JP3237244B2 - Calculation method of short-term forecast coefficient - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for calculating short-term prediction coefficients which is preferably applied to, for example, a linear prediction encoder.

[0002]

2. Description of the Related Art In recent years, low bit rates, that is, 4.8
Speech coding at 9.69.6 kbps includes a vector sum excited linear prediction (VSELP).
code excitation linear prediction (CELP: Code)
Excited Linear Prediction) has been proposed.

[0003] The technical content of this VSELP is described in JP-T-Hei 2-50 by Motorola Incorporated.
No. 2135, “Digital Speech Coder with Improved Vector Excitation Source” and “VECTOR SUM EXCITED LIN”
EAR PREDICTION (VSELP): SPEECH CODING AT 8 KBPS: I
ra A.Gerson and Jasiuk: Paper presented at the In
t.Conf.on Acoustics, Speech and Signal Processing-
April 1990 ".

There is a VSELP encoder as a speech encoding apparatus to which the speech encoding method using VSELP is applied. The processing of the VSELP encoder is roughly divided into calculation of a short-term prediction coefficient and calculation of a quantized vector.

[0005] The calculation of the short-term prediction coefficients is a process of extracting a portion corresponding to a formant of a voice. That is,
This is a process in which a plurality of reflection coefficients are obtained from the autocorrelation value obtained from the input voice data, and a plurality of short-term prediction coefficients are obtained based on the plurality of reflection coefficients to predict the formant of the voice.

[0006]

By the way, in order to obtain a plurality of reflection coefficients from the autocorrelation value, a reflection coefficient closest to the reflection coefficient value obtained by calculation is searched from the table of VSELP itself. The above operation is performed by a DSP (Digital Signal Processor).

The above fixed-point DSP calculates, for example, ten reflection coefficients when the prediction order is tenth order.
If the DSP has a fixed decimal point calculation format, the prediction error will gradually decrease during the calculation of the higher-order reflection coefficient. As a result, a calculation error will occur. As a result, an erroneous reflection coefficient will be obtained. In some cases, it made noise.

An object of the present invention is to provide a method of calculating a short-term prediction coefficient that does not use an erroneous reflection coefficient and does not generate abnormal noise.

[0009]

A method of calculating a short-term prediction coefficient according to the present invention comprises the steps of: obtaining a covariance matrix based on input speech data; and sequentially calculating a plurality of reflection coefficients from the covariance matrix in ascending order. A step of calculating a plurality of short-term prediction coefficients from a plurality of reflection coefficients obtained from the above-mentioned step, and a step of calculating a plurality of short-term prediction coefficients from the low-order reflection coefficients in the step of sequentially calculating the plurality of reflection coefficients from a low order When the sum of the autocorrelation of the forward residual and the autocorrelation of the backward residual of a certain order is equal to or less than a predetermined value, the reflection coefficient of the order and the higher order is forcibly set to a value near zero. The above-mentioned problem is solved by the feature of the present invention.

Here, when the sum of the autocorrelation of the forward residual and the autocorrelation of the backward residual becomes equal to or less than a predetermined value, it is not necessary to calculate the reflection coefficient after the order.

Further, the method of calculating short-term prediction coefficients according to the present invention comprises the steps of: obtaining a covariance matrix based on input speech data; and sequentially calculating a plurality of reflection coefficients from the covariance matrix from a low order. Obtaining a multi-order short-term prediction coefficient from the multi-order reflection coefficient obtained from the step, wherein the multi-order reflection coefficient is sequentially calculated from a low order. When the average value of the autocorrelation of the forward residual and the autocorrelation of the backward residual is equal to or smaller than a predetermined value, the reflection coefficient of the order and the higher order is forcibly set to a value near zero. The feature is solved to solve the above problem.

Here, when the average value of the autocorrelation of the forward residual and the autocorrelation of the backward residual becomes equal to or less than a predetermined value, it is not necessary to calculate the reflection coefficient after that order. .

In the method of calculating a short-term prediction coefficient according to the present invention, the reflection coefficient is calculated by a fixed-point operation.

[0014]

According to the present invention, there is provided a method for calculating a short-term prediction coefficient, wherein the sum or the average value of the autocorrelation of the forward residual and the autocorrelation of the backward residual of a certain order is equal to or less than a predetermined value. Since the multi-order reflection coefficient from the covariance matrix obtained based on the audio data is forcibly calculated to be a value close to zero at the order and the higher order, it is sequentially calculated from the lower order, so an incorrect Does not calculate reflection coefficient and does not generate abnormal noise.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a vector sum excited linear prediction (VSELP) which is a linear prediction encoder to which the method for calculating short-term prediction coefficients according to the present invention can be applied.
ediction) An encoder will be described as an embodiment with reference to the drawings.

FIG. 1 shows a VSELP according to an embodiment of the present invention.
3 shows a schematic configuration of a main part (short-term prediction coefficient calculation unit) of the encoder.

The VSELP encoder calculates short-term prediction coefficients using 170 samples of input audio data as one frame. First, in FIG. 1, an input terminal 11 is supplied with a digitally converted audio signal (audio sample data) at a predetermined sampling rate of about 8 kHz, for example.
It is sent to the covariance matrix calculator 12.

The covariance matrix calculator 12 calculates a covariance matrix from the autocorrelation of the audio sample data. The covariance matrix is windowed by the windowing processing unit 13 to extend the bandwidth. The windowed covariance matrix is supplied to the initial correlation calculator 14.

[0019] The initial correlation calculation unit 14, a F ₀ calculator 14a for calculating the autocorrelation F ₀ of the initial forward residuals, B ₀ calculator 14b for calculating the autocorrelation B ₀ of the initial backward residuals
And a C ₀ calculator 14c for calculating the mutual correlation C ₀ between the initial forward residual and the backward residual. The autocorrelation F ₀ , the autocorrelation B _0, and the mutual correlation C ₀ obtained by the initial correlation calculator 14 are supplied to a predetermined value calculator 15.

The predetermined value calculator 15 calculates the autocorrelation F
_A predetermined value that is a fixed threshold is calculated from ₀ and B ₀ . The predetermined value calculated by the predetermined value calculation unit 15 is supplied to the prediction error level determination unit 16.

The prediction error level determination unit 16 calculates an autocorrelation F _j of a forward residual of the prediction order j and an autocorrelation B _j of a backward residual of the prediction order j obtained by a correlation calculation unit 19 described later. The calculated prediction error level is also supplied, and the predetermined value is compared with the value. When the prediction error is equal to or less than the predetermined value, the process branches to the reflection coefficient fixing unit 18. When the prediction error does not decrease to the predetermined value or less, the process branches to the reflection coefficient calculation and quantization unit 17.

The reflection coefficient calculation & quantization unit 17 calculates the reflection coefficient r _j at the prediction order j (1 ≦ j ≦ 10), quantizes it, and records it in a memory unit (not shown). The reflection coefficient r _j obtained by the reflection coefficient calculation & quantization section 17 is supplied to the correlation calculation section 19. Also, j =
In the case of 10, the process branches to the short-term prediction coefficient calculation unit 20.

[0023] The correlation calculating unit 19, and F _j calculator 19a for calculating the autocorrelation F _j of the forward residual of the prediction order j, B _j calculated for calculating the autocorrelation B _j of backward residuals prediction order j And a C _j calculator 19c for calculating a mutual correlation C _j between the forward residual and the backward residual. The reflection coefficient r _j and the autocorrelation F _j−1 of the forward residual in the preceding stage are calculated in the preceding stage. Autocorrelation B _j−1 of the backward residual and the cross-correlation C of the preceding stage
from _j-1, is calculated autocorrelation F _j of the forward residual of the prediction order j, the correlation C _j mutual autocorrelation B _j and prediction order j of the backward residual prediction order j.

After the branch from the prediction error level judging unit 16, the reflection coefficient fixing unit 18 forcibly fixes the reflection coefficient r _j of the order and the higher order to a value near zero, and after that order. Is not calculated. This reflection coefficient r
_j is supplied to the short-term prediction coefficient calculation unit 20 while being repeated until j = 10. This short-term prediction coefficient calculation unit 20
A short-term prediction coefficient is calculated based on the reflection coefficient r _j and output from the output terminal 21.

In the VSELP encoder of this embodiment, a covariance matrix is calculated from the autocorrelation of the audio sample data.
When calculating the reflection coefficient from the covariance matrix, when the sum of the autocorrelation of the forward residual and the autocorrelation of the backward residual of a certain order is equal to or smaller than a predetermined value, the order and the higher order The reflection coefficient at is forced to a value near zero,
The calculation of the reflection coefficient after that order is not performed.

In other words, the reflection coefficient is determined from the autocorrelation of the audio sample data, and the short-term prediction coefficient α _k is determined from the reflection coefficient. First, the meaning of calculating the autocorrelation coefficient will be described.

The audio sample data S (n) has the following (1)
As shown in the equation, it can be estimated by the gain parameter G, the filter coefficient a _k, and the excitation u (n).

[0028]

(Equation 1)

A predictor having a short-term prediction coefficient α _k is
The prediction output S '(n) as shown in the following equation (2) is output.

[0030]

(Equation 2)

At this time, the prediction error e (n) is

[0032]

(Equation 3) Becomes

In the above equations (1) to (3), if α _k = _ak and e (n) = G · u (n), the audio sample data S
(n) can be predicted exactly. Here, if the short-term prediction coefficient α _k is excellent, a predicted output S ′ (n) that well represents the spectral characteristics of the audio signal S (n) can be obtained. To do this, the mean square error E _n of the prediction error e (n),

[0034]

(Equation 4) And then, it is necessary to make the mean square error E _n of the equation (4) to a minimum. Here, S (m) is a voice signal of sample data number m for voice section n. Also, what we want to obtain here is the short-term prediction coefficient α _k for the short-term audio signal, so the audio signal is a finite section.

[0035] (4) When the short-term prediction coefficients αk are given to minimize the mean square error En of _{formula, E n / α i = 0} (i = 0,1, ···, p) ·· (5), the following equation (6) is obtained.

[0036]

(Equation 5) Here, alpha _k ^* is the alpha _k that minimizes the mean square error E _n. (Hereinafter, “*” of α _k ^* is abbreviated as α _k .)

Here, as shown in the following equation (7), φ _n (i, k)
Is defined.

[0038]

(Equation 6)

Then, the above equation (6) becomes the following equation (8).

[0040]

(Equation 7)

Therefore, if φ _n (i, k) (k = 0... P) of the linear equation of the above equation (8) is given, the short-term prediction coefficient α _k is obtained. That is, the calculation of the autocorrelation coefficient is significant.

In the VSELP encoder of this embodiment, as described above, the covariance calculator 12 calculates the covariance matrix from the autocorrelation of the audio sample data. Hereinafter, the principle of the covariance matrix calculation unit 12 will be described.

In solving the linear equation of the above equation (8), the section for calculating φ _n (i, k) and the voice sample data S
The definition of _n (m) gives rise to two methods of linear predictive analysis. One of these is the covariance method. In this covariance method, the interval for calculating the mean square error is fixed, and φ _n (i, k)
This is a way to consider the effect of the calculation.

First, the mean square error E (n) is calculated as

[0045]

(Equation 8)

When φ _n (i, k) is defined as

[0047]

(Equation 9) Becomes

When the suffix of the sum in equation (10) is changed, φ _n
(i, k) is

[0049]

(Equation 10)

Or

[0051]

[Equation 11] Becomes

From the above equations (11) and (12), -p≤m≤N-1
And S _n (m) in the range of
Individuals will be supplemented from the outside.

Here, in the above equation (8), φ _n (i,
Since k) = φ _n (k, i), this is a p × p symmetric matrix. An analysis method based on a method of calculating φ _n (i, k) as in this equation is a covariance method, and the covariance matrix calculation unit 12
Has become the principle.

The solution of the above covariance method will be described below.
First, a lattice filter for time-series linear prediction analysis is usually represented by a cascade having p lattices as shown in FIG. In the linear prediction analysis, the spectrum H (Z) of the signal is modeled by the transfer function A (Z) of all poles as shown in the following equation (13).

[0055]

(Equation 12)

The reflection coefficient K _m has a relationship with the prediction coefficient a _{k according} to the following formula (14) according to the well-known Levinson method.

[0057]

(Equation 13)

In the grid form, the reflection coefficient is usually estimated step by step to minimize the forward residual f _m (n), the backward residual b _m (n), or both.

Therefore, from FIG. 2, the following equation (15) holds. _{f 0 (n) = b 0} (n) = S (n) f m + 1 (n) = f m (n) + k m + 1 b m (n-1) b m + 1 (n) = k m ₊₁ f _m (n) + b _m (n−1) ・ ・ ・ ・ (15)

S (n) is an input signal and e (n) = f
_p (n) is the prediction error. Usually, since the lattice-shaped computes k _m as the square of the functions, the following (16) is established. _{F m (n) = E (} f m 2 (n)) B m (n) = E (b m 2 (n -1)) C m (n) = E (f m (n) b m (n- 1)) (16) Here, E (•) is an expected value.

Also, φ _n (i, k) = E (S (n−k) S (n−
i)), the following equation (17) holds.

[0062]

[Equation 14]

Here, F ₀₀ ^m = F _m (n), B ₀₀ ^m = B _m
(n−1), C ₀₀ ^m = C _m (n), and from the above equation (14), the above equation (17) is converted into the following equation (18).

[0064]

(Equation 15)

In this way, the autocorrelation F of the forward residual, the autocorrelation B of the backward residual, and the mutual correlation C can be sequentially calculated from the covariance matrix to determine the reflection coefficient.

That is, the covariance matrix calculator 12 in FIG. 1 calculates the covariance matrix φ (i, k) from the voice sample data.
To

[0067]

(Equation 16) Calculate as Here, N _A is the number of audio samples (for example, 170 samples), and N _p is the total number of prediction orders (for example, 10 order).

Next, the initial correlation calculator 14 in FIG.
The autocorrelation F ₀ of the forward residual, the autocorrelation B ₀ of the backward residual, and the cross-correlation C _{0 of} both are calculated as in the following equation (20). F ₀ (i, k) = φ (i, k) 0 ≦ i, k ≦ N _p B ₀ (i, k) = φ (i + 1, k + 1) 0 ≦ i, k ≦ N _p C ₀ (i, k) = φ (i, k) 0 ≦ i, k ≦ N _p (20)

Next, the predetermined value calculation section 15 calculates a predetermined value as a threshold from the sum of the autocorrelation of the forward residual and the autocorrelation of the backward residual. This predetermined value is supplied to the prediction error level determination unit 16.

Then, in order to obtain the prediction error level,
It is assumed that the prediction order j = 1. Then, the reflection coefficient calculation & quantization unit 17 calculates the reflection coefficient r _j from the following equation (21).

[0071]

[Equation 17]

The denominator of the equation (21) is the prediction error. This prediction error is obtained from the sum of the autocorrelation of the forward residual and the autocorrelation of the backward residual.

The reflection coefficient r _j obtained by the above equation (21) is
The reflection coefficient is calculated and quantized by the reflection coefficient calculator & quantizer 17.

[0074] Then, the prediction order j is Once you become equal to N _p, comes to an end. Here, the correlation calculator 19 calculates the following
(22) for calculating autocorrelation F _j of the forward residual of the prediction order j with respect to the reflection coefficient of the next stage, the correlation C _j mutual autocorrelation B _j and prediction order j of the backward residual prediction order j from the equation.

[0075]

(Equation 18)

Then, the prediction order j is incremented (+
1) Then, the process returns to the reflection coefficient calculation & quantization unit 17.

In the correlation calculator 19, the values of F _j , B _j and C _j decrease as the prediction order advances to higher orders. In some cases, the calculation error becomes so small as to have a significant effect, and the calculation of the reflection coefficient in the subsequent prediction order is greatly erroneous.

Then, the prediction error level determination section 16 determines whether or not the calculation result of the denominator of the above equation (21) has become smaller than a predetermined value set in advance. Then, the reflection coefficient fixing unit 18 forcibly fixes the reflection coefficient of the order and the higher order to near zero, and stops the calculation of the reflection coefficient after the order. The preset predetermined value is about 1 of the original value of F ₀ (0,0).
/ 1000, which is a sufficiently small value.

In the above description, the prediction error and the predetermined value are obtained from the sum of the autocorrelation of the forward residual and the autocorrelation of the backward residual, but the autocorrelation of the forward residual and the backward It may be obtained from the average value of the autocorrelation of the residual.

FIG. 3 is a flow chart showing the flow of operation up to the calculation of the reflection coefficient of the VSELP encoder of this embodiment.

First, in step S1, the covariance matrix calculation unit 12 calculates a covariance matrix from the autocorrelation of the voice sample data.

In step S2, the initial correlation calculating section 1
In step 4, the autocorrelation F ₀ of the initial order forward residual, the autocorrelation B ₀ of the initial order backward residual, and the mutual correlation C _{0 of} the initial forward residual and the backward residual are calculated.

In step S3, the prediction order j is set to 1. In step S4, the autocorrelation F ₀ and the autocorrelation B ₀
Then, a predetermined value serving as a threshold value is calculated from.

In step S5, it is determined whether or not the prediction error obtained by the sum or the average value is smaller than the predetermined value obtained in step S4. Here, YES
Is determined, the process proceeds to step S11. If NO is determined, the process proceeds to step S6.

In step S6, the reflection coefficient calculation & quantization section 17 calculates the reflection coefficient r _j .

In step S7, the reflection coefficient calculation and quantization section 17 quantizes the reflection coefficient r _j .

In step S8, it is determined whether or not the predicted order j has reached 10. Here, if YES is determined, this flowchart ends, and if NO is determined, the process proceeds to step S9.

In step S 9, the correlation calculator 19 calculates the autocorrelation F _j of the forward residual, the autocorrelation B _j of the backward residual, and
The mutual correlation C _j between the forward residual and the backward residual is calculated.

In step S10, the predicted order is incremented, and the process returns to step S5.

In step S11, in response to the determination result in step S5, a quantized reflection coefficient near zero is substituted for the reflection coefficients from the j-th order to the tenth order. Here, the quantization near 0 means a value closest to 0 in the table of the quantized reflection coefficients.

As described above, the VSELP encoder of this embodiment calculates the covariance matrix from the autocorrelation of the audio sample data and, when obtaining the reflection coefficient from this covariance matrix, uses the autocorrelation of the forward residual of a certain order with the autocorrelation. When the sum or average value of the autocorrelation of the backward residual becomes equal to or less than a predetermined value, the reflection coefficient at the order and the higher order is forcibly set to a value near zero, and the order after the order Since no reflection coefficient is calculated, no abnormal noise is generated without using an incorrect reflection coefficient.

The method of calculating the short-term prediction coefficients according to the present invention is not limited to the above-described embodiment.
However, the present invention is not limited to those using.

[0093]

The method of calculating the short-term prediction coefficient according to the present invention is applied to the case where the sum or the average value of the autocorrelation of the forward residual and the autocorrelation of the backward residual of a certain order is equal to or smaller than a predetermined value. The multiple-order reflection coefficient from the covariance matrix obtained based on the input voice data is forcibly set to a value close to zero in the order and the higher order, and the reflection coefficient after the order is not calculated. Since the calculation is performed sequentially from the lower order, it is not necessary to use an erroneous reflection coefficient. Therefore, the occurrence of abnormal noise caused by the calculation error can be removed beforehand, and the communication quality can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a schematic configuration of a main part of a VSELP encoder according to an embodiment to which a calculation method of a short-term prediction coefficient according to the present invention is applicable.

FIG. 2 is a diagram showing a lattice filter used for explaining a covariance method.

FIG. 3 is a flowchart illustrating a flow of an operation of a main part of the VSELP encoder according to the embodiment.

[Explanation of symbols]

 12... Covariance matrix calculation section 14... Initial correlation calculation section 15... Predetermined value calculation section 16... Prediction error level determination section 17... Reflection coefficient Calculation & quantization section 18: Reflection coefficient fixing section 19: Correlation calculation section 20: Short-term prediction coefficient calculation section

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-55-65999 (JP, A) JP-A-5-55994 (JP, A) JP-A-1-126700 (JP, A) JP-A-59-65999 171997 (JP, A) IEICE Technical Report Vol. 92 No. 277 (DSP92 67-76) Improvement of speech quality in 8 kb / s LD-CELP speech coding system (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 19/00 G10L 11/00

Claims (5)

(57) [Claims] 1. A step of obtaining a covariance matrix based on input speech data and sequentially calculating a plurality of order reflection coefficients from the lower order from the covariance matrix; Calculating a short-term prediction coefficient having a step of calculating a next-order short-term prediction coefficient. When the sum of the autocorrelation is equal to or less than a predetermined value,
A method for calculating a short-term prediction coefficient, wherein the order and the higher-order reflection coefficients are forcibly set to values near zero. 2. The method according to claim 1, wherein the sum of the autocorrelation of the forward residual and the autocorrelation of the backward residual does not calculate a reflection coefficient of an order lower than or equal to a predetermined value or less. 1. A method for calculating a short-term prediction coefficient according to 1. 3. A step of obtaining a covariance matrix based on the input voice data and sequentially calculating a plurality of order reflection coefficients from the lower order from the covariance matrix; Calculating a short-term prediction coefficient having a step of calculating a next-order short-term prediction coefficient. A method for calculating a short-term prediction coefficient, wherein when an average value of an autocorrelation is equal to or smaller than a predetermined value, reflection coefficients of the order and higher orders are forcibly set to values near zero. 4. The method according to claim 1, wherein the calculation of the average value of the autocorrelation of the forward residual and the autocorrelation of the backward residual, and the calculation of the reflection coefficient after the order smaller than a predetermined value are not performed. 3. The calculation method of the short-term prediction coefficient described in 3. 5. The method according to claim 1, wherein the reflection coefficient is calculated by a fixed-point operation.