JP3275247B2 - Audio encoding / decoding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is applied to code-driven linear predictive coding and vector-sum driven linear predictive coding using a noise codebook, and is a highly efficient speech for digitally encoding a speech signal sequence with a small amount of information. The present invention relates to an encoding method and a decoding method thereof.

[0002]

2. Description of the Related Art At present, as a method for encoding speech with high efficiency, an original speech is divided into sections called frames, which have a constant interval of about 5 to 50 ms, and the speech of one frame is divided into an envelope shape of a frequency spectrum and It has been proposed to separate the information into two pieces of information, that is, a drive excitation signal for driving a linear filter corresponding to the envelope shape, and encode each of them. In that case, as a method of encoding the drive excitation signal,
There has been known a method of separating and encoding a drive excitation signal into a periodic component considered to correspond to a fundamental frequency (pitch cycle) of a voice and another component (in other words, an aperiodic component). Code-Excited Linear Prediction Cod (Code-Excited Linear Prediction Cod)
ing: CELP) and vector sum-driven linear prediction coding (Vecto
r Sum Excited Linear Prodiction Coding (VSELP) method. About each technology, MRSchroeder and
BSAtal: "Code-ExcitedLinear Prediction (CELP): H
igh-quality Speech at Very Low Bit Rates ", Proc.IC
ASSP'85, 25.1.1, pp. 937-940, 1985, and IAGerson
and MAJasiuk: "Vector Sum Excited Linear Predict
ion (VSELP) Speech Coding at 8 kbps ", Proc. ICASS
P'90, S9.3, pp. 461-464, 1990.

In these encoding methods, as shown in FIG. 9, a parameter representing an envelope shape of a frequency spectrum is calculated in a voice analysis unit 12 for an original voice input to an input terminal 11. Usually, a linear prediction method is used for this analysis. The linear prediction parameter is encoded by a linear prediction parameter encoding unit 13, the encoded output is branched, and decoded by a linear prediction parameter decoding unit 14, and the decoded linear prediction parameter is converted to a linear prediction synthesis filter. It is set as 15 filter coefficients.

In the adaptive codebook 16, the immediately preceding past excitation vector is cut out at a length corresponding to a certain period (pitch period), and the cut-out vector is repeated until the length of the frame becomes equal to the frame length. Is output. Noise codebook 1
7, 18 output time-series code vector candidates corresponding to the non-periodic components of the voice. The noise codebooks 17 and 18 are usually based on white Gaussian noise as shown in FIG.
Various code vectors of the length of the frame are stored in advance independently of the input speech.

The time series vector candidates from the adaptive codebook 16 and the noise codebooks 17 and 18 are weighted by an adder 19 in multipliers 21 ₁ , 21 ₂ and 21 ₃ , respectively.
₁ , g ₂ , and g ₃ are multiplied, and the output of these multiplications is
It is added by two. This added output is supplied to the linear prediction synthesis filter 15 as a drive excitation vector, and the synthesis filter 15 outputs a synthesized (reproduced) voice. The distortion of the synthesized speech with respect to the original speech from the input terminal 11 is calculated by the distance calculation unit 23, and the codebook search unit 2 is operated in accordance with the calculation result.
4, a candidate having a different cutout length in the adaptive codebook 16 is selected, another code vector is selected from the noise codebooks 17 and 18, and the weights g ₁ , g ₂ , and g of the weighted addition unit 19 are further selected. ₃ is changed so that the distortion calculated by the distance calculation unit 23 is minimized. A periodic code indicating the cut-out length of the adaptive codebook 16 when the distortion is minimized, a noise code indicating each code vector of the noise codebooks 17 and 18, and a weight code indicating weights g ₁ , g ₂ and g _3. , Linear prediction parameter codes are output as encoded outputs and transmitted or stored.

[0006] In the decoding, as shown in FIG.
6 and the prediction parameters are set as filter coefficients in the linear prediction synthesis filter 27. A time-series code vector obtained by cutting out the previous driving excitation vector from the adaptive codebook 28 in the cycle using the immediately preceding past driving excitation vector obtained up to that time and the input periodic code, and repeating this for the number of frames. Are output, and the code vector indicated by the input noise code is a random codebook 29, 31.
Are read out as time-series vectors. These time-series vectors are weighted in accordance with the weighting code input by the weighted addition unit 32, and then added, and the added output is supplied to the synthesis filter 27 as a drive excitation vector, and reproduced from the synthesis filter 27. Voice is obtained.

The random codebooks 29 and 31 are the same as the random codebooks 17 and 18 used for encoding. Only one or more noise codebooks may be used. In code-driven linear predictive coding, all code vectors to be candidates are directly stored in a random codebook. That is, if the number of code vectors to be candidates is N, the number of code vectors stored in the noise codebook is also N.

[0008] In the vector sum excited linear predictive coding, the noise codebook, as shown in FIG. 12, (referred to as a basic vector) all code vectors stored are read out simultaneously, multiplying unit 33 ₁ ~ 33 _M Are multiplied by +1 or −1 by the noise codebook decoder 34, and the multiplied outputs are added and output as an output code vector. Therefore, the number of output code vectors becomes 2 ^M by a combination of +1 and −1 by which each basic vector is multiplied, and one of the 2 ^M output code vectors is selected so as to minimize distortion.

The weight in the weighted adder 19 is determined logically optimally at the time of periodic search (when searching the adaptive codebook 16) and at the time of code vector search (when searching the noise codebooks 17 and 18). There is a method of performing scalar quantization and a method of having a codebook for weight and performing a search on the codebook to determine a codebook with minimum distortion.

[0010]

However, in these conventional methods, since the periodicity of the driving sound source signal is limited to only the components of the previous frame, the expressiveness of the periodicity is weak, and the reproduced sound is rough. It had the disadvantage of lacking smoothness. An object of the present invention is to provide a method for enhancing the expressiveness of the periodicity of a driving sound source, which has been conventionally expressed only with a component related to the previous frame, and more smoothly and accurately expressing reproduced sound.

[0011]

According to the present invention, in order to enhance the expressiveness of the periodicity of speech, a part or all of the code vector output from the noise codebook which has not had the periodicity conventionally, Alternatively, a part of a component of the output code vector or a part of a plurality of noise codebooks has the same periodicity as the periodicity of the output time-series code vector of the adaptive codebook.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First, the case where the present invention is applied to a coding unit of CELP (Code Driven Linear Predictive Coding) will be described. The overall configuration is the same as that of FIG. 9. As in the conventional case, first, using the adaptive codebook, a basic period search (which is considered to correspond to the pitch period) of the target frame from the driving excitation vector of the previous frame and a search based on it Create a driving sound source periodic component.
Subsequently, a search for the random codebook is performed. In the present invention, the code vector of the random codebook is made periodic. That is, as shown in FIG. 1, one code vector is cut out from the noise codebook 17 by a length 36 of the basic period L obtained by the basic period search. This extraction is performed by a method of extracting a length L from the beginning of the code vector to the rear, and a method of extracting the length L from the end to the front.
There is a method of cutting out the image, and the figure shows the method of cutting out from the beginning. As shown in FIG.
6 is repeatedly arranged until the frame length is reached,
A periodic code vector is created and used as an output code vector. This is performed for all the code vectors in the noise codebook 17, and among those, the one that minimizes the distance between the reproduced sound passed through the synthesis filter and the original sound is defined as the optimum code vector. Subsequent determination of the weight of each driving sound source component is performed in the same manner as in the related art.

When a plurality of random codebooks are used, some of the random codebooks are provided with periodicity by the method shown in FIG. 1, code vectors are output, and the remaining random codebooks are output. May be used as non-periodic. FIG. 2 shows an example of this configuration. In this case, the random codebook 17 outputs a code vector with periodicity, and the random codebook 18 outputs a non-periodic code vector. This makes it possible to arbitrarily set the degree of freedom between the periodic component and the non-periodic component in the driving excitation by appropriately allocating the number of noise codebooks to be periodicized and the number of noise codebooks not to be periodicized. Can be approached.

Further, in the coding section of the CELP, only a part of the code vectors in one noise codebook may be made periodic by the above-described method, and the rest may be made non-periodic. For example, as shown in FIG. 3, reference numeral vectors 1 to N _S in the noise codebook 17 outputs each remembering periodicity,
The other code vectors N _S +1 to N are output as non-periodic. According to this configuration, as the drive excitation signal for a certain frame, which of the periodicized code vector and the non-periodicized code vector is suitable is determined by the same codebook search method as the conventional method, It can be automatically determined at the same time as the code vector search. In other words, it is possible to change the degree of freedom between the periodic component and the non-periodic component for each frame and make it closer to the optimum.

The periodicity shown in FIGS. 1 and 2 can be similarly applied to a noise codebook in VSELP (Vector Sum Driven Linear Predictive Coding). Next, an example in which the present invention is applied to an encoding method based on VSELP will be described. As shown in FIG. 4, a predetermined one of the M basic vectors is output in a periodic manner by the above-described method, and the others are output in a non-periodic manner. Each of the non-periodic code vectors is multiplied by +1 or -1, and then added to obtain an output code vector. Multiplying section 33 _to 333
_The code change for _M is performed in the same manner as in the prior art, and the output code vector is optimized. In this way, only a part of the basic vectors in the random codebook are given a periodicity, and the rest are made aperiodic as they are, whereby the ratio of the number of the periodic basic vector to the aperiodic basic vector, That is, the degree of freedom between the periodic component and the aperiodic component of the driving sound source vector can be set arbitrarily, and the ratio can be made closer to the optimum. This ratio is set in advance.

According to this method, after the search for the optimal code vector, the periodic component of the code vector (weighted addition of only the periodicized basic vector by the VSELP method) and the non-periodic component (also the periodic Is weighted and added only by the VSELP method). So, for example, FIG.
As shown in (1), in the weight coding of each drive excitation component after the search for the optimal code vector, the periodic component and the aperiodic component in one output code vector can be given different weights. Given periodicity basic vector 1 to M _S, are added in the addition portion 37 multiplied by the ± 1 to, for the remaining basis vectors M _S + 1 to M multiplied by the a orゝ± aperiodic Are added by an adder 38, and the outputs of the adders 37 and 38 are respectively multiplied by a multiplier 21 ₂ ′,
The weights g ₂ ′ and g ₂ ″ are multiplied by 21 ₂ ″ and supplied to the addition unit 22. In this case, the optimum code vector is determined first, and thereafter, the optimum weight is searched for for each of the periodic component and the non-periodic component in the code vector by the above-described method. In this way, by changing the component ratio between the periodic component and the aperiodic component of the driving sound source vector for each frame,
It is possible to set an optimum value for the frame.

As shown in FIG. 6, for example, each of four consisting of codevectors by-noise codebook 39 _1, 39 ₂ each one of the selected respective basis vectors each code vector, a portion, in this example the sub the output of the noise codebook 39 ₁ to periodic, others aperiodic orゝcity multiplies ± 1 each of these may be output code vector by adding the two basis vectors.

In the configuration of FIG. 6, as shown in FIG. 7, only a part of the code vectors in each of the sub-noise codebooks 39 ₁ and 39 ₂ may be made periodic. In FIG. 7, each of two code vectors in the four code vectors is periodic.
In the above description, the present invention has been described with respect to encoding, but it is assumed that decoding is the same as the random codebook in encoding.

[0019]

As described above, according to the present invention,
Since the noise code vector in the driving sound source signal is made periodic, the reproduced sound becomes smooth. In that case, the ratio between the periodic component and the non-periodic component of the driving sound source signal can be set arbitrarily, and the ratio can be made closer to the optimum. In addition, by terminating some code vectors of one noise codebook, the degree of freedom can be changed for each frame. Furthermore, a different weight can be assigned to each component of the period and the non-period for each frame, and the weight ratio can be optimized by searching the weight codebook.

FIG. 8 shows an example of the sound quality improvement effect in the case of voice coding of about 4 kbit / s. FIG. 8A shows one VSELP-type noise codebook of M _S basic vectors subjected to the periodic processing and one VSELP-type noise codebook of (12−M _S ) basic vectors not subjected to the periodic processing. SNR and segmental SNR when each is used. FIG. 8B is FIG. 4 where the number M of basic vectors is 12
Using one of the noise codebook and an SNR and segmental SNR when treated inner M _S number of periods of the basic vectors. According to these, the present invention provides 4 kbit / s
The conventional method (M _S =
0), the quantization noise can be reduced by about 1 dB, and it can be seen that the present invention can improve the synthesized speech quality. Judging from hearing, it is particularly preferable that M _S = 9 or 10.

[Brief description of the drawings]

FIG. 1 is a diagram showing a code vector periodizing unit in a CELP noise codebook according to a first embodiment of the present invention.

FIG. 2 is a diagram illustrating a codebook and a codebook search unit according to a second embodiment of the present invention when a plurality of noise codebooks are used and all code vectors in some codebooks are subjected to a periodic process; FIG.

FIG. 3 is a diagram showing a codebook in a case where a periodic process is performed on a part of code vectors in one noise codebook in the third embodiment of the present invention.

FIG. 4 is a diagram showing a code vector periodicization unit in a VSELP noise codebook according to a fourth embodiment of the present invention.

FIG. 5 is a diagram illustrating a noise codebook, a noise codebook search unit, and a driving excitation weight search unit in a case where a periodic component and an aperiodic component of a VSELP noise codebook are separately weighted according to a fifth embodiment of the present invention; FIG.

FIG. 6 shows a sixth embodiment of the present invention, in which each code vector of the main noise codebook is configured as a linear combination of a plurality of CELP-format auxiliary noise codebook outputs, The figure which shows the example at the time of making all code vectors in the codebook of a part periodic.

FIG. 7 shows a seventh embodiment of the present invention, in which each code vector of the main noise codebook is configured as a linear combination of a plurality of CELP-format subnoise codebook outputs, and a part of each subnoise codebook. The figure which shows the example at the time of making the code vector of periodical.

FIG. 8 is a diagram of an SNR and a segmental SNR showing the effect of the present invention.

FIG. 9 is a block diagram showing a general configuration of a linear prediction encoding device.

FIG. 10 is a diagram showing a noise codebook in CELP.

FIG. 11 is a block diagram showing a general configuration of a decoding device for linear predictive coding.

FIG. 12 is a diagram showing a noise codebook in VSELP.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-2-84699 (JP, A) JP-A-63-37724 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 19/00-19/14 H03M 7/30 H04B 14/04

Claims (6)

(57) [Claims] 1. An output from an adaptive codebook for each frame.
Code vector and the output code vector from the random codebook
In a speech encoding method that uses a driving filter obtained by adding and driving a synthesis filter to reproduce speech, a pitch period is obtained for each frame, and a past driving sound source vector is cut based on the pitch period.
Output from the adaptive codebook as an output code vector, and the code vector of the noise codebook is cycled repeatedly for each of the pitch periods, so that distortion of the reproduced sound output from the synthesis filter with respect to the input sound is minimized. that searches over SL noise codebook output sign-vector, the noise code book has been encoded with the search indicating the pitch period
And a code indicating the code vector of the speech signal. 2. A method according to claim 1, wherein a plurality of random code books are stored on a frame basis .
Fetch the code vector, and fetch each code vector
Multiplied by a combination of +1 or -1
Force code vector and output code vector from adaptive codebook
In the speech coding method utilizing the driving of the synthesis filter with the drive vector obtained by weighting and adding the weights to reproduce the speech , a pitch cycle is obtained for each frame , and the past drive sound source vector is cut based on the pitch cycle.
Output from the adaptive codebook as an output code vector, and at least the code vector extracted from the noise codebook.
Is also partly cycled for each of the pitch periods, and is used as the input sound of the reproduced sound output from the synthesis filter.
The combination of +1 and -1 that minimizes the distortion
And the code indicating the combination and the pitch cycle
Speech encoder Goka method and outputting a code indicating. 3. The noise codebook according to claim 2,FromOf the above periodic
Linear between code vector and non-periodic code vector
Combine the above noise codebookOutput fromSignal vector
2. The method according to claim 1, whereinOrSpeech code described in 2Transformation
Law. Wherein the said periodic been code not vector and periodic code vector from the noise code book by the weight ratio between constant heavy Mitsuki adds the output sign-vector of the noise codebook, the noise code On the book sign vector
Search for the code vector of the random codebook that minimizes the recording distortion ,
3. The speech encoding method according to claim 1, wherein said weight ratio is changed with respect to a code vector constituting said code vector having a minimum distortion to determine a weight ratio having a minimum distortion. (5) Output code from adaptive codebook for each frame
Weighting the output vector from the random codebook
Drive the synthesis filter with the added drive vector
In a voice decoding method for reproducing voice, Past driving sound source vector based on the input pitch period
And output code vector from the above adaptive codebook
age, The code vector from the noise codebook is
The output code from the above random codebook
Signal vector A speech decoding method characterized by the above-mentioned. 6. The periodic code of the random codebook
Linear combination of vector and non-periodic code vector
Output from the random codebook
The speech decoding method according to claim 5, wherein