JP3031765B2 - Code-excited linear predictive coding - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a code-excited linear predictive coding method, and can be applied as a high-quality compression coding method for, for example, audio signals.

[0002]

Code Excited Linear Prediction encoding method of the Prior Art Conventionally, for example, has been shown in the literature 1. Reference 1: 『1989, P
roc.IEEE Int.Conf.on Acoustics, Speech and Signal P
rocessing, PP.65-pp.68, NSJayant and JHChen, "Spe
ech Coding with Time-VaryingBit Allocations to Exc
FIG. 2 is disclosed in Reference 1 .
FIG. 3 is a functional block diagram of an example of a code-excited linear prediction encoder that summarizes what is performed.

In FIG. 2, an input original speech vector S is
It is supplied to the analysis circuit 102, where the PARCO (PARti
al auto-CORrelation (partial autocorrelation) analysis, and supplies the quantized short-term prediction coefficient αjp to the multiplexing circuit 111,
The vocal tract prediction coefficient αjq is supplied to the short-time filter 103.

Next, the switches 108 and 113 are opened, and the adaptive excitation codebook 104 in this state stores the adaptive excitation code vector ea (i) (i = 1 to n) in the supplied adaptive excitation code index Ia. Selective output based on The gain control circuit 119 calculates an original speech vector S,
The adaptive excitation gain control code βap is generated from the adaptive excitation code vector ea (i) and supplied to the multiplexing circuit 111.
The gain control circuit 119, adaptive excitation gain control co
Supplies over de βap to the multiplier 122, the adaptive excitation code vector βea (i) and adaptive excitation gain control code βap and allowed to multiply <br/> calculated by and controls the gain, the adaptive excitation code vector after multiplication βea: (i) Ru is supplied to the adder 110.

The adder 110 supplies the adaptive excitation code vector βea (i) as it is to the short-time filter 103 as the excitation vector e (i), and the short-time filter 103 supplies the synthesized speech vector Sw to the subtractor 109. I do. The subtracter 109 performs subtraction on a component basis between the original speech vector S and the synthesized speech vector Sw, and supplies the error vector er (i) to the auditory filter 105. The auditory filter 105 performs filtering and converts the vector ew (i) into the auditory error calculation circuit 1
06. Hearing error calculation circuit 106 calculates a root mean gi of each component unit vector ew (i), the square
I that minimizes the average gi is the optimal adaptive excitation code vector
As the index Ia Le, adaptive excitation codebook 1
04 and the multiplexing circuit 111.

Next, the switch 113 is closed, and the stochastic excitation gain control code βsq generated by the gain control circuit 123 is applied to the stochastic excitation code vector es (i) output from the stochastic excitation codebook 107. To the multiplier 126
To control the gain, and supplies the multiplied stochastic excitation code vector βes (i) to the adder 110. At this time,
The gain control circuit 123 also supplies the stochastic excitation gain control code βsp to the multiplexing circuit 111. The adder 110 adds the probabilistic excitation code vector βes (i) and the adaptive excitation code vector βea (i) , and supplies the excitation vector e (i) to the short-time filter 103 as described above .
A subtracter 109 in the above and similar processing methods, the auditory filter 105, and treated with auditory error calculation circuit 106, the optimal
Seeking index Is of stochastic excitation code vector, a multiplexing circuit 111, stochastic excitation codebook 10
7

[0007] Next, by closing the switch 108 and 113, and updates the contents of the adaptive excitation codebook 104 by the excitation vector e (i).

[0008] The probabilistic excitation code index Is and the adaptive excitation code index Ia obtained as described above.
The stochastic excitation gain control code βsp the adaptive excitation gain control code βap quantization short prediction coefficients αjp and, multiplexing
The circuit 111 multiplexes and outputs the total code C.

[0009] In the code-excited linear prediction decoder,
In other words, by performing processing symmetrical to the above processing,
Issue.

[0010]

However, in the above conventional method, the PARC for the input voice signal is not used.
When the number of quantization bits in the OR analysis is small, there is a problem that distortion or the like is generated in the reproduced synthesized voice, and that a satisfactory S / N cannot be obtained, such as a sufficient S / N cannot be obtained. Therefore, the number of quantization bits cannot be reduced, which has been an obstacle to reducing the transmission bit rate.

[0011] For P ARCOR analysis, in order to reduce the information amount, extending the frame interval for example to 20 msec,
In the meantime, in order to smoothly change the parameters, linear interpolation is performed, for example, every 2.5 msec. However, this linear interpolation does not always adapt to the PARCOR coefficient, and therefore, the distortion of the reproduced synthesized speech spectrum increases, and the sound quality is degraded. This is PARC
It is also considered that the OR coefficient does not correspond to a single definite physical quantity and is a complex compound parameter.

Further, since the gain for the adaptive excitation codebook and the gain for the stochastic excitation codebook are independently controlled by coefficients that are scalar-quantized, when the number of quantization bits is reduced, distortion and the like are generated, and the reproduced image is synthesized. Since the S / N of the sound cannot be sufficiently obtained, the number of quantization bits cannot be reduced, and as a result, there has been an obstacle to lowering the transmission bit rate.

Since the adaptive excitation gain control code and the stochastic excitation gain control code are independently generated and transmitted,
There is a tendency that the transmission bit rate tends to be high, and if the number of excitation source codebooks increases, the number of gain control codes increases, and the transmission bit rate must be increased.

The demand for such an encoding system at present is that the quality of synthesized speech reproduced at a low transmission bit rate can be obtained well, but it has been difficult to realize the above problem due to the above problems.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and has as its object to provide a code-excited linear predictive coding method capable of obtaining a sufficient synthesized speech quality even at a low transmission bit rate. It is to provide.

[0016]

SUMMARY OF THE INVENTION The present invention, in order to achieve the above object, against the code excited linear predictive coding scheme using a plurality of types of excitation source information, introducing the following characteristic units and methods did.

That is , a line spectrum pair analysis is performed on a speech signal to generate line spectrum pair analysis information used for generating a synthesized voice, and the line spectrum pair analysis information is generated .
The code corresponding to the information is one element of the multiplexed coded audio signal
A line spectrum pair analysis means which is a, the plurality of types
Of which stores a plurality of sets of gain control information set comprising a plurality of gain control information for each of the excitation source information, its
Index that specifies any pair of
The gain control codebook, which is an element of the audio signal, and the output from the gain control codebook
Each type of excitation source information after gain control is added.
The gain for the calculated excitation information
From the additional excitation source information.
Gain control means for performing gain control, performing gain control on the plurality of types of excitation source information by a plurality of gain control information in an optimal gain control information set,
Adds and adds multiple types of excitation source information with gain control
Gain control is also performed on the excitation source information,
The audio signal is encoded and decoded using the added excitation source information and the line spectrum pair analysis information.

[0018]

According to the present invention, since the line spectrum pair analysis is performed on the input voice signal, the line spectrum pair corresponding to the main formant has a small deviation from the formant frequency and corresponds relatively well. Therefore, the linear interpolation characteristics are good, the spectral distortion due to the linear interpolation is small, and the deterioration of the sound quality is reduced. Then, even if the coefficients of the line spectrum pair are quantized to a low bit number, sound quality can be improved as compared with the conventional PARCOR analysis.

Further, there is no need to transmit gain information for each of a plurality of types of excitation source information. That is, since the respective gain control information for each excitation source information is formed as a set , it is only necessary to transmit an index designating this set, so that the transmission bit rate can be reduced as compared with the related art. Moreover, by setting the combination of the gain control information to the optimum combination of the gain control information in advance, it is possible to efficiently obtain good sound quality.

Furthermore, a plurality of types of excitations with gain control
Source information is added, and gain control is also performed on the added excitation source information.
Therefore, improvement in the sound quality of the reproduced sound can be expected.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of the code excitation linear predictive coding system according to the present invention will be described below with reference to the drawings.

[0022] This embodiment is a code excited linear predictive coding of forward type and row Uco over de-excited linear prediction encoder that decoder.

[0023] The object of this example is to realize a code excited linear predictive coder is also obtained good synthetic speech quality at low bit rates the decoder.

In order to realize this object, in this embodiment, the vocal tract prediction parameters are mainly set to LSP (Line Sp).
(Electrum Pair) parameters and a circuit for performing quantization, and LSP parameters are inversely quantized and interpolated.
A circuit for converting to a vocal tract prediction parameter, a VQ (vector quantization) gain codebook in which the gains of the adaptive excitation codebook and the stochastic excitation codebook are vector-quantized, and a backward-type excitation gain prediction circuit are provided. I have.

FIG. 1 is a functional block diagram of a code excitation linear prediction encoder according to this embodiment. FIG. 3 shows a functional block diagram of the code-excited linear prediction decoder of this embodiment.

In FIG. 1, the code-excited linear predictive encoder according to this embodiment includes a vocal tract analysis circuit 301, an LSP quantizer 302, an LSP inverse quantizer 303, a multiplexing circuit 304, and an adaptive excitation. Codebook 305, VQ gain codebook 306, stochastic excitation codebook 307
, A multiplier 308, an adder 309, and a multiplier 310
, A gain control circuit 311, a multiplier 312, a synthesis filter 313, a subtractor 314, and an auditory filter 315.
And a hearing error calculation circuit 316.

[0027] In here, the vocal tract analysis circuit 301, the LSP quantizer 302, LSP dequantizer 303 forms an LSP analysis system.

Next , the code-excited linear prediction code shown in FIG.
Illustrating an operation of the encoder.

The frame units summarized original speech vector S is supplied to the vocal tract analysis circuit 301, it is required vocal tract predictive parameter aj. This vocal tract prediction parameter aj
Is supplied to the LSP quantizer 302, and the LSP quantizer
302 converts the vocal tract prediction parameter aj into an LSP parameter and quantizes the LSP code, and converts the LSP code Iw into an LS
It supplied to P inverse quantizer 303 and the multiplexing circuit 304. L
The SP inverse quantizer 303 calculates the LSP from the LSP code Iw.
After the parameters are inversely quantized and interpolated, they are inversely transformed into vocal tract prediction parameters aqj and supplied to the synthesis filter 313.

On the other hand, the adaptive excitation codebook 305 converts the adaptive excitation code vector eai (i = 1 to n) into the multiplier 30.
8 Also, the stochastic excitation codebook 307
Supplies the stochastic excitation code vector esl (l = 1 to m) to the multiplier 310. Further , the VQ gain codebook 306 outputs a pair of excitation gains gak and gsk (k = 1 to p). The excitation gain gak is supplied to the multiplier 308.
The excitation gain gsk is supplied to the multiplier 310.

[0031] Accordingly, the multiplier 308 obtains a vector eaik supplied to the adder 309 by multiplying the adaptive excitation code vector eai the excitation gain GAK, multiplier 310
Is the stochastic excitation code vector esl and the excitation gain g
The vector eslk is obtained by multiplying by sk and supplied to the adder 309.

The adder 309 performs addition of the component unit vector eaik and vector Eslk, supplied to the multiplier 312 obtains an excitation vector e. Gain control circuit 311
Calculates the excitation gain σ from the past optimal excitation vector and supplies it to the multiplier 312. The multiplier 312 multiplies the excitation vector e by the excitation gain σ to obtain a vector eg and supplies the vector eg to the synthesis filter 313.

The synthesis filter 313 filters the vector eg using the supplied vocal tract prediction parameter aqj as, for example, a tap coefficient of the filter, obtains a synthesized speech vector Sw, and supplies the resultant to the subtractor 314. Subtractor 314
Performs subtraction between the original speech vector S synthesized speech vector Sw component unit, and supplies the auditory filter 315 seeking error vector er. Auditory filter 315 performs filtering on the error vector er, filtering
The resulting vector ew is supplied to the hearing error calculation circuit 316.

[0034] Hearing error calculation circuit 316, a vector ew
The root mean square of each component of
The combination of j and k is converted to the optimal adaptive excitation code vector
Index Ia, the optimum stochastic excitation codevector
Index Is, as indenyl <br/> box Ig optimal excitation gain code, and supplies them to the multiplexing circuit 304 supplies the adaptive excitation code index Ia to the adaptive excitation codebook 305, stochastic excitation code index Supplying Is to the stochastic excitation codebook 307,
The excitation gain code index Ig is supplied to the VQ gain codebook 306.

Next, the adaptive excitation codebook 305 is updated.
The method will be described. The adaptive excitation codebook 305 outputs an optimal adaptive excitation code vector eao according to the adaptive excitation code index Ia and supplies the output to the multiplier 308.
Further, the stochastic excitation codebook 305 outputs an optimal probabilistic excitation code vector eso according to the stochastic excitation code index Is and supplies it to the multiplier 310. Also,
The VQ gain codebook 306 outputs an optimum pair of the excitation gains gao and gso according to the excitation gain code index Ig. One excitation gain gao is supplied to the multiplier 308, and the other excitation gain gso is supplied to the multiplier 310. Sa
Re that.

The multiplier 308 multiplies the optimal adaptive excitation code vector eao and the optimal excitation gain gao by multiplication.
The subsequent vector eao · gao is supplied to the adder 309. Also, the multiplier 310 performs multiplication of the optimum stochastic excitation code vector eso and optimal excitation gain gso, base after multiplication
The vector eso · gso is supplied to the adder 309. Adder 3
09 adds these vectors eao · gao and eso · gso, and adds the added value e opt to the adaptive excitation codebook 3.
It is supplied to the 05, Ru to update the content.

[0037] In addition, each codebook 305,306,3
There are various methods for searching the optimum code index according to 07, but there is no particular limitation. For example, the ea1 the adaptive excitation code vector, in case of a es1 stochastic excitation code vector, searches by changing the excitation gain (ga1, gsl) ~ (gap , gsp) in the order of, optimal adaptive excitation code vector And the stochastic excitation code vector and the excitation gain.

The multiplexing circuit 304 multiplexes the LSP code Iw, the adaptive excitation code index Ia, the stochastic excitation code index Is, and the excitation gain code index Ig obtained as described above, and sums them. Output as code C and supply to code-excited linear prediction decoder.

FIG. 3 shows a functional block diagram of the code-excited linear prediction decoder of this embodiment.

[0040] In FIG. 3, the code excited linear predictive decoder of this embodiment includes a demultiplexing circuit 402, an adaptive excitation codebook 403, VQ gain codebook 404
, A stochastic excitation codebook 405 and a multiplier 406
, A multiplier 407, an adder 408, a gain control circuit 409, a multiplier 410, and an LSP dequantizer 411.
And a synthesis filter 412.

Next , the code-excited linear predictive decoding shown in FIG.
Illustrating an operation of the encoder.

The total code C that is input, in the demultiplexer 402 is separated into each code, LSP code Iw is supplied to the LSP dequantizer 411, the adaptive excitation code index Ia adaptive excitation codebook 403
, And the probabilistic excitation code index Is is supplied to the probabilistic excitation codebook 405, and the excitation gain code index Ig is supplied to the VQ gain codebook 404.

The LSP inverse quantizer 411 inversely quantizes the LSP parameters from the LSP code Iw , further performs interpolation, converts the LSP parameters into vocal tract prediction parameters aj, and supplies the parameters to the synthesis filter 412.

The adaptive excitation codebook 403 supplies the adaptive excitation code vector ea corresponding to the adaptive excitation code index Ia to the multiplier 406. Further, the stochastic excitation codebook 405 supplies the stochastic excitation code vector es corresponding to the stochastic excitation code index Is to the multiplier 407. The VQ gain codebook 404 is
Excitation gains ga and gs corresponding to the excitation gain code index Ig are output.
6 and the excitation gain gs is supplied to the multiplier 407.
Ru is.

The multiplier 406 calculates a vector eag obtained by multiplying the adaptive excitation code vector ea by the excitation gain ga.
Is supplied to the adder 408. Further, the multiplier 407 supplies a vector esg obtained by multiplying the stochastic excitation code vector es and the excitation gain gs to the adder 408. The adder 408 adds the vector eag and the vector esg ,
The obtained vector e is supplied to the multiplier 410. Also,
The vector e is also supplied to the adaptive excitation codebook 403, and the content is updated by the vector e.

The gain control circuit 409 is used to determine the past optimum excitation
The excitation gain σ is obtained from the vector and supplied to the multiplier 410. A multiplier 410 multiplies the excitation vector e by an excitation gain σ to obtain a vector eg, and obtains a synthesis filter 412
To supply.

The synthesis filter 412 filters the vector eg using the supplied vocal tract prediction parameter aj as, for example, a tap coefficient of the filter, and obtains and outputs a synthesized speech vector S.

According to the above-described embodiment, the encoder side
And LSP parameter code Iw by LSP analysis
And the LSP code Iw is
Since the vocal tract prediction parameter aj is obtained by SP inverse quantization and the synthesized speech is reproduced by the synthesis filter, the conventional PARC even if the number of quantization bits of the LSP parameter is reduced.
Compared with the analysis by the OR method, the S / N of the reproduced synthesized speech
Can be improved.

[0049] In addition, the optimal excitation gain by the excitation gain code index Ig in the VQ gain codebook g
Since a pair of ao and gso is output, the adaptive excitation code vector is multiplied by the optimal excitation gain gao, and the stochastic excitation code vector is multiplied by the optimal excitation gain gso to encode and decode. Only the optimum excitation gain code index Ig needs to be performed. Conventionally, two systems of gain code indexes have been transmitted. Therefore, sufficient voice quality can be obtained even at a low transmission bit rate.

In the above embodiment, the forward type code-excited linear predictive encoder and decoder have been described. However, the present invention is not limited to this configuration. The present invention can be applied to a decoder.

The VQ gain codebook stores two types of excitation gains as a pair. However, the present invention is not limited to this, and a plurality of excitation gains are stored as a set according to the number of types of excitation codebooks to be controlled. The configuration may be as follows.

[0052]

As described above, according to the present invention, a line spectrum pair analysis means, a gain control codebook ,
With gain control means to optimize the gain control information set
Excitation by multiple gain control information
Gain control is performed on the source information, and the gain-controlled
Add several types of excitation source information, and add
Also performs gain control, and performs encoding and decoding of the audio signal using the gain-controlled addition excitation source information and the line spectrum pair analysis information. Synthesized speech quality can be obtained.

[Brief description of the drawings]

1 is a functional block diagram of a code excited linear prediction encoder according to actual施例.

FIG. 2 is a functional block diagram of a code excitation linear prediction encoder according to a conventional example.

3 is a functional block diagram of a code excited linear predictive decoding apparatus according to the actual施例.

[Explanation of symbols]

301 vocal tract analysis circuit, 302 LSP quantizer, 30
3, 411 ... LSP inverse quantizer, 304 ... multiplexing circuit,
305: adaptive excitation codebook, 306: VQ gain codebook, 307: stochastic excitation codebook, 30
8, 310, 312, 406, 407, 410 multiplier, 309, 408 adder, 311, 409 gain control circuit, 313, 412 synthesis filter, 314 subtractor, 315 audio filter, 316 audio Error calculation circuit, 402: demultiplexing circuit.

Continuation of front page (72) Inventor Kenichiro Hosoda 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) References JP-A-3-243998 (JP, A) JP-A-4- 73699 (JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) G10L 19/12 H03M 7/30 JICST file (JOIS)

Claims (1)

(57) [Claims] 1. A plurality of types of code excited linear predictive coding method using excitation source information, performs a line spectrum pair analysis for the audio signal, to generate a line spectrum pair analysis information utilized to produce synthesized speech, the Code corresponding to line spectrum vs. analysis information
A plurality of sets of line spectrum pair analysis means, which is one element of the multiplexed coded audio signal, and a plurality of sets of gain control information including a plurality of pieces of gain control information for each of the plurality of types of excitation source information are stored.
And the index that defines one of the pairs is
And gain control codebook that is the one component of the multiplexed encoded audio signal, the output from the gain control codebook, respectively
Of multiple types of excitation source after gain control
The gain for the added excitation source information
The above-mentioned additional excitation source information is formed from
And a gain control means for down control, the optimal gain control information by a plurality of gain control information in the set performs gain control on said plurality of types of excitation source information, a plurality of types of excitation source information gain control Add
Gain control for the added excitation source information,
A code excitation linear predictive encoding method for encoding and decoding a speech signal using the in-controlled addition excitation source information and the line spectrum pair analysis information.