JPS62209499A - Encoding/decoding system and apparatus - 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 (Field of Industrial Application) The present invention relates to band compression of audio signals, and particularly to digital transmission and storage.

(Prior Art) In order to efficiently encode audio or image signals, it is effective to utilize the redundancy of signals, especially correlation.

Typical examples are linear predictive coding and orthogonal transform coding.

The limit of the compression rate of these encoding methods is determined by the amount of correlation that the signal has, that is, the determinant of the covariance matrix. Paper by Nitori: [Linear Transform Coding and Linear Predictive Coding], Electronic Communication Academic journal, vol, 53-A, PP, 97-10
4 (1970). According to the theory described therein, highly correlated signals can be encoded more efficiently (the smaller the determinant of the covariance matrix is).

(Problems to be solved by the invention) Conventional high-efficiency speech encoding methods have achieved band compression by relying only on the correlation of input speech signals, and there has been a limit to coding efficiency. .

An object of the present invention is to improve the encoding efficiency in waveform encoding by encoding a cross-correlation function between an input audio signal and a signal that is known or determined depending on the characteristics of the input audio signal. be.

(Means for Solving the Problem) According to the present invention, when encoding a series of input discrete signal sequences, the input signal sequence and a signal sequence that is fixed or adaptively determined based on the signal are Coding/decoding characterized in that a cross-correlation function is determined, the cross-correlation function sequence is encoded, the encoded cross-correlation function is decoded, and a reproduced signal sequence is generated from the decoded cross-correlation function. method is obtained.

Further, according to the present invention, means for inputting a series of discrete signal sequences, means for determining a cross-correlation function between the input signal sequence and a signal sequence that is fixed or adaptively determined according to the signal, and the cross-correlation function and means for outputting a code sequence representing the cross-correlation function.

Furthermore, according to the invention, means for inputting a code sequence;
A decoding device is obtained, comprising means for decoding a cross-correlation function from the code sequence and means for generating a reproduced signal sequence from the decoded cross-correlation function.

(Operation) The input audio signal divided into blocks of N samples is x
(n), n=1. ..., N. This x(n) and
A signal h(n) that is determined in advance or adaptively to x(n)
) and the cross-correlation function is calculated as follows.

! (1;)=Σx(n+t)h(n)
(1) The two transformations of n-1ρ x(n) are expressed as X(z), and the two transformations of h(n) are expressed as H(
z), then the two transformations A(z) of the cross-correlation function type (m)
is V(z)=X(z) H(z-1)
(2), and the electric waste vector is: PW(Z) = V(z) V(z) =P)((z) PH(z)
(3) However, Px(z)=X(z) X(z-1) PH(Z)=H(z) H(z-1) As shown in n) and the electric waste vector. On the other hand, according to the theory in Reference 1, the compression limit of x(n) is determined depending on the i-square mean of the electric debris vectors of x(n), that is, the bias of the electric debris vectors. Therefore, as can be seen from equation (3) above, h(n
) is taken appropriately, the cross-correlation function y(n) between x(n) and h(n) can become a signal with a higher correlation than x(n) (i.e., the geometric mean of the electric scum vector is smaller). sexual,
It can be expected to obtain high encoding efficiency. However, h(n
) is determined by adapting it to x(n), it is necessary to encode information representing h(n) as auxiliary information. To generate the reproduced signal x(n) from the cross-correlation function, as can be seen from equation (2), the spectrum of the cross-correlation function is
A simple method is to divide the spectrum by the spectrum and then convert it to a time signal. A diagram of the principle of the present invention is shown in FIG. In the figure, 1. is the input signal x(n) and the defined signal h(n)
2o is 1. To encode the cross-correlation function t17(m) obtained in 3.
．． is 2. Decoding the cross-correlation function encoded in 4. to obtain the cross-correlation function tp(T,). Indicates that the basic constituent elements are to obtain the reproduced signal 'u(n) from the decoded cross-correlation function.

This concludes the explanation regarding the principle of the present invention.

(Embodiment) FIG. 2 is a block diagram of an Ondokari encoding/decoding device showing a first embodiment of the present invention. In the figure, 10 is an input terminal, and 1 is sampled at regular intervals, for example, at 8 kHz.
A 60-sample discrete audio signal x(n) is input, and x(n) is supplied to the cross-correlation function calculator 11. A cross-correlation function calculator 11 inputs a predetermined signal sequence h(n) from 12 memories, calculates a cross-correlation function type CC) between x(n) and h(n), and performs linear prediction on it. Output to encoder 20. Inside the linear predictive encoder 20, the memory 12
The signal sequence stored in is a signal whose electric cassette vector is equal to the long-term average spectrum of the audio signal. The linear predictive encoder 20 is a linear predictive encoder having a so-called backward type predictor, 21 is a buffer memory that stores a cross-correlation function of 160 samples, and 22 is a cross-correlation function W(n) generated by an adaptive prediction circuit 25. 23 is the prediction residual e
A quantizer 24 quantizes the quantized prediction residual δ(n) and a prediction 6(n). Various algorithms are known for the adaptive prediction 525, but here we will use Gibson's algorithm.
) paper [Sequentially Adaptive Backward Prediction in ADPM Speech Coders]
ive Backward Prediction
inADPCM 5peech Coders IEEETrans, on Comm
PP, 145-150 (19
78) This will be explained using the method described in (Reference 3). The quantized prediction error δ(n) is converted into a code sequence by a 17 encoder and then transmitted to the receiving side from a 19 output terminal. On the receiving side, first, the code sequence received from 40 is decoded into a quantized prediction error δ(n) by a decoder 44, and output to the linear prediction decoder 30. At 30, δ
(n) and the predicted value amount n) obtained from the adaptive prediction circuit 31 are added together using 35 adder circuits to obtain a cross-correlation function distribution (n).
is created and output to the buffer memory 45. Conversion circuit 46
is the reproduced signal Q using the method already explained using the signal h(n) having the same characteristics as the transmitting side stored in the memo 1 and the cross-correlation function W(n) input from the buffer memory 45. (n) Generate and output to output terminal 49. The operation of the adaptive prediction circuit 31 is the same as that of the adaptive prediction circuit 25 on the transmission side.

3(a) and 3(b) are block diagrams of a speech encoding/decoding apparatus showing a second embodiment of the present invention, and FIG.
) shows the configuration on the encoding device side, and FIG. 3(b) shows the configuration on the decoding device side. In Fig. 3(a), input terminal 1
00, a discrete audio signal of 160 samples sampled at, for example, 8 kHz at regular time intervals x(n) n=1
．． ....

160, are input, 110 linear predictive analyzers and 1
50 cross-correlation calculators. Linear prediction analyzer 1
10 is, for example, an eighth-order vocal tract cross-sectional area function k(i), i=1.10, corresponding to the input signal x(n). ..., 8. is calculated and output to the linear prediction parameter similarity comparator 120.

The linear prediction parameter similarity comparator 120 uses, for example, 1024 vocal tract cross-sectional area function patterns kj(i), which are stored in advance in the linear prediction parameter codebook 130.
(i=1. . . , 8°j=1. . . , 1024) and k(i) input from 110 are compared for similarity.

If this is the standard pattern with the highest degree of similarity or the first one, then 120 is kj(i).

The pattern (i=1..., 8) is sent to the 140 impulse response meter Kh, and the sign j of the pattern is sent to the multiplexer 1.
Output to 90. Impulse response calculator 140 includes 13
The impulse response h(n) of the linear prediction filter is calculated from the vocal tract cross-sectional area function supplied from 0 and output to the cross-correlation function calculator 150. 150 calculates the 160th-order cross-correlation function tP(n) between the audio signal x(n) input from 100 and the impulse response h(n) input from 140, and outputs it to the buffer memory 160. The buffer memory 160 divides the 160th-order cross-correlation function into four sub-blocks every 40 samples, and the normalization circuit 1
Output to 61.

The cross-correlation function input to the normalization circuit 161 is normalized so that its square norm becomes 1, and is output to the cross-correlation function similarity comparator 170. The normalization coefficients σs, (8"1)..., 4) at this time are encoded using the normalization coefficient codebook 162 by 165 normalization coefficient similarity comparators. Normalization coefficient codebook For example, out of 256 standard patterns, σs, (""+...,
The code S of the pattern most similar to the pattern 4) is output to the multiplexer 190. The cross-correlation function similarity comparator 170 uses, for example, 1024 cross-correlation functions vi(n) stored in the cross-correlation function codebook 180.
s(i=ts...

1024, n=1. ..., 40) standard pattern and 16
The t-th subblock ψ(n+40
・Calculate the similarity with t) and (t: 0, 1, 2, 3). If the standard pattern with the highest degree of similarity with the th sub-block is the 1st standard pattern, the cross-correlation function codebook 180 has the code 1(t) of the selected standard pattern, (t=
0.1, 2, 3) are output to the multiplexer 190.

The multiplexer 190 is divided into four mutual phases f! 'l + 5' Code 1(t) representing the number J, code j representing the linear prediction parameter, and code S representing the normalization coefficient
are multiplexed and transmitted from the output terminal 101 to the decoding side. Next, the operation on the decoding device side will be explained with reference to FIG. 3(b). The demultiplexer 260 inputs the code sequence through the reception terminal 201, separates it into a code representing the cross-correlation function, a code representing the vocal tract cross-sectional area function, and a code representing the normalization coefficient. t), (
t=0.1, 2.3) in the cross-correlation function codebook 25
0, the code j representing the vocal tract cross-sectional area function is output to the vocal tract cross-sectional area function codebook 240, and the code S representing the normalization coefficient is output to the normalization coefficient codebook 251.

In the cross-correlation function codebook 250, the cross-correlation function corresponding to the received code 1(t) (t=0.1, 2, 3.4) is searched and outputted to the correction circuit 255. In the normalization coefficient codebook 251, the code S is used to search the codebook for a pattern of normalization coefficients corresponding to the pattern, and output to the correction circuit 255. The correction circuit 255 has 25
The normalized cross-correlation function input from 0 is multiplied by the normalization coefficient input from 251 to obtain an equivalent correlation function, which is output to the prediction residual calculator 230. Similarly, in the linear prediction parameter codebook 240, a pattern of the vocal tract cross-sectional area function corresponding to the code j supplied from 260 is searched,
It is output to the linear prediction coefficient calculator 220. The linear prediction coefficient calculator 220 converts the vocal tract cross-sectional area function supplied from 240 into a linear prediction coefficient, and outputs it to the impulse response calculator 221 and reproduction signal generator 210. Impulse response calculator 221 calculates the impulse response of the linear prediction filter based on the linear prediction coefficients received from 220 and outputs it to prediction residual calculator 230 . The prediction residual calculator 230 solves the following simultaneous equations from the cross-correlation function Φ(n) supplied from the correction circuit 255 and the impulse response h(n) supplied from the impulse response calculator 221, and calculates the residual. Calculate e(n).

1ω Σφ(n, n+m) e(m)=Ko(n), n=1.
..., 160. (4) cl However, +Nu The determined prediction residual is output to the reproduction signal generator 210. In the reproduced signal generator 210, a linear prediction coefficient calculator 22
Linear prediction coefficient input from 0 and prediction residual calculator 230
A reproduced signal x(n) is generated from the prediction residual inputted from the output terminal 200 and outputted from the output terminal 200.

Although the embodiments described so far are based on linear predictive coding, orthogonal transform coding can also be used as cross-correlation function type coding.

This concludes the description of the embodiments of the present invention.

(Effects of the Invention) The present invention does not directly encode an input audio signal, but converts the input audio signal into a cross-correlation function between the input signal and a signal that is known or adapted to the properties of the input audio signal. It is characterized in that it performs encoding. Since the cross-correlation function obtained in this way can have a higher correlation than that of the input audio signal, the configuration of the invention that selects the cross-correlation function as the target signal for encoding has a higher encoding efficiency than the conventional one. It has the effect of realizing the following.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the principle of the present invention, FIG. 2 is a block diagram showing the first embodiment of the invention, and FIGS.
b) is a block diagram showing a second embodiment of the invention.

Claims (3)

[Claims] (1) When encoding a series of input discrete signal sequences, find a cross-correlation function between the input signal sequence and a signal sequence that is fixed or adaptively determined according to the signal, and encode the cross-correlation function sequence. An encoding/decoding method comprising: decoding the encoded cross-correlation function, and generating a reproduced signal sequence from the decoded cross-correlation function. (2) means for inputting a series of discrete signal sequences; means for determining a cross-correlation function between the input signal sequence and a signal sequence that is fixed or adaptively determined according to the signal; and means for encoding the cross-correlation function. and means for outputting a code sequence representing the cross-correlation function. (3) A decoding device comprising means for inputting a code sequence, means for decoding a cross-correlation function from the code sequence, and means for generating a reproduced signal sequence from the decoded cross-correlation function. .