JPS58193598A - Voice coding system and apparatus provided therefor - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a coding method for audio signals, particularly a high-efficiency coding method in which the amount of information is 16 hits/second or less.

A subband coding method is known as one of the methods for encoding an audio signal with four efficiency at a bit rate of about 16 bits/second. This method is a method that efficiently performs encoding by utilizing the power bias i in the spectrum of the audio signal.

In other words, an audio signal has a statistical property in which the power of the low frequency component of the spectrum is large and the power of the high frequency component is small. Taking advantage of this property, the frequency band of the audio signal is divided into K multiple subbands (the bandwidth of one subband is often chosen to be 1 degree), and low frequencies with high power are divided into K subbands. Assign the number of quantization bits to the subbands in the domain (for example, 3 to 4 bits), and assign a smaller number of quantization bits to the subbands in the frequency domain (for example, 1 to 2 bits). Perform encoding.

In this way, the number of quantization bits is assigned according to the power of each subband, so it is possible to compress the transmission information #A, and 16
If the bit rate is in the order of bits per second, it is possible to obtain high-quality playback audio.

Details of the subband coding method, Bell Systems Technical Journal (BliL) by Mr. R, E, C piece CHIHRg
L 8Y8TBM8 TgCHNICAL JOURN
AL) magazine.

1977 Tokyo No. 5, 747-770 Makoto's ``On Yuza Dezu In Op T Boo Band Coders for -- Levit Rate To Speech Fist Comienik -- /
117 J (”ON THlg DI!8IGN
OF 8UB-HAND C0DE&31i'OR'L
DN-BI'r LTB aPFflK 01MJNI
This is explained in the paper titled CATI Side (Reference 1.) 41Kl12. Therefore, a brief explanation will be given here.

FIG. 1 is a block diagram of the subband coding method described in the above-mentioned document 1. The unified audio signal x(fi) inputted from the input terminal 100 at An example of the frequency characteristics of the °pass filter (BPI-BPN: BPI-BP4) is shown in 1st Nlb, quoted from the above-mentioned document 1. In Fig. 1(b), the horizontal axis represents the frequency saturation (Hz); The vertical axis is the level (dB)
shows.

1st m la) K'R, the output signal of the bandpass filter 2001-20ON sampler (RE8
AMPLBR) 2051 to 205N. Here, the resampler is 2 times the bandwidth of each bandpass filter.
Perform resampling + (REaAMPLFt) at twice the frequency, then remove the bandpass filter 1 (BPFI:
Let fllb be the bandwidth of component number 200s), sampler 1 (-B-P-!'4-) is component 11
No. 205. ) will perform the reconversion at a frequency of 2 fIHz O. The output signals of renanpla (2051 to 205N) are encoder 1 to encoder N (component vehicle number 2101 to
210, ), quantization and encoding are performed, and 1
Output to Luteprek F-15/220. During quantization, the number of quantization bits is assigned in advance for each expansion. For example, subband IK is 4 bits, ? If subband 2 is assigned 3 bits, and the subband in the low frequency region is assigned a larger number of quantization bits than the subband in the high frequency region, the overall P)M)B can be calculated using the following formula. .

B=2 Σ f,・Flies (bits/second)・・・・・・・
...(1)・-1 The value of B is the bit rate value to which it belongs (for example, 16 bits/
It is necessary to predetermine the number of quantization bits and the bandwidth J so that

Returning to glFIAIJLl again, Multiple Funa 22
0 inputs quantization codes for each subband, combines them to create a code sequence, and outputs it to the output terminal 230.

On the receiving side, a code sequence is manually inputted from an input terminal 240 and outputted to a demultiplexer 250 . Demultigrek t250
separates the quantized codes of N subbands from the input code system and outputs them to decoders 1 to N (constituent vehicle numbers 26G, ~260-). Cocoon symbols 2701 to 270- 27
0M). Here, the band pass filter 270
．． ~270M is bandpass filter 2 on the transmitting side
00. ~ Performs the same operation as 20ON, and performs bandpass replacement on the input signal. band pass filter 270
．． The output signal of ~27G□ is output to the adder circuit 280. An addition path 280 adds these to produce a reproduced audio signal! (
11 is obtained and outputted from the output terminal 300.

Encoder 1 to encoder N (configuration!! Prime numbers 2101 to 210
．． ), the method of quantization in the above-mentioned document IK is to convert the quantization step size Δ, to the quantization code i at j-1 time,
It is determined according to the following formula based on .

j, =M (II,, )・Δ, -8...
α) Here M(1,,) is 1. −, is a multiplier determined by . Please refer to Gyre7 for details of this quantization method.
) (JAYANT) et al. K Pell Systems Technical Journal (B, S, T, J,) 19
Published in the September 1973 issue, 1119-1144 members [7.
word memory” (“As1aptiv* Qllam
tizatian With a 0ntA%brdl
hmory'') (Reference 2.), further explanation will be omitted here.

The subband coding method explained above is a method that reduces the amount of transmitted information by dividing the audio signal into multiple subbands and performing quantization using the uneven power of each subband. However, there was a drawback that the quality of the reproduced audio signal was greatly influenced by the filter %4& for dividing into subbands. In other words, after dividing the audio signal into subbands by a filter, Kt
Re-sampling is performed for each subband at a frequency that is at least twice the width of each band. If this IIIK; y filter f) characteristic is poor and the cut-off characteristic is not good, there is a drawback that aliasing noise and interference noise with adjacent spunds occur. Furthermore, if the characteristics within the pass band are not good, there is a drawback that noise is superimposed on the audio signal within the pass band.

Furthermore, when performing resampling, if the sampling frequency is set to a value sufficiently larger than twice the subband bandwidth, it is possible to prevent noise caused by the cutoff characteristics of the filter to some extent; In this case, the overall bit rate increases and the effect of reducing the amount of transmitted information is small.

Furthermore, when trying to obtain a filter with good characteristics, the amount of calculation increases significantly and the configuration becomes complicated.

An object of the present invention is to provide an audio encoding system and an apparatus for the same, which not only allows high-quality reproduced audio to be obtained with a simple filter configuration, but also reduces the amount of transmitted information compared to conventional systems. be.

According to the present invention, on the transmitting side, a means for inputting a discretized audio signal sequence and extracting a parameter representing the fundamental period of the l14I distribution audio signal sequence, Means for generating a subband signal sequence divided into several subbands with a predetermined bandwidth based on the model excluding the predicted value of the sequence, and a means for generating a subband signal sequence that is divided into several subbands having a predetermined bandwidth, excluding the predicted value of the sequence, means for generating a rattan 10 signal sequence converted to a predetermined temple speed;
- means for generating a j12o signal sequence encoded with the first signal i-yupo; - means for transmitting a combination of the JII2 signal sequence of the JII2 meter and the JII2 signal sequence of each nap band; From signal series 2! l
means for reproducing a 111 subband value code sequence and generating a 30 signal sequence by combining the reproduced subband signal sequences; and a means for generating a predicted value of the audio signal sequence from the transmission signal sequence, and the receiving side generates a parameter representing the fundamental period of the audio signal sequence from the transmission signal sequence and a signal sequence of Recommendation 4 which is a signal sequence encoded for each subband. and a means for separating the
means for regenerating the subband signal sequence for each subband from the signal sequence of 1 and 4, and generating a fifth signal system by combining the reproduced subband signal sequences, based on the parameters; inputting the speech encoding method characterized in that it has a means for generating a predicted value of the speech signal sequence using the IE50 (No. 1 sequence), and calculating the basic period of the speech signal sequence. means for extracting a parameter representing a parameter; means for generating a predetermined value of the audio signal sequence based on the parameter; means for generating a difference signal sequence between the audio signal sequence and the predicted value; means for inputting a No. I signal and generating a subband signal sequence divided into a plurality of subbands having a predetermined bandwidth; means for generating a signal sequence converted to a sampling rate determined by the speed; means for inputting the signal sequence and generating a signal sequence encoded with a predetermined quantization pitch number for each subband; means for transmitting a combination of heavy parameters and the coded signal sequence for each subband; and means for inputting the coded signal sequence for each subband,
G11 of the entire band by combining means for reproducing the starting band signal sequence and the previously distributed and finished subband signal sequence.
Using the above-mentioned all-band signal sequence based on the child actor who generates the signal system and the WM distribution 2 meters.

and means for generating a precognitive value of the audio signal sequence5.
A device is desired which is used for the speech encoding method IX'F, which uses K-K as a nuisance.

Furthermore, according to the present invention, the coded signal sequence is manually processed and the code system of the parameter representing the fundamental period of the audio signal sequence and the signal sequence coded for each subband are separated from the coded signal sequence. A means for manually inputting the encoded signal sequence for each band to be recorded every KII! A means for generating a coded signal sequence, a means for inputting the decoded signal sequence and reproducing a subband signal, and a signal sequence for the entire band by combining the parked subband signal sequence. means for generating a parameter that is manually decoded by manually decoding a code sequence of a parameter representing the fundamental period of the audio signal sequence; It is characterized in that it further includes means for reproducing an audio signal sequence, and means for inputting the reproduced audio signal sequence and generating detailed values of the audio signal sequence based on the parameters decoded by the decoder. A decoding device for the speech encoding method is obtained.

Next one is f! The configuration of the audio encoding method using AK will be explained in detail using one aspect.

FIG. 2 is a block diagram showing an embodiment of a speech encoding system according to the configuration of the present invention. Here, figure J112 (gradient indicates a block diagram of the transmitting side, and figure 1m2 (kll indicates a block diagram of the receiving side).

The audio signal sequence X (S) sampled at the sampling frequency (for example, II K&lz ) is input to the buffer 405 through the input terminal 400 .

A buffer 405 stores an audio signal sequence of a certain length,
This is outputted to the pitch detection circuit 520, and also outputted to the subtracter 410 at each compiling time. Pip detection - path 52
0 is an audio signal sequence for a constant time period, the pitch period M and the pitch gain are calculated by a well-known technique (for example, the autocorrelation method), and these are output to the encoder 630. The encoder 630 quantizes the pitch period M and the pitch gain - using predetermined quantization bits aK, obtains the pitch period M and the pitch gain -, and outputs them to the decoder 9540 and the multi-brevner 450. Decoding @ 540 is ml! ,,
1 is decoded to obtain M' and a', which are output to the preview circuit 510. The subtracter 410 subtracts the output value of the /17a 405 and the output value of the pre-#1 circuit 5100 = (11) to obtain the band pass filter group BPI~1PN (component numbers 420.~420-). Each group of bandpass filters has a predetermined bandwidth I.

Now, assume that each band is 11 to 11. The calculation of the band pass filter may be performed by the well-known 15K calculation, for example, as shown below.

S1=Σh1− − ・・・・・・・・・・・・
(3)・-Iris no breath here 8. (m is bandpass filter 1 (BPI)
The output value value island, (!≦i≦N) is an impulse response value representing the frequency response of the bandpass filter 1 (BPI). Also, N is the order.

111g2tgl4Kjl, the output signal value of the band pass filter #evaluation BPl~BPH is the re-number plar 4~4
30, resampled at twice the frequency of the bandwidth 11-9 of each bandpass 74 router, each code! !
1 to N (output to component numbers 4401 to 440-). Encoders 1 to N are encoders using well-known technology (for example, the above-mentioned JAVA
Quantization and encoding are performed using a predetermined number of quantization bits from K, and the quantization bits are output to the multipreference unit 450, and each is decoded.
~N (component numbers 4701 to 470) are also output. The decoder 1-N inputs -4 to the quantization code 1 of the corresponding nap band, decodes it, and then outputs the bandpass filters 1lipBP1 to BPN (component numbers 4801 to 4), respectively.
8G, ). Here the tension cable consists of 48G, ~4BG
, has the same operation as structure II & element 42G, -420,. Bandpass 74 router (configuration 1! al1 No. 480.~
The output signal value of 480- is output to the adder circuit 49G.
− is obtained. .-) is output to the adder circuit SOO. The adder circuit s00 adds the .- and -711 output values of the extension circuit 615, and outputs the S- result to the pre-Ml- path 510. Input fsin route 51 route 1G and use a series of M' to predict the fundamental periodic component of the audio signal according to the following formula.
141-Do! Output l-.

Q<smtg'*? (1l-(M'-1>)
-------+4) Here, l is the delay time necessary for the calculation from the output of the subtracter 41G to the output of the addition@600e, and the bandpass filter # (420, to 420.
, 68G, ~580. ) depends on the order, for example 2~S
It is extremely low. Further, when (M-I) takes a negative value, the value KN-M may be used instead of M. Here N is 1
It is 1m.

Delay path 515 inputs S- and outputs it to SUM 500, which provides the distance of Itgle time. Here, I is the above (4
It is the same as I written in 2. Multiplex? 450 inputs the quantization codes 11 to jlll and -, arranges them temporally, and outputs them through the transmitting output terminal 460.

Next is Joy 211! (Give an explanation of the receiving side shown to the town.

The demultiplexer tsso inputs the code through the receiving side input terminal 550, separates the quantization code of each subband into ~, and 4h, and sends the quantization code of each subband to the decoders 1 to N(
It outputs to the component numbers 5701 to 57O-, and outputs M and M to the decoder 575. The decoder 575 is
The decoder 540 performs the same operation as the decoder 540 in the
Output to θ□~58G,). Here these bandpass filters have a bandpass of 74 Le/B at the transmitting side.
PI ~ BPN (configuration liI Qiao No. 480. ~ 480
The adder circuit 59 performs the same operation as □) for each subband.
Output to 0. Addition circuit 590 is bandpass filter 5
The output values of 801 to 580M are added to obtain @(b), which is output to the addition circuit 600. The adder 1 circuit 600 adds . Note that the pre-adjustment 1 circuit 610 and the delay circuit 615 operate in the same manner as the pre-adjustment #1 circuit 510 and the continuous kite circuit 515 shown in FIG.

By adopting the configuration of this embodiment, it is possible not only to obtain high-quality reproduced audio even when using a band-pass filter with a simple configuration (for example, a small order), but also to improve the quality of rll'J@ compared to the conventional method. This has the effect of reducing transmission ***.

This will be explained next.

Here, to simplify the explanation, processing for dividing an input signal into subbands, performing encoding and decoding, and reproducing the signal will be expressed as shown in the wk3 diagram.

ji3MIJLl is a simplified representation of the processing on the transmitting side of this embodiment, and corresponds to t&lk in figure j12. In 3Ii14, 650 indicates the encoding and decoding paths, and the encoder (4401 to 440.
) and the decoder (470, to 470.) are shown in -k. The bandpass filter path 660 is w62
Diagram 0 bandpass filter (420, -420, and 5
801 to 5801) in one. The fIIIgl path 67G is a simplified representation of the preorder circuit 510 of FIG. 2 111. In addition, in Fig. 3 IJLI, the pre-i11 circuit 6700 Å force value is the reproduced signal M.
(represents 4. Wisteria 3111kl is a simplified representation of the conventional system's transmission side and reception side. (210,~21G, ) and decoder (to 26~
uo, ) is expressed in one.

Moreover, the band pass film circuit 665 is shown in FIG.
270k) by **.

When the configuration of this embodiment is adopted, the reproduced signal! (tlJ
If you refer to FIG. 3 (&l), Ks is calculated as follows.

? (s) = 'jfnJ+ F (@ (IQ > +
F←rm (tlJ) ・= ... (151 Here, the symbol V(-) represents the output signal when the signal . Now, the characteristic of filter F is 1lil
Suppose that it is not l-like.

In this case, the K(5) equation is changed to the following equation.

i (b) = z@-D (・(b)) 10F (ugly (11) -
・・・・・・−) (However, $ 116) − 曾 (ro) +・(−
) ・・・・・・・・・・・・・・・ (7)(6)
In the formula, D(・1 is the division of the signal φ input to the filter PK by K11
It shows the distortion that occurs due to the deviation KM from the 1m1 filter characteristic.

According to (6) 2, the power Pd of the signal other than the input signal included in the reproduced signal Zh> is D(・(11)'s power Pd
It is expressed as the sum of III and the power P, - of F(m(→)), and can be written as the following equation.

p, wealth p, ,, +p, -,, ・・・・・・・・・・・・
(6) On the other hand, in the case of the conventional subband coding method, the reproduced signal i- can be written as follows with reference to FIG. 3 (b+).

? (-F(g(yl)+F(m(ro))...
・・・・・・・・・ (9) Here, the symbols F(・) and m
轡 has the same meaning as equation (5). When the characteristics of the filter F are not ideal, K is expressed as the first-order equation.

1-x(←D(s(◆F(ntm)・・・・・・・・・
... Elephant time Here, the symbol D (.) has the same meaning as in equation (6). According to Eijin Ni, is it a playback signal? The noise included in (b) can be written as P, like the Oji formula.

P, = P familiarity + P compensation , , , , -, , -----------
−°−°゛If the precolumn index is operated in advance, (
Since the power of (b) is smaller than the power of z-, the following equation holds true.

P91al < PIIIIXI ・・・・・・・・・
・・・・・・・・・・・・・・・・・・ Therefore, according to the configuration of this embodiment, distortion caused by a defect in the filter am included in the reproduced signal is reduced compared to the conventional method. I understand that. Therefore, according to the configuration of this embodiment, even if a band-pass filter with a simple configuration is used, reproduced audio of higher quality can be obtained compared to the conventional system.

Furthermore, if the quality is equivalent to that of the conventional method, there is an effect that the amount of transmitted information can be reduced.

In addition, in this example, the pre-11 circuit 51G and the 6-tightening (
4) Although the first-order side shown in No. 2 was performed, the characteristics will be further improved if the higher-order side is performed. For example, on the tertiary side, B.
EEE TRANSACTIONS ON EIETRAN8ACTIO by Mr. Nis Attar (B, S, MTAL) et al.
"PREDICTIVE CODING OF Speech Signals and Nanojective Errors"("PREDI"), published in the June 1979 issue of N8 ON A, 8jl, P"), pp. 247-254.
CTIVg C0DING 8P] 1ECH5IGNMUL8 AND8υMJWCTIVE ERRORCRI
TERIA') (Reference 3) etc.
It has been stated.

According to the above-mentioned document 3, the tertiary element - may follow the following formula.

i←/@”; (s-(M'-1-I))+a'e?
(s-(M'-1))+rs(s-(M'4-1-
I)) , , , , , , , , , , , Qm General K
In the transmission band of 1,611,147 seconds or less, a method is used in which the quantization noise is given a frequency characteristic to make it audibly easier to hear. This method uses noise shaping (N0
I8E 8HAPING), and it is often done to make the frequency characteristics of the quantization noise close to the frequency characteristics of the audio signal. - As an example, the frequency characteristics of an audio signal and the frequency characteristics of quantization noise (with and without noise shaping) are shown below.

11, the horizontal axis represents frequency (Kllz) and the vertical axis represents level. Also, the solid line represents the frequency characteristics of the audio signal.
The dashed-dotted line and the broken line show the frequency characteristics when no Y-shaping is applied and when it is applied, respectively.

FIG. 5 is a pre-12 diagram showing an example of the second embodiment of the speech encoding system according to the configuration of the present invention. This figure is a combination configuration in which noise shaping is added to the configuration shown in Diagram 2. This is an example. The symbol (&) in the same figure shows the transmitting side, and the symbol 1b) in the same figure shows the receiving side. Tree number 700.
710 is a noise shaping circuit. The 0-line circuit is composed of a filter, and each k of 700.710
(1) Transfer function C4Zl, F (Zl is expressed by the following equation.

Here ~ is the coefficient of the noise shaping filter,
N indicates the order of the filter. In the shaft type, the ratio a
φ may be a predetermined constant, or KL'C may be used to adaptively convert Kf according to the audio signal.

Components in FIG. 5 with the same numbers as in FIG. 2 perform the same operations as in FIG. Not only the effect, but also the audible KJ! There are five effective ways to obtain an improved reproduced audio signal.

1s211e In Fig. 5, DP is used as an encoder.
Of course, other encoding methods such as the CM method may be used.

[Brief explanation of the drawing]

Iris 111 (ILl is an example of the conventional split band bouding method 81!, !1 figure (bl is the frequency a% of the bandpass filter for dividing into nap bands)
Figure 112 shows an example of the audio encoding method according to the present invention. Block diagram illustrating an embodiment of the present invention, No. 38111 Town #l'+ is used to describe the detailed description of the present invention. 4 is a diagram showing an example of noise shaping, and FIG. 5 is a diagram showing an example of noise shaping. In wJ, 20G, ~200..27G, ~270
．． 420. ~420. . 480, ~48ON, 5801~580. −・・・・
・Band pass filter, 21Q, ~S! 1G,, 44
01~44G, --- encoder,! 601-26G,. 57012570. ...Decoder, 205. −”
206. ．． 4301-430.・−・−Resampler,
520...Pitch detection circuit, 410...
・Subtraction circuit, 280.490.500.510.600
...Addition circuit, 510.610.670--
Child a+tii path, 515.615...Delay circuit, 450...Multiplex, 560...Figure 1 (αU(b) Figure 2 (oL) Cb) 3rd Figure (α) (b) Fourth bile

Claims (2)

[Claims] (1) On the transmitting side, the discretized audio signal sequence is manually processed,
means for extracting a parameter representing a fundamental period of the audio signal sequence; and a plurality of subbands having a predetermined band Ii &- after taking a difference between the audio signal sequence and a preconditioning value of the audio signal sequence field. means for generating a subband signal sequence divided into subband signal sequences; means for generating a Jllll signal sequence converted from the subband signal sequence into a number uniformity determined by the subband; and encoding the first signal sequence. did. #! means for generating a signal sequence of 2; means for transmitting a combination of the m distribution parameter and the 2nd signal sequence of each of the subbands; and a means for generating a third signal sequence by combining the reproduced band value system set, and a preconditioning value of the audio signal sequence using the arranged 30 signal sequence based on the SIl period parameter. On the receiving side, a parameter representing the basic period of the audio signal sequence from the transmission signal sequence and a JI signal that is a KII-coded signal system signal for each nap band are provided.
means for separating the signal sequence ζ of II4, reproducing each subband signal sequence from the signal sequence II4, and reproducing the subband signal sequence ζ to be reproduced! and a means for generating a signal sequence of 1s5 by combining the above parameters, and a means for generating a preliminary adjustment value of the voice signal sequence using the signal sequence of the pre-arrangement 5 based on the parameters. A speech encoding method characterized by: (2) Means for inputting a discretized audio signal sequence and extracting a parameter representing the basic period of the audio signal sequence, and means for generating a preconditioning value of the Zengji phonetic code sequence based on the parameter. a means for generating a differential signal sequence between the audio signal sequence and the preconditioning value; and a subband signal sequence divided into a plurality of subbands each having a predetermined bandwidth by λ of the differential signal sequence. means for generating a koji sequence from the IMI Ki-tpupando-ban-go sequence; and generating a signal sequence in which the signal sequence is input and encoded with a predetermined number of quantization bits for each nanoband. means for transmitting a combination of the parameters and the coded signal sequence for each subband; and a means for transmitting a combination of the parameters and the encoded signal sequence for each subband, and generating a signal sequence for the entire band by combining the reproduced subband signal sequence (( 3) means for manually inputting the coded signal sequence and separating from the coded signal sequence into a puff meter code sequence representing the fundamental period of the audio signal system and a signal system sequence coded for each subband; , means for manually generating a signal sequence coded for each subband and generating a decoded signal sequence for each subband; and means for inputting the decoded signal sequence and reproducing the subband signal. - Means for generating a signal sequence of the entire band from K by combining the reproduced Spupando signal sequence, and inputting and decoding a puff meter code sequence representing the fundamental period of the 17M audio value consideration sequence. means for generating a converted puff meter, means for reproducing an audio signal sequence by combining the signal sequence of the entire band and the preset value of the audio signal sequence, and inputting the reproduced audio signal sequence and the decoded parameter. A decoding device for use in a speech encoding method, characterized in that the decoding device includes means for generating a preliminary value of the speech signal sequence based on the speech signal sequence.