The above situation is exactly the same even if the number of bands increases by 2 or more. From the above, a subband encoding method that adaptively allocates bits according to the signal size of multiple bands is a device. There are problems in that the process becomes complicated and there is also a delay in transmitting information regarding bit allocation.

An object of the invention is to provide a subband coding method that uses only i4j and has a small delay.

According to the present invention, there is provided a means for inputting an audio signal and dividing it into a plurality of subbands, and an encoding means including adaptive quantization for inputting a signal of each of the subbands and quantizing it while sequentially updating the step size. and a bit allocation means for determining the number of quantization bits of each band based on information on the step size in each adaptive quantization, and an output of the bit allocation means, and outputs after sorting the outputs of the respective encoding means. an encoding section having means for allocating the signal from the encoding section, means for allocating and outputting the signal from the encoding section so as to correspond to each of the sub-bands based on decoding side bit allocation information, and an output of the allocating means;
A decoding means including adaptive inverse quantization that performs decoding while sequentially updating the step size, a means for outputting the decoding side bit allocation information based on the step size in the adaptive inverse quantization, and outputs of the respective decoding means are combined. Output 1
A subband encoding method is obtained having a decoding section having means for decoding.

The present invention further provides a code including a filter that inputs an audio signal and divides it into a plurality of sub-bands, and an adaptive quantizer that inputs the signal of each sub-band and quantizes it while sequentially updating the step size. a bit allocation means for inputting step size information in each of the inverse quantizations and determining the allocation of quantization bits for each subband by considering it as a short-time signal level of each subband; and an output of the bit allocation means. A subband encoder having a multiplexer that rearranges and outputs the outputs of the respective encoders according to the above is obtained.

The present invention further includes a demultiplexer that inputs a received signal, divides it into a plurality of subbands based on bit allocation information, and outputs the divided signal; a decoding means including adaptive inverse quantization that performs decoding while updating the size; and a decoding unit that inputs the step size in each necessary inverse quantization and outputs the bit allocation information by regarding it as a short-time signal level of each subband. A subband decoder is obtained having an allocation means and a means for combining the outputs of the respective decoding means to produce a reproduced audio signal.

Embodiments of the present invention will be explained in detail with reference to the drawings.

FIG. 2 shows a first embodiment according to the present invention. Also, the third
The figure shows an example of adaptive quantization 21, 22,...2N.
The figure shows an embodiment of adaptive inverse quantizers 61, 62, . . . , 6N.

A signal sampled at 8KI-iz input from terminal 1 is divided into N subbands by filters 1, 12, . . . IN, and resampled according to the bandwidth.

The method of band division can be determined by referring to the aforementioned paper by Mr. Ota. Here, N=4, the interval is set as i, and each band is sampled at 42 kHz. The band-divided signals are input to adaptive quantizers 21, 22, . . . 2N. One of them, the f-signal 100, is quantized in an adaptive quantizer 21. The signal 100 is adjusted in step size Δ- by the gain adjustment circuit 210 of FIG.
Multiplied by the reciprocal of 1/△j. J represents zumbling time. Δj operates according to an algorithm described later, and the gain adjustment circuit 210 works as an AGO. The output of the gain adjustment circuit 210 is applied to a quantization circuit 211, where it is quantized, encoded as a quantization level number l, and outputted as a signal 103. The signal 103 is also input to the adaptation circuit 212 .

The number of quantization bits in the quantization circuit 211 is assumed to be 6 bits. The adaptation circuit 212 receives only the upper two bits of the six bits and updates the step size according to the following equation.

Δj+1−(Δj)β・M(l)−(1) Here, β is a positive value smaller than 1, to 1. (1) is a value around 1 determined by l. This is D, J Goodman
A paper (“IA ro
“bustadaptive quantizer”, I
EEETransactions on Commu
nications, Vol, 00M 23, pp.
1362-1365.1975).

To put it simply, the result of qi and childization is inside (closer to 0)
When the level t of
(1) is made larger than 1. If you do that, △j＋□
changes to a larger value when a large signal is input, and to a smaller value when a small signal is input. Also, β is a constant introduced to make you forget the past state,
The influence of bit errors during transmission can be gradually reduced. Therefore, gain adjustment circuit 210 can act as an AGO. The output signal 101 of the adaptive quantizer 21 is input to the adaptive bit allocation circuit 45. The signal 103 is also input to the multiplexer 41. Signals in other bands are also input to the adaptive S: child generators 22...2N, and the same operations as described above are performed. The adaptive bit allocation circuit 45 assumes that the step size output from each adaptive quantization circuit 21, 22, . decide. However, the maximum allocated bits is 6 bits, and the minimum is 2 bits.
Bit. Each band is sampled at 2KHz (500μSeC period), 5 times, 32kB
To achieve a transmission rate of ps, 16 bits per 500μ formula can be used. Therefore, all channels are 16 bits. The reason why the minimum number of allocated bits is set to 2 is to enable updating of the step size at all times. Bit allocation information is input to multiplexer 41. multiplexer 4
1 rearranges and outputs the output signals of the adaptive quantizers 21, 22, . . . 2N.

The situation is shown in FIG. al, bi, C1, di
are the quantization results of each subband, and the one with smaller 1 is the upper bit. Here, the bits f used for updating the step size are arranged for four channels toward the beginning, and the lower bits of each channel are arranged as shown in FIG. 5 based on bit allocation information. This makes the device a little simpler during decoding. Of course, the bits to be transmitted in one band may be grouped together. The signals rearranged by the multiplexer 41 in this manner are outputted from the terminal 2. In the decoder, the received signal is inputted from the terminal 3 and inputted to the demultiplexer 51. The demultiplexer 51 distributes the received signal to each band according to the bit allocation information from the bit allocation circuit 55. Bit allocation circuit 5
5 inputs the step size from each adaptive quantizer 61, 62, . . . 6N, and performs the same operation as the bit allocation circuit 45 of the encoder. As described later, each step size matches that on the encoder side, so the received signal is correctly distributed to each band. The output of the demultiplexer 51 is input to each adaptive inverse quantizer 61, 62, . . . 6N. In adaptive dequantizer 61, signal 200 is applied to dequantizer circuit 610 and adaptive circuit 612. At that time, the adaptive circuit 612
Only the upper 2 bits are input to . Inverse quantization circuit 61
The output of 0 is input to the gain adjustment circuit 611, where it is multiplied by a step size Δ. If there is no transmission path error, the step sizes △・ and △ found by the adaptive circuit 212 are
, matches. If quantization is performed with reasonable precision, the output 203 of the gain adjustment circuit 611 will be the signal 100.
The value is sufficiently close to . Adaptive inverse quantization circuit 62,...
6N) The same processing as in the adaptive inverse quantization circuit 61 is performed here as well.

The outputs of the adaptive inverse quantizers 61.62, . . . 6N are input to frequency conversion filters 81.82, . . . 8N and are frequency shifted to the original frequency band. The outputs of the respective frequency conversion filters 81, 82, .

As described above, when this embodiment is used, there is no need to transmit bit allocation information, no need to measure the signal level for bit allocation, the equipment is simple, and there is no signal delay. A band coding method is obtained.

FIGS. 6 and 7 are diagrams for explaining a second embodiment of the present invention, and are diagrams showing a part of an encoder and a part of a decoder.

In this embodiment, the adaptive quantizer 21 of the first embodiment and)
In place of the reciprocal quantizer 61, the examples shown in FIGS. 6 and 7 will be used. The explanation will be centered on FIGS. 6 and 7. In
The difference between the number 100 and the predicted value which is the output of the predictor 802 is obtained by a subtracter 801 and added to the adaptive sentence generator 800. The adaptive quantizer may be exactly the same as the adaptive quantizer 21 of the first embodiment.
3 and input only the upper 2 bits to the local inverse quantizer 8036. The stiffness may be the same as the adaptive inverse quantizer 61 of the first embodiment, but only 2 bits are added to the input ζ. The sum of the output of the local inverse quantizer 803 and the output of the predictor 802 is calculated by the adder 804 and the predictor 802
As input. Predictor 802 may be a normal predictor.

On the decoder side, the top two of the signals 200. An adder 812 adds the output of an adaptive inverse quantizer 811 that inputs only bits and the output of a predictor 813. All bits of signal 20o are also input to adaptive inverse quantizer 810. These adaptive inverse quantizers may be the same as the adaptive inverse quantizer 62 of the second embodiment. Adder 814 adds the output of adaptive quantizer 810 and the output of predictor 813 and outputs the result as reproduced signal 203. Note that, unlike this embodiment, the same reproduced signal can be obtained by calculating the sum of the output of the adder 812 and the result obtained by dequantizing only the lower bits.

In the second embodiment, each band is subjected to ADPOM, but only some bands may be subjected to ADPOM, and other bands may be subjected to ADPOM as in the first embodiment.

Can be mixed. Moreover, predictive coding other than ADPOM may be used.

As described above, by using the second embodiment of the present invention, a simple subband encoding method for an apparatus using ADPOM% can be realized.

[Brief explanation of drawings]

FIG. 1 is a diagram showing a conventional embodiment, FIG. 2 is a diagram showing a first embodiment according to the present invention, and FIGS. 3 and 4 are diagrams showing adaptive quantization of the first embodiment according to the present invention. FIG. 5 is a diagram showing an example of the output of the multiplexer 41, and FIGS. 6 and 7 are diagrams for explaining the second embodiment according to the present invention, and the reference numerals are used to explain the second embodiment of the present invention. 1.2.3.4 is a terminal, 11.12, IN is a filter,
21.22.2N is an adaptive quantizer, 41 is a multiplexer, 45 is a bit allocation circuit, 51 is a demultiplexer,
52 is an adder, 55 is a bit M circuit, 61.62.6
N is an adaptive inverse childizer, 81.82.8N is a frequency conversion filter, 100.101.103°200.201.2
03, appearance + 仁箆脹yu is a signal, 210 is a gain adjustment circuit, 2
11 is a multiplexing circuit, 212 is an adaptive circuit, 610 is an inverse quantization circuit, 611 is a gain adjustment circuit, 612 is an adaptive circuit, 800 is an adaptive quantizer, 801 is a subtracter, 802 is a predictor, 803 is an adaptive inverse quantizer, 804 is an adder, 81
0.811 is an adaptive inverse quantizer, 812 is an adder, 813
is a predictor, and 814 is an adder. -1 O 3 Flash Gu Sword 7 S Bile 76 Figure O 7 Figure G/'3

Claims (1)

[Claims] 1. Means for inputting an audio signal and dividing it into a plurality of subbands, and adaptive quantization for inputting the signal of each subband and quantizing while sequentially updating the step size. encoding means; bit allocation means for determining the number of quantization bits for each band based on step size information in each adaptive quantization; and rearranging the outputs of the respective encoding means according to the outputs of the bit allocation means. an encoding unit having a means for outputting, a means for allocating and outputting a signal from the encoding unit so as to correspond to each of the sub-bands based on decoding side bit allocation information, and outputting the output of the allocating means sequentially. a decoding means including adaptive inverse quantization that decodes without updating the step size; a means for outputting the decoding side bit allocation information based on the step size in the adaptive inverse quantization; and a combination of the outputs of the respective decoding means. Subband encoding method (2) has a decoding unit having means for outputting, a filter that inputs an audio signal and divides it into a plurality of subband numbers, and inputs the signal of each of the subbands and updates the step size sequentially. An encoder including adaptive quantization that performs quantization while inputting the step size information for each inverse quantization ζ, and considering it as the short-time signal level of each subband, allocates the quantization bits of each subband. A subband encoder comprising a bit allocation means for determining, and a multiplexer for rearranging and outputting the outputs of the respective encoders according to the output of the bit allocation means. 3. A demultiplexer that inputs a received signal, divides it into a plurality of subbands based on bit allocation information, and outputs it; and an adaptive inverter that inputs the signal divided into each subband and decodes it while sequentially updating the step size. a decoding means including quantization; a decoding bit allocation means for inputting a step size in each adaptive inverse quantization and outputting the bit allocation information by regarding it as a short-time signal level of each subband; and an output of each of the decoding means. and a subband decoder comprising means for combining the two to produce a reproduced audio signal.