JP2001134298A - Speech encoding device and speech decoding device, and speech encoding/decoding system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the amount of computations during a code search on an encoding side while obtaining high quality synthesized speech on a decoding side in a speech encoding device, a speech decoding device and a speech encoding/decoding system which employ a pulse spreading code table. SOLUTION: The configuration of a pulse spreading code table A 101 of a speech encoding device is made different from that of a pulse spreading code table B 111 of a speech decoding device. A spreading pattern that is obtained by replacing components with zero for every one sample is stored in a spreading pattern storage section A 102 of the encoding side to conduct code search computations. Fixed waveforms frequently included in the noise source target obtained through learning, for example, are stored as spreading patterns in a spreading pattern storage section B 112 of a decoding side to conduct decoding. Thus, an encoding device in which the amount of computations is reduced in the encoding side and a decoding device capable of obtaining high quality synthesized speech on a decoding side are obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech encoding device, a speech decoding device, and a speech encoding / decoding system for efficiently encoding / decoding speech information.

[0002]

2. Description of the Related Art As a low bit rate speech coding apparatus and speech decoding apparatus, many coding apparatuses and decoding apparatuses based on the CELP system have been developed. As a representative document on the CELP system, for example, “Code Excit
ed Linear Prediction: "HighQuality Speech at Low B
it Rate "", MR Schroeder et al, Proc. ICASSP'8
5, pp. 937-940 (Reference 1). CE
In an encoding / decoding device of the LP system, input speech is divided into fixed time intervals (this time interval may be hereinafter referred to as a frame), and an audio signal is encoded / decoded for each frame. Become

First, an outline of a CELP speech coding apparatus will be described with reference to FIG. In the CELP speech coding apparatus in FIG. 6, first, the linear prediction analysis unit 12 performs a linear prediction analysis on the input speech 11 to calculate a linear prediction coefficient,
The calculated linear prediction coefficient is output to the linear prediction coefficient encoding unit 13. Next, the linear prediction coefficient encoding unit 13 encodes the linear prediction coefficient (vector quantization), and outputs a quantization index (hereinafter, referred to as a linear prediction code) obtained by the vector quantization to the code output unit 24 and the linear prediction coefficient. Output to the predictive code decoding unit 14. Next, the linear prediction code decoding unit 14
The linear prediction code obtained by the linear prediction coefficient coding unit 13 is decoded (dequantized) and output to the synthesis filter 15. The synthesis filter 15 forms a synthesis filter of an all-pole model having, as coefficients, a decoded linear prediction code obtained by decoding by the linear prediction code decoding unit 14. Then, a vector obtained by multiplying the adaptive excitation vector selected from the adaptive codebook 17 by the adaptive excitation gain 19 and a vector obtained by multiplying the noise excitation vector selected from the noise codebook 18 by the noise excitation gain 20 are vectorized. The addition is performed by the addition unit 22 to generate a driving sound source vector. Then, the distortion calculator 16 calculates a distortion (Equation 1) between the output vector when the synthesis filter 15 is driven by the driving sound source vector and the input speech 11, and outputs the distortion ER to the code specifying unit 23.

[0004]

(Equation 1)

In Equation (1), u is an input speech vector in a processing frame, H is an impulse response matrix of a synthesis filter, ga is an adaptive excitation gain, gc is a noise excitation gain, p is an adaptive excitation vector, and c is an adaptive excitation vector. This is a noise source vector.

Here, the adaptive codebook 17 is a buffer (dynamic memory) storing driving excitation vectors for the past several frames, and the adaptive excitation vector selected from the adaptive codebook 17 synthesizes input speech. It is used to represent a periodic component in a linear prediction residual vector obtained through an inverse filter of the filter.

On the other hand, the noise codebook 18 is a static memory for storing a plurality of types of fixed vectors, and the noise excitation vector selected from the noise codebook 18 is newly added to the linear prediction residual vector in the current processing frame. Is used to represent the non-periodic component (the component obtained by removing the periodicity (adaptive excitation vector component) from the linear prediction residual vector) added to. Note that the random codebook stores a plurality of types of fixed vectors, and is therefore sometimes called a fixed codebook. And
Adaptive excitation vector weighting section 19 and noise excitation vector weighting section 20 read from excitation codebook 21 the adaptive excitation vector selected from adaptive codebook 17 and the noise excitation vector selected from noise codebook 18. It has a function of multiplying the adaptive sound source gain and the noise sound source gain. The weight codebook 21 is a static memory storing a plurality of types of sets of an adaptive excitation gain multiplied by an adaptive excitation vector and a noise excitation gain multiplied by a noise excitation vector.

The code specifying unit 23 optimizes the index of the above three codebooks (adaptive codebook, noise codebook, and weighted codebook) to minimize the distortion ER of (Equation 1) calculated by the distortion calculator 16. Select Then, the distortion specifying unit 23
Outputs the index of each codebook selected when the distortion is minimized to the code output unit 24 as an adaptive excitation code, a noise excitation code, and a weight code, respectively. And finally, the code output unit 24 outputs the linear prediction coefficient coding unit 13
And the adaptive excitation code, the noise excitation code, and the weighting code specified by the code specifying unit 23 are all collected into a code (bit information) representing the input speech in the current processing frame, and decoded. Output to the conversion unit. In addition,
An adaptive excitation code, a noise excitation code,
The specification of the weight code may be performed after dividing a frame at a fixed time interval into shorter time intervals called subframes. However, in this specification, the following description will be given without distinction between a frame and a subframe (after unifying them as frames).

Next, an outline of the CELP speech decoding apparatus will be described.
This will be described with reference to FIG. In the CELP decoding device of FIG. 7, first, the code input unit 31 receives a code (bit information for code-representing a speech signal in a frame section) specified by the CELP speech coding device (FIG. 6), and receives the code. The resulting code is decomposed into four types of codes: a linear prediction code, an adaptive excitation code, a noise excitation code, and a weight code. Then, it outputs the linear prediction code to the linear prediction coefficient decoding unit 32, the adaptive excitation code to the adaptive codebook 33, the noise excitation code to the noise codebook 34, and the weight code to the weight codebook 35. Next, the linear prediction coefficient decoding unit 32 decodes the linear prediction code input from the code input unit 31 to obtain a decoded linear prediction code, and outputs the decoded linear prediction code to the synthesis filter 39.

The synthesis filter 39 constitutes an all-pole model synthesis filter having the decoded linear prediction code obtained by the linear prediction coefficient decoding unit 32 as a coefficient. Also, adaptive codebook 33
Outputs an adaptive excitation vector corresponding to the adaptive excitation code input from code input section 31. Further, the noise codebook 34 outputs an adaptive excitation vector corresponding to the noise excitation code input from the code input unit 31. Furthermore,
The weighting codebook 35 reads out the adaptive excitation gain and the noise excitation gain corresponding to the weighting code input from the code input unit 31, and outputs them to the adaptive excitation gain multiplication unit 36 and the noise excitation gain multiplication unit 37, respectively.

The adaptive excitation gain multiplying section 36 multiplies the adaptive excitation vector output from the adaptive codebook 33 by the adaptive excitation gain output from the weighting codebook 35, and the noise excitation gain multiplication section 37 The noise excitation vector output from the codebook 34 is multiplied by the noise excitation gain output from the weighting codebook 35. And the vector adder 38
Generates the drive excitation vector by adding the output vectors of the adaptive excitation gain multiplication unit 36 and the noise excitation gain multiplication unit 37. Then, with the driving sound source vector,
The synthesizing filter 39 is driven to output the synthesized speech 40 of the received frame section.

In the above-described CELP-type speech coding apparatus / speech decoding apparatus, in order to obtain a high-quality synthesized speech, it is necessary to suppress the distortion ER of (Equation 1) to a small value. For that purpose, it is desirable to specify a combination of the adaptive excitation code, the noise excitation code, and the weight code in a closed loop so as to minimize (Equation 1). However, when trying to identify the distortion ER of (Equation 1) in a closed loop, the amount of calculation becomes too large, and therefore, it is general to specify the above three types of codes in an open loop.

Specifically, first, an adaptive codebook search is performed. Here, the adaptive codebook search process means that a periodic component in a prediction residual vector obtained by passing an input voice through an inverse filter is converted into an adaptive excitation source output from an adaptive codebook storing a driving excitation vector of a past frame. This is a process of performing vector quantization by a vector. Then, the entry number of the adaptive excitation vector having a periodic component in the linear prediction residual vector and a close periodic component is specified as the adaptive excitation code.
Note that the ideal adaptive excitation gain is provisionally determined at the same time by the adaptive codebook search.

Then, a random codebook search is performed. The noise codebook search is a component obtained by removing the periodic component from the linear prediction residual vector of the processing frame, that is, a component obtained by subtracting the adaptive excitation vector component from the linear prediction residual vector (hereinafter, also referred to as a noise excitation target). Is a vector quantization using a plurality of noise excitation vector candidates stored in the noise codebook. Then, by this noise codebook search processing, the entry number of the noise excitation vector that encodes the noise excitation target with the least distortion is specified as the noise excitation code. At the same time, the ideal noise gain is provisionally determined by the noise codebook search.

Finally, a weight codebook search is performed. The weighted codebook search stores in the weighted codebook a vector composed of two elements, an ideal adaptive gain tentatively obtained during the adaptive codebook search and an ideal noise gain tentatively obtained during the noise codebook search. This is a process of encoding (vector quantization) so that distortion is minimized in the obtained gain candidate vector (a vector candidate including two elements of an adaptive excitation gain candidate and a noise excitation gain candidate). Then, the entry number of the gain candidate vector selected here is output to the code output unit as a weight code.

Here, of the above-mentioned general code search processing in the CELP speech coding apparatus, the noise codebook search processing (the processing of specifying the adaptive excitation code and then specifying the noise excitation code) will be described in more detail. Give an explanation.

As described above, in a general CELP encoding apparatus, at the time of performing a random codebook search, the linear prediction code and the adaptive excitation code have already been specified. Here, H is the impulse response matrix of the synthesis filter composed of the linear prediction codes already specified, p is the adaptive excitation vector corresponding to the adaptive excitation code, and ideal adaptive excitation gain that is obtained simultaneously when the adaptive excitation code is specified. Assuming that the (provisional value) is ga, the distortion ER of (Equation 1) is transformed into (Equation 2) below.

[0018]

(Equation 2)

Here, the vector v in (Equation 2) is the input speech signal u in the frame section, the impulse response matrix H of the synthesis filter (default), the adaptive excitation vector p (default),
A noise source of the following (Equation 3) using the ideal adaptive sound source gain ga (provisional value).

[0020]

(Equation 3)

In Equation (1), the noise source vector is c
On the other hand, in (Equation 2), the noise source vector is expressed as ck. This is because (Equation 1) does not specify the entry number (k) of the noise source vector, whereas (Equation 2) specifies the entry number. Despite the differences above, the meaning of the object is the same.

Therefore, the random codebook search is a process of obtaining the entry number k of the noise excitation vector ck so as to minimize the distortion ERk of (Equation 2). Then, when specifying the entry number k of the noise excitation vector ck that minimizes the distortion ERk of (Equation 2), it can be assumed that the noise excitation gain gc can take an arbitrary value. Therefore, the process of finding the entry number that minimizes the distortion of (Equation 2) is replaced by the process of specifying the entry number k of the noise source vector ck that maximizes the following fractional expression Dk of (Equation 4) Can be

[0023]

(Equation 4)

The noise codebook search is performed by (1) calculating the fractional expression Dk of (Equation 4) for each entry number k of the noise excitation vector ck by the distortion calculation unit 16 and outputting the value to the code identification unit 23 . (2) In the code identification unit 23, the entry number k
The value of (Equation 4) for each is compared in magnitude, the entry number k at which the value becomes maximum is determined as the noise excitation code, and output to the code output unit 24. Such a two-stage process is performed.

In the early CELP system, a random codebook in which a plurality of random numbers were entered as noise excitation vectors, that is, a random codebook in which a plurality of random numbers were directly recorded in a memory, was used. On the other hand, in recent low-bit-rate CELP encoding / decoding devices, a noise excitation vector including a small number of non-zero elements whose amplitudes are +1 or −1 (the amplitude of elements other than the non-zero elements is zero) is generated. Many that have an algebraic codebook in the noise codebook have been developed. The algebraic codebook is called "Fast CELP Codi
ng based on Algebraic codes '', J. Adoul et al, Pro
c. IEEE Int. Conf. Acoustics, Speech, Signal Proce
ssing, 1987, pp. 1957-1960 (Reference 2) and “Comparison
of Some Algebraic Structure for CELP Coding of Sp
eech ", J. Adoul et al, Proc. IEEE Int. Conf. Acous
tics, Speech, Signal Processing, 1987, pp. 1953-195
6 (Reference 3).

The algebraic codebook disclosed in the above document can (1) produce a high-quality synthesized sound when applied to the CELP system having a bit rate of about 8 kb / s.
This codebook has excellent features such that a noise excitation codebook can be searched with a small amount of calculation, and (3) a data ROM capacity for directly storing a noise excitation vector is not required. CS-ACELP (bit rate 8 kb / s) or ACELP (bit rate 5.3 kb / s), which is characterized in that an algebraic codebook is provided in the noise codebook section, is described in G.99. 72
9 and g723.1 from ITU-T to 19
It was recommended in 1996. For CS-ACELP, refer to “Design and Description of CS-ACELP: AT
oll Quality 8 kb / s Speech Coder '', Redwan Salamie
t al, IEEE trans. SPEECH AND AUDIO PROCESSING, vo
l. 6, no. 2, March 1998 (Reference 4), etc., discloses the detailed technology.

An algebraic codebook is a codebook having excellent features as described above. However, on the other hand, the algebraic codebook is
When applied to the noise codebook of the ELP encoding / decoding device, the noise excitation target is always encoded (vector quantized) with a noise excitation vector including only a small number of zero elements. There is also a problem that it is impossible to faithfully represent a sound source target with a code. This problem becomes particularly significant when the processing frame corresponds to an unvoiced consonant section, a background noise section, or the like.

This is because the noise source often has a complicated shape in an unvoiced consonant section or a background noise section. Furthermore, when an algebraic codebook is applied to a CELP encoding / decoding device having a bit rate lower than about 8 kb / s, the number of non-zero elements in the noise excitation vector is reduced. Even in a voiced section in which the noise source target tends to have a pulse shape, the above problem may be a problem.

As a method for solving the above problem of the algebraic codebook, as a method, a vector including a small number of non-zero elements (elements other than the non-zero elements have a value of zero) output from the algebraic codebook is used. There is disclosed a pulse spread codebook in which a vector obtained by superimposing a fixed waveform called a diffusion pattern is used as a driving sound source of a synthesis filter. The pulse spread codebook is disclosed in Japanese Patent Laid-Open No. Hei 10-232696 (Reference 5), "ACELP Coding Using Pulse-Spread Structure Exciter" Yasunaga et al., Proceedings of the 1997 IEICE Spring Conference, D-14.
-11, p. 253, 1997-03 (Reference 6), "Low-rate speech coding using pulse spread source" Yasunaga et al., Proc. Of the 1998 Autumn Meeting of the Acoustical Society of Japan, pp. 281-282. , 1998-1
0 (Reference 7).

Next, an outline of the pulse spreading codebook disclosed in the above document will be described with reference to FIGS. FIG. 9 shows a more detailed example of the pulse spreading codebook of FIG.

In the pulse spread codebook 50 of FIGS. 8 and 9, reference numeral 41 denotes a small number of non-zero elements (having an amplitude of +1 or-).
This is an algebraic codebook that generates a pulse vector 42 composed of 1). In the CELP encoding apparatus / decoding apparatus described in Literatures 2, 3, and 4, the pulse vector 42 (constituted by a small number of non-zero elements) output from the algebraic codebook 41 is directly used as a noise source. Used as a vector. Reference numeral 43 in FIGS. 8 and 9 denotes a diffusion pattern storage unit. The diffusion pattern storage unit 43 stores one or more types of fixed waveforms called diffusion patterns for each channel. The diffusion pattern stored for each channel may be either a case where a diffusion pattern having a different shape is stored for each channel or a case where a diffusion pattern having the same shape (common) is stored for each channel. . The case where the diffusion pattern stored for each channel is common corresponds to a simplified case where the diffusion pattern stored for each channel is stored. The case where the shapes of the diffusion patterns stored in the.

The pulse spread codebook 50 is an algebraic codebook 4
Instead of directly outputting the output vector 42 from 1 as a noise excitation vector, the pulse spreading section 45 converts the vector 42 output from the algebraic codebook 41 and the spreading pattern 44 read from the spreading pattern storage section 43 by the pulse spreading section 45. A vector 46 obtained by superimposing for each channel and adding vectors obtained by the superimposition operation is used as a noise source vector.

The pulse spread codebook shown in FIG.
When used in the noise codebook section of the ELP encoding / decoding device, the noise codebook portion of the CELP encoding device of FIG. 6 and the CELP decoding device of FIG. 7 (18 in FIG. 6 and 34 in FIG. 7)
Is replaced by the pulse spread codebook of FIG. 8 (FIG. 9).
0 and FIG. 11 are the configurations of the encoding device and the decoding device, respectively.

The CELP encoding / decoding device disclosed in References 5, 6, and 7 is an encoding device (FIG. 1).
0) and the decoding apparatus (FIG. 11) have the same configuration (the number of channels in the algebraic codebook section, the number and type of spreading patterns registered in the spreading pattern storage section, etc.) between the coding apparatus side and the decoding apparatus. (Common on both sides). Then, by setting the shape, the number of types, and the selection method of a plurality of types of diffusion patterns registered in the diffusion pattern storage unit 43 efficiently, the quality of synthesized speech is improved. I'm trying.

The description of the pulse spreading codebook herein is based on the pulse vector 42 composed of a small number of non-zero elements.
Is described using an algebraic codebook in which the amplitude of the non-zero element is limited to +1 or −1, but the codebook that generates the pulse vector is a non-zero element codebook. It is also possible to use a multi-pulse codebook with unlimited amplitude or a regular pulse codebook, and in such a case, improve the quality of synthesized speech by using a pulse vector superimposed on a spreading pattern as a noise source vector. Can be realized.

Up to now, [1] statistical learning has been performed on the shapes of many noise source targets, and the diffusion patterns of the shapes (References 5, 6, and 7) statistically included in the noise source targets at a high frequency. 2] A random-shaped diffusion pattern for efficiently expressing unvoiced consonant sections and noise sections (Refs. 5, 6,
7), [3] pulse-shaped diffusion patterns for efficiently expressing voiced stationary sections (References 5 and 7), [4] energy of pulse vectors output from algebraic codebooks (non-zero elements) (Where energy is concentrated at the position shown in FIG. 6)). A diffusion pattern having a function of dispersing it to the surroundings (Reference 6), [5] Encoding a speech signal for some appropriately prepared diffusion pattern candidates , Decoding and viewing evaluation of synthesized speech are repeated, and a diffusion pattern selected to output high quality synthesized speech (Reference 5), [6] A diffusion pattern created based on phonetic knowledge (Reference 5) ) Are registered for each non-zero element (channel) in the excitation vector output from the algebraic codebook, and at least one of the registered diffusion patterns and the vector generated by the algebraic codebook are registered. (A few non-zero elements Thus by superimposing and configured) for each channel, by using a material obtained by adding the superimposed results of each channel as a noise source vector, it has been shown to be effective in improving the quality of synthesized speech.

In particular, when the diffusion pattern storage unit 43 registers a plurality of (two or more) types of diffusion patterns per channel, <1> is registered as a method of selecting the plurality of diffusion patterns. Encoding / decoding is performed for all combinations of the spread patterns that have been obtained, and a method of performing closed selection of a spread pattern that minimizes the resulting coding distortion, or <2> when performing a noise codebook search Speech information that has already been clarified (the speech information referred to here is, for example, voiced dynamic information determined using dynamic or static fluctuation of a weight code, or
A method of open-selecting a diffusion pattern using voiced strength information determined using dynamic fluctuation of a linear prediction code) is disclosed (References 5 and 6).

In the following description of the conventional technique, for the sake of simplicity, it is assumed that the spreading pattern storage unit 45 in the pulse spreading codebook of FIG. 9 registers only one type of spreading pattern per channel. Description of the related art will be limited to the pulse spread codebook 50 of FIG.

Here, the noise codebook search processing when the pulse spread codebook is applied to the CELP coding apparatus is compared with the noise codebook search processing when the algebraic codebook is applied to the CELP coding apparatus. The processing will be described. First, a codebook search process in the case where an algebraic codebook is used for the random codebook will be described.

The number of non-zero elements in the vector output by the algebraic codebook is N (the number of channels in the algebraic codebook is N), and one non-zero element with an amplitude of +1 or -1 is output for each channel. (The amplitude of elements other than the non-zero element is zero) di (i is the channel number: 0 ≦ i ≦ N−
1) When the subframe length is L, the noise excitation vector c of the entry number k output by the algebraic codebook
k is represented by the following (Equation 5).

[0041]

(Equation 5)

Then, the following (Equation 6) is obtained by substituting (Equation 5) into (Equation 4).

[0043]

(Equation 6)

The process of specifying the entry number k that maximizes the following (Expression 7) obtained by rearranging the (Expression 6) is the noise codebook search process.

[0045]

(Equation 7)

Where xt = vtH,
M = HtH (v is a noise source). Here, when calculating the value of (Equation 7) for each entry number k, xt = vtH and M = HtH are calculated in the preprocessing stage, and the calculation results are expanded (stored) in a memory.
By introducing this preprocessing, it is possible to greatly reduce the amount of calculation when calculating (Equation 7) for each candidate entered as a noise excitation vector, and as a result, the total amount of noise codebook search required Reducing the amount of calculation is disclosed in (References 2, 3, and 4) and the like, and is generally known.

Next, a description will be given of a random codebook search process when a pulse spread codebook is used as a random codebook.

The number of non-zero elements output by the algebraic codebook, which is a part of the pulse spread codebook, is N (the number of channels of the algebraic codebook is N), and the amplitude output for each channel is +1 or -1. (Where i is the channel number: 0 ≦ i ≦ N−1), the vector containing only one non-zero element (i is the channel number: 0 ≦ i ≦ N−1), and the channel number stored in the diffusion pattern storage unit Assuming that the spreading pattern for i is wi and the subframe length is L, the noise excitation vector ck of the entry number k output by the pulse spreading codebook becomes the following (Equation 8).

[0049]

(Equation 8)

Therefore, in this case, the following (Equation 9) is obtained by substituting (Equation 8) for (Equation 4).

[0051]

(Equation 9)

The process of specifying the entry number k of the noise excitation vector that maximizes the following (Expression 10) obtained by rearranging the (Expression 9) is a noise codebook search using a pulse spread codebook. Processing.

[0053]

(Equation 10)

However, in (Equation 10), xit = vt
Hi (where Hi = HtWi: Wi is a diffusion pattern superposition matrix), and R = HitHj. When calculating the value of Equation 10 for each entry number k, Hi = HtW
It is possible to calculate i and xit = vtHi and R = HitHj and record them in memory. Then, for each candidate entered as a noise source vector (Equation 1)
The amount of calculation when calculating (0) is the same as the amount of calculation when calculating (Equation 7) using an algebraic codebook ((Equation 7) and (Equation 10) are isomorphic) ), The noise codebook search can be performed with a small amount of calculation even when the pulse spread codebook is used (References 5, 6, and 7).

[0055]

In the above prior art, the effect of using a pulse spread codebook for a noise codebook section of a CELP encoding apparatus / decoding apparatus and the effect of using a pulse spread codebook as a noise codebook are described. It has been shown that, when used in the noise codebook section, a random codebook search can be performed in the same manner as when the algebraic codebook is used in the random codebook section. The difference between the amount of computation required for noise codebook search when using an algebraic codebook for the random codebook unit and the amount of computation required for noise codebook search when using the pulse spread codebook for the noise codebook unit is as follows: (Equation 7) and (Equation 10) Differences in the amount of computation required for each preprocessing stage, ie, preprocessing (xt = vtH, M = HtH) and preprocessing (Hi = HtWi, xit = vtHi, R = HitHj) Is the difference in the amount of computation required.

In general, in the CELP encoding apparatus / decoding apparatus, the number of bits that can be allocated to the noise codebook tends to decrease as the bit rate decreases. This tendency leads to a decrease in the number of non-zero elements when constructing a noise excitation vector when using an algebraic codebook or a pulse spread codebook for the noise codebook section. Therefore, CE
As the bit rate of the LP encoder / decoder decreases, the difference in the amount of calculation between the case where an algebraic codebook is used and the case where a pulse spread codebook is used becomes smaller. However, when the bit rate is relatively high, or when it is necessary to minimize the amount of calculation even at a low bit rate, the increase in the amount of calculation in the preprocessing stage caused by using the pulse spreading codebook cannot be ignored. Sometimes.

According to the present invention, in a speech coding apparatus and speech decoding apparatus of the CELP system using a pulse spread codebook for a noise codebook section, and a speech coding / decoding system, an algebraic codebook is used on the coding side. It is an object of the present invention to obtain a high-quality synthesized speech on the decoding side while suppressing the amount of calculation at the time of code search in the preprocessing stage, which is increased as compared with the case where is used for the noise codebook unit.

[0058]

SUMMARY OF THE INVENTION The present invention is to solve the above-mentioned problems which may occur when a pulse spread codebook is used for a noise codebook section of a CELP coding apparatus / decoding apparatus. A speech encoding device characterized by using different spreading patterns on the encoding device side and the decoding device side.
It is an invention of a decoding device. In the present invention, the diffusion pattern storage unit on the speech decoding device side registers the diffusion patterns described in References 2, 3, and 4 as described in the related art, and uses them to store an algebraic codebook. On the other hand, a synthesized speech having higher quality than the case of using it is generated. On the encoding device side, a diffusion pattern registered in the diffusion pattern storage section of the decoding device is simplified (for example, a diffusion pattern thinned out at regular intervals). Pattern, or a diffusion pattern censored at a certain length)
A random codebook search is performed by using this.

Thus, when the pulse spread codebook is used for the noise codebook section, the coding side increases the code search time in the preprocessing stage, which is increased as compared with the case where the algebraic codebook is used for the noise codebook section. Can be reduced, and a high-quality synthesized speech can be obtained on the decoding side.

[0060]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention according to claim 1 of the present invention provides a vector including at least one non-zero element (elements other than the non-zero element have a value of zero) and a fixed waveform called a diffusion pattern. And a pulse spread codebook that generates a vector by superimposing the pulse spread codebook, wherein the pulse spread codebook has a configuration different from the configuration of the pulse spread codebook on the voice decoding device side. There is an effect that the pulse spread codebook of the speech coding apparatus can be configured to reduce the amount of calculation when performing a code search calculation, even though the pulse spread codebook is different from that of the speech decoding apparatus.

According to a second aspect of the present invention, the spreading pattern storage unit, which is a component of the pulse spreading codebook, stores a spreading pattern different from the spreading pattern stored in the spreading pattern storage unit on the speech decoding device side. 2. A speech coding apparatus according to claim 1, wherein the spreading pattern stored in the pulse spread codebook of the speech coding apparatus is different from the spreading pattern stored in the speech decoding apparatus. However, there is an effect that the configuration can be made such that the amount of calculation is reduced when performing the code search calculation.

According to a third aspect of the present invention, the diffusion pattern storage section stores a diffusion pattern obtained by simplifying the diffusion pattern stored in the diffusion pattern storage section of the speech decoding device. 3. The speech coding apparatus according to claim 2, wherein the spreading pattern stored in the pulse spread codebook of the speech coding apparatus is a simplified version of the spreading pattern stored in the speech decoding apparatus. This has the effect of reducing the amount of calculation when performing the code search calculation.

According to the fourth aspect of the present invention, the spreading pattern storage unit reduces the constituent elements of the spreading pattern stored in the spreading pattern storage unit of the speech decoding device to zero at appropriate intervals. Preferably, the speech encoding apparatus according to the second or third aspect stores a diffusion pattern obtained by replacement.

As described in the fifth aspect of the present invention, the spreading pattern storage unit uses N samples (N is a natural number) for the constituent elements of the spreading pattern stored in the spreading pattern storage unit of the speech decoding device. ) Stores a diffusion pattern obtained by replacing it with zero for each case, and the speech coding apparatus according to any one of claims 2 to 4 has a similar effect.

Further, as in the invention according to claim 6, the spreading pattern storage unit reduces the constituent elements of the spreading pattern stored in the spreading pattern storage unit of the speech decoding device to zero for each sample. It is more preferable that the speech encoding apparatus according to the fifth aspect stores a diffusion pattern obtained by replacement.

According to a seventh aspect of the present invention, in the spread pattern storing section, the spread pattern obtained by truncating the constituent elements of the spread pattern stored in the spread pattern storage section of the speech decoding device at an appropriate length. 4. A speech encoding apparatus according to claim 2, wherein the speech pattern is stored in a pulse spreading codebook of the speech encoding apparatus by truncating the speech pattern at an appropriate length. In addition, although the spreading pattern is different from the spreading pattern stored in the speech decoding device, the amount of calculation for code search calculation can be reduced.

Further, as in the invention according to claim 8, the spreading pattern storage unit stores the components of the spreading pattern stored in the spreading pattern storage unit of the speech decoding device as N samples (N is a natural number). ) Stores the diffusion pattern obtained by truncation at the length, and the same effect is exhibited as the speech encoding apparatus according to any one of claims 2, 3, and 7.

According to the ninth aspect of the present invention, the diffusion pattern storage section cuts off the constituent elements of the diffusion pattern stored in the diffusion pattern storage section of the speech decoding device into half length. It is preferable that the speech coding apparatus according to any one of claims 2, 3, and 7 stores the obtained diffusion pattern.

According to a tenth aspect of the present invention, there is provided a speech decoding apparatus for decoding a speech signal having a speech code generated by the speech coding apparatus according to any one of the first to ninth aspects. The decoding device can decode a speech signal coded with a reduced amount of operation and a high-quality synthesized speech on the decoding side.

According to an eleventh aspect of the present invention, there is provided a signal processing processor in which a software program for realizing the audio encoding device according to any one of the first to ninth aspects is described, and the amount of arithmetic operation of the audio encoding is performed. By using a software program in which the signal processing is reduced, it is possible to form a signal processing processor that realizes high speed or low bit rate.

According to a twelfth aspect of the present invention, there is provided a signal processing processor in which a software program for realizing the speech decoding device according to the tenth aspect is described. By using a software program that can decode high-quality speech on the decoding side while using the obtained speech signal, it has an effect that a signal processing processor that realizes high-quality synthesized speech can be formed.

According to a thirteenth aspect of the present invention, the configuration of the pulse spread codebook of the speech encoder is different from the configuration of the pulse spread codebook of the speech decoder. And a system in which the pulse spread codebook of the speech coding apparatus is configured to reduce the amount of calculation when performing a code search operation while using a pulse spread codebook different from that of the speech decoding apparatus. It has the action of:

According to the fourteenth aspect of the present invention, the difference between the structure of the pulse spread codebook of the speech coding apparatus and the structure of the pulse spread codebook of the speech decoding apparatus is that each pulse spread codebook has 14. The speech encoding / decoding system according to claim 13, wherein the speech decoding apparatus has a shape of a diffusion pattern provided, and the diffusion pattern stored in the pulse spread codebook of the speech encoding apparatus is transmitted to the speech decoding apparatus. Although the spreading pattern is different from the stored one, there is an effect that it is possible to form a system configured to reduce the amount of calculation when performing the code search calculation.

According to a fifteenth aspect of the present invention, the shape of the diffusion pattern on the speech encoding device side is a simplified version of the shape of the diffusion pattern on the speech decoding device side. The speech encoding / decoding system according to claim 1, wherein the spreading pattern stored in the pulse spread codebook of the speech encoding apparatus is a simplified version of the spreading pattern stored in the speech decoding apparatus, and a code search operation is performed. This has the effect that a system configured to reduce the amount of computation when performing the operation can be formed.

And, as in the invention of claim 16,
16. The method according to claim 13, wherein the shape of the diffusion pattern on the voice encoding device side is a shape obtained by replacing constituent elements of the diffusion pattern on the voice decoding device side with zero at appropriate intervals. It is preferable to use any one of the speech encoding / decoding systems described above.

Also, as in the invention of claim 17,
The shape of the diffusion pattern on the voice encoding device side is a shape obtained by replacing components of the diffusion pattern on the voice decoding device side with zero for every N samples (N is a natural number). The speech coding / decoding system according to any one of 13 to 16 has a similar effect.

Further, according to the invention of claim 18,
18. The speech code according to claim 17, wherein the shape of the diffusion pattern on the speech encoding device side is a shape obtained by replacing components of the diffusion pattern on the speech decoding device side with zero for each sample. It is more preferable to use an encryption / decryption system.

According to a nineteenth aspect of the present invention, the shape of the diffusion pattern on the audio encoding device side is a shape obtained by truncating the components of the diffusion pattern on the audio decoding device side to an appropriate length. The speech encoding / decoding system according to any one of claims 13 to 15, wherein a speech pattern stored in a pulse spreading codebook of the speech encoding device is truncated at an appropriate length to perform speech decoding. However, there is an effect that a system configured to reduce the amount of calculation at the time of code search calculation can be formed even though the spreading pattern is different from the spreading pattern stored in the coding device.

Further, according to the invention of claim 20,
The spread pattern on the speech encoding device side is a shape obtained by truncating the constituent elements of the diffusion pattern on the speech decoding device side by a length of N samples (N is a natural number). The speech coding / decoding system according to any one of 13, 14, 15, and 19 exhibits the same operation.

Further, according to the invention of claim 21,
2. The speech decoding apparatus according to claim 1, wherein the shape of the diffusion pattern is a shape obtained by truncating a component of the diffusion pattern on the speech decoding apparatus to half its length.
Preferably, the speech encoding / decoding system according to any one of 4, 15, and 19 is used.

According to a twenty-second aspect of the present invention, there is provided a communication base station including the signal processing processor according to the eleventh or twelfth aspect, wherein a software program for reducing the amount of operation on the encoding side is provided. A high-speed or low bit rate can be achieved by a signal processing processor that uses a CDMA, and a high-quality synthesized speech can be realized by a signal processing processor that uses a software program that can decode high-quality speech on the decoding side. Has the effect of realizing a base station as follows.

According to a twenty-third aspect of the present invention, there is provided a communication terminal comprising the signal processing processor according to the eleventh or twelfth aspect, wherein a software program for reducing the amount of operation on the encoding side is provided. High-speed or low-bit-rate can be achieved with the signal processing processor used, and high-quality synthesized speech can be realized with the signal processing processor using a software program that can decode high-quality speech on the decoding side. This has the effect of realizing a terminal that performs

According to a twenty-fourth aspect of the present invention, there is provided a wireless communication system in which the communication base station according to the twenty-second aspect and the communication terminal according to the twenty-third aspect are connected by a wireless network. High-speed synthesis or low-bit-rate processing can be achieved with a signal processing processor using a software program that reduces the noise, and high-quality synthesis can be achieved with a signal processing processor that uses a software program that can decode high-quality audio on the decoding side. This has the effect of realizing a wireless communication system capable of realizing voice.

The use of different spreading patterns on the encoding device side and the decoding device side means that a previously prepared spreading vector (for the decoding device) is modified while retaining its characteristics, so that it is used for the encoder. It is to obtain the diffusion vector.

Here, as a method of preparing a spreading vector for a decoding device in advance, a method disclosed in a patent (Japanese Patent Application Laid-Open No. 10-63300) filed by the present inventors, that is, a method for searching for a sound source, A method of preparing by learning the statistical tendency of the target vector, a method of actually encoding the sound source target, and a method of preparing by repeatedly performing an operation of gradually deforming in a direction to reduce the sum of the encoding distortion generated at that time, A method of designing based on phonetic knowledge to improve the quality of synthesized speech, a method of designing for the purpose of randomizing a high-frequency phase component of a pulse sound source, and the like can be considered. All of these details are included here.

The diffusion vector obtained in this way is
In any case, the amplitude of the sample (front sample) close to the first sample of the diffusion vector is relatively larger than the amplitude of the rear sample. In particular, the amplitude of the first sample is often the largest among all the samples in the spreading vector (this is almost always the case).

As a specific method for obtaining a spreading vector for an encoder by deforming a spreading vector for a decoding device while maintaining its characteristics, the following method can be mentioned. 1) A spreading vector for an encoder is obtained by replacing a sample value of a spreading vector for a decoding device with zero at an appropriate interval. 2) The spreading vector for the encoder is obtained by truncating the spreading vector for the decoding device of a certain length at an appropriate length. 3) A threshold value of the amplitude is set in advance, and a sample whose amplitude is smaller than the threshold value set for the spreading vector for the decoding device is replaced with zero, thereby obtaining a spreading vector for the encoder. 4) The spreading vector for the encoding device is stored by storing the sample value at an appropriate interval including the first sample for the decoding device of a certain length and replacing the values of the other samples with zero. To win.

Here, for example, when several samples from the front of the diffusion vector are used as in the above method 1),
While preserving the general shape (rough characteristics) of the diffusion vector,
It is possible to newly obtain a spreading vector for the encoding device.

Further, for example, as in the above method 2), even if the sample value is replaced with zero at an appropriate interval, the original shape (rough characteristics) of the original diffusion vector is maintained.
It becomes possible to newly obtain a spreading vector for the encoding device. In particular, in the case of the above method 4), the limitation that the amplitude of the first sample, which often has the largest amplitude, is always stored as it is is imposed, so that the outline of the original diffusion vector can be stored more reliably. It is possible to keep.

Further, as in the method 3), the sample having the amplitude equal to or larger than the specific value is stored as it is, and the amplitude of the sample having the amplitude equal to or smaller than the specific value is replaced with zero. Characteristics) are preserved,
It is possible to obtain a spreading vector for the encoding device.

Hereinafter, embodiments of the present invention will be described with reference to FIGS. 1 to 4. (Embodiment 1) FIGS. 1 and 2 show a CELP-type speech coding apparatus and speech decoding apparatus, respectively, in which a pulse spread codebook is used for a noise codebook section in this embodiment. .

The difference between FIG. 1 and FIG. 2 and FIG. 10 and FIG. 11 used in the prior art is the same configuration (the number of channels of the algebraic codebook unit and the spreading pattern storage unit) in FIG. 10 and FIG. 1 and 2 use pulse spread codebooks having different configurations, whereas pulse spread codebooks having the same number of registered diffusion patterns and the same shape are used.

Then, the pulse spreading codebook A in FIG.
101 and the configuration of the pulse spreading codebook B111 in FIG.
3 (a) and 3 (b) respectively. FIG.
When comparing the pulse spread codebook of FIG. 3A with the pulse spread codebook of FIG. 3B, the difference in the configuration is that the shape of the spreading pattern registered in the spreading pattern storage unit is different. FIG.
In the audio decoding device side of (b), the diffusion pattern storage unit B1
12, a diffusion pattern similar to the diffusion pattern described in the related art, that is, [1] statistical learning of the shape of many noise source targets, and the shape of a shape that is statistically frequently included in the noise source target. A diffusion pattern, [2] a random-shaped diffusion pattern for efficiently expressing unvoiced consonant sections and noise sections, [3] a pulse-shaped diffusion pattern for efficiently expressing voiced stationary sections, [4] A diffusion pattern of a shape giving an effect of dispersing the energy of the excitation vector output from the algebraic codebook (energy is concentrated at the position of the non-zero element) to the surroundings, [5] appropriately prepared For some of the diffusion pattern candidates, encoding, decoding, and listening / listening evaluation of synthesized speech are repeated,
A diffusion pattern selected to output a high-quality synthesized speech, [6] one of diffusion patterns created based on phonetic knowledge (noise source vector B1
13) is registered for each channel.
On the other hand, on the speech encoding device side in FIG. 3A, the diffusion pattern registered in the diffusion pattern storage portion on the speech decoding device side in FIG. A diffusion pattern (noise source vector A103) replaced with zero for each sample is registered.

The CELP constructed as described above
In the speech encoding device / speech decoding device, the speech is decoded in the same manner as described in the related art, without being aware that different spreading patterns are registered on the encoding device side and the decoding device side. Encode and decode signals.

Then, the coding apparatus can reduce the amount of pre-processing calculation at the time of searching for a random codebook when the pulse spread codebook is used for the random codebook section (calculation of Hi = HtWi and xit = vtHi). The amount can be reduced to about half), and the decoding device can diffuse the energy concentrated at the non-zero element position to the surroundings by superimposing the conventional diffusion pattern on the pulse vector, and synthesize speech. Quality can be improved.

In this embodiment, as shown in FIGS. 3 (a) and 3 (b), the speech encoding device sets the spreading pattern used by the speech decoding device to zero every other sample. Although the case where the replaced diffusion pattern is used has been described, the speech coding apparatus uses a diffusion pattern obtained by replacing the elements of the spreading pattern used by the speech decoding apparatus with zeros every N (N ≧ 1) samples. If you have
The present embodiment can be applied as it is, and in that case, the same operation can be obtained.

Further, in the present embodiment, as in the description of the prior art, the diffusion pattern storage unit has one channel per channel.
Although the embodiment in which the diffusion pattern is registered for each type has been described, two or more types of diffusion patterns are registered for each channel, and a pulse spreading code characterized in that these diffusion patterns are selected and used. The present invention can be applied to a CELP speech coding apparatus / decoding apparatus using a book as a noise codebook unit, and the same operation and effect can be obtained in such a case.

Further, in the present embodiment, as in the description of the prior art, the algebraic codebook section uses a pulse spread codebook that outputs a vector including three non-zero elements. However, the present embodiment can be applied to the case where the number of non-zero elements in the vector output by the algebraic codebook unit is M (M ≧ 1). Function and effect can be obtained.

Further, in this embodiment, as in the description of the prior art, a case where an algebraic codebook is used as a codebook for generating a pulse vector composed of a small number of non-zero elements has been described. As a codebook for generating a pulse vector, the present embodiment can be applied to a case where other codebooks such as a multi-pulse codebook and a regular pulse codebook are used.・ Effects can be obtained.

(Embodiment 2) FIGS. 1 and 2 show a CELP speech encoding apparatus and speech decoding apparatus according to the present embodiment, respectively, wherein a pulse spread codebook is used for a noise codebook section. Device.

The difference between FIGS. 1 and 2 and FIGS. 10 and 11 is that FIGS. 10 and 11 have the same configuration (the number of channels in the algebraic codebook section, the spreading pattern registered in the spreading pattern storage section). 1 and 2 use pulse spread codebooks having different configurations, while the pulse spread codebooks having the same number of types and the same shape are used.

Then, the pulse spreading codebook A in FIG.
FIGS. 4A and 4B show the configuration of the pulse spreading codebook B in FIG. 2 and FIG. FIG. 4A and FIG.
When comparing the configuration of the pulse spread codebook in (b), the difference in the configuration is that the length of the spreading pattern registered in the spreading pattern storage unit is different. On the side of the speech decoding apparatus in FIG. 4B, the diffusion pattern storage unit B112 stores, in the diffusion pattern storage unit B112, a diffusion pattern similar to the diffusion pattern described in the related art, that is, [1] the shape of many noise sound source targets. And a diffusion pattern having a shape that is statistically included in the noise source at a high frequency, [2] a diffusion pattern having a random shape for efficiently expressing unvoiced consonant sections and noise sections,
[3] A pulse-shaped diffusion pattern for efficiently expressing a voiced stationary section, [4] Energy of an excitation vector output from an algebraic codebook (energy is concentrated at the position of a non-zero element) ) Is distributed around the surroundings, [5] For some appropriately prepared diffusion pattern candidates, the audio signal is encoded, decoded, and the listening / listening evaluation of the synthesized voice is repeated. One of diffusion patterns (noise source vector B113) selected from a diffusion pattern selected to output a high-quality synthesized speech and [6] a diffusion pattern created based on phonetic knowledge is 1 per channel. Each type is registered. And meanwhile,
In the speech encoding device side of FIG. 4A, the diffusion pattern registered in the speech pattern storage unit of the speech decoding device side of FIG. Diffusion pattern (noise source vector A105)
Is registered.

The CELP constructed as described above
In the speech encoding device / decoding device, without being aware that different spreading patterns are registered on the encoding device side and the decoding device side, in the same manner as described in the related art,
Encode and decode audio signals.

Then, the coding apparatus can reduce the amount of pre-processing calculation at the time of searching for a random codebook when the pulse spread codebook is used for the random codebook section (calculation of Hi = HtWi and xit = vtHi). The amount can be reduced to about half), and the decoding apparatus can improve the quality of synthesized speech by using the conventional diffusion pattern.

In this embodiment, as shown in FIGS. 4 (a) and 4 (b), the speech coding device cuts off the spreading pattern used by the speech decoding device at half the length. Although the case where the spread pattern is used has been described, the speech coding apparatus side searches for a noise codebook when the spreading pattern used in the speech coding apparatus is cut off at a shorter length N (N ≧ 1). The effect that it becomes possible to further reduce the amount of pre-processing calculation at the time is obtained. However, in this case, the case where the spreading pattern used on the side of the speech coding apparatus is cut off at the length of 1 corresponds to a speech coding apparatus that does not use a spreading pattern (a spreading pattern is applied to a speech decoding apparatus).

Further, in the present embodiment, as in the description of the prior art, the case where the diffusion pattern storage unit registers one type of diffusion pattern per channel has been described. The present embodiment is also applied to a speech coding apparatus / speech decoding apparatus using a pulse spread codebook for a noise codebook unit, characterized in that the above-mentioned spread patterns are registered and the spread patterns are selected and used. It is possible to apply, and in that case, the same operation and effect can be obtained.

Further, in the present embodiment, as in the description of the prior art, the case where the algebraic codebook unit uses a pulse spread codebook that outputs a vector including three non-zero elements is described. However, the present embodiment can be applied to the case where the number of non-zero elements in the vector output by the algebraic codebook unit is M (M ≧ 1). Function and effect can be obtained.

In the present embodiment, a case has been described in which the speech coding apparatus uses a diffusion pattern obtained by truncating the spreading pattern used by the speech decoding apparatus by half the length. Is used in combination with (Embodiment 1), that is, the speech encoding device cuts off the spreading pattern used by the speech decoding device at a length N (N ≧ 1), and furthermore, Is M (M ≧ 1)
It is also possible to replace each sample with zero, in which case the amount of code search operation can be further reduced.

(Embodiment 3) FIG. 5 is a conceptual diagram showing a configuration of a wireless communication system according to the present embodiment. 20
1 is a communication terminal, 202 is a signal processing processor that describes a software program that implements a speech encoding device, 203 is a signal processing processor that describes a software program that implements a speech decoding device, and 204 is a processor 202 And 203, a communication terminal 205 having the same configuration as 201, a communication base station 206, and a signal processing unit 207 for describing a software program for realizing the speech coding apparatus. A processor 208 is a signal processing processor that describes a software program for realizing the speech decoding device;
Reference numeral 09 denotes a speech encoding / decoding system including processors 207 and 208.

Communication terminals 201 and 205 include a signal processing processor 202 described in (Embodiment 1) or (Embodiment 2) and describing a software program for realizing a speech encoding apparatus, and speech decoding. And an audio encoding / decoding system 204 having a signal processing processor 203 in which a software program for realizing the encoding device is described, and transmits and receives an audio signal via a communication base station 206.

The communication base station 206 communicates with the communication terminal 201.
Has a speech encoding / decoding system 209 having a configuration similar to that of the speech encoding / decoding system 204, and realizes a signal processing processor 207 and a speech decoding device in which a software program for implementing the speech encoding device is described. Signal processing processor 208 describing a software program
Includes

By connecting a communication terminal and a communication base station as shown in FIG. 5 with a wireless network, high-quality synthesized speech can be realized by decoding while decoding at high speed and low bit rate. A wireless communication system can be constructed.

In FIG. 5, a radio communication system is constructed by two communication terminals and one communication base station, but the number of terminals and base stations is not limited to this.

[0114]

As described above, according to the present invention, a speech coding apparatus and a decoding apparatus of the CELP system using a pulse spread codebook for a noise codebook unit, and a speech coding and decoding system obtain by learning. A fixed waveform that is frequently included in the generated noise source target is registered as a diffusion pattern, and the diffusion pattern is superimposed (reflected) on the pulse vector to use a noise source vector closer to the noise source target. Therefore, it is possible to improve the quality of synthesized speech on the decoding side, and further, to perform a noise codebook search operation, which may be a problem when the pulse spreading codebook is used for the noise codebook section on the encoding side. This has the advantageous effect that the amount can be kept lower than in the prior art (References 5, 6, 7).

The same operation and effect can be obtained when other codebooks such as a multi-pulse codebook and a regular pulse codebook are used as a codebook for generating a pulse vector composed of a small number of non-zero elements. be able to.

[Brief description of the drawings]

FIG. 1 is a configuration block diagram of a CELP speech encoding device according to a first embodiment of the present invention.

FIG. 2 is a configuration block diagram of a CELP speech decoding device according to Embodiment 1 of the present invention.

FIG. 3A is a conceptual diagram showing an example of a pulse spreading codebook A used in the speech encoding device according to the first embodiment of the present invention. FIG. 3B is a conceptual diagram showing a pulse used in the speech decoding device according to the first embodiment of the present invention. Conceptual diagram showing an example of spreading codebook B

FIG. 4 (a) is a conceptual diagram showing an example of a pulse spreading codebook A used in a speech coding apparatus according to Embodiment 2 of the present invention. (B) Pulses used in a speech decoding apparatus according to Embodiment 2 of the present invention. Conceptual diagram showing an example of spreading codebook B

FIG. 5 is a conceptual diagram showing a configuration of a wireless communication system according to a third embodiment of the present invention.

FIG. 6 is a block diagram showing the configuration of a conventional CELP speech coding apparatus.

FIG. 7 is a configuration block diagram of a conventional CELP speech decoding device.

FIG. 8 is a configuration block diagram of a conventional pulse spreading codebook.

FIG. 9 is a block diagram showing an example of a detailed configuration of a conventional pulse spreading codebook.

FIG. 10 is a block diagram showing the configuration of a conventional CELP speech coding apparatus using a pulse spread codebook as a noise codebook unit.

FIG. 11 is a block diagram showing a configuration of a conventional CELP speech decoding apparatus using a pulse spread codebook as a noise codebook unit.

FIG. 12 is a block diagram showing an example of a detailed configuration of a conventional pulse spreading codebook.

[Explanation of symbols]

 Reference Signs List 11 input speech 12 linear prediction analysis unit 13 linear prediction coefficient encoding unit 14, 32 linear prediction code decoding unit 15, 39 synthesis filter 16 distortion calculation unit 17, 33 adaptive codebook 18, 34 noise codebook 19, 36 adaptive sound source Gain 20, 37 Noise excitation gain 21, 35 Weighted codebook 22, 38 Vector addition unit 23 Code identification unit 24 Code output unit 31 Code input unit 40 Synthetic speech 41 Algebraic codebook 42 Pulse vector 43 Diffusion pattern storage unit 44 Diffusion pattern 45 pulse spreading section 46 noise excitation vector 50 pulse spreading codebook 101 pulse spreading codebook A 111 pulse spreading codebook B 102, 104 spreading pattern storage section A 103, 105 noise excitation vector A 112 diffusion pattern storage section B 113 noise excitation vector B 201,205 Communication terminal 202 Signal processing Processor (encoding) 203 signal processing processor (decoding) 204 voice coding / decoding system (terminal side) 206 communication base station 207 signal processing processor (coding) 208 signal processing processor (decoding) 209 Speech coding / decoding system (base station side)

Claims (24)

[Claims] 1. A pulse spreading codebook that generates a vector by superimposing a vector including at least one non-zero element (elements other than the non-zero element have a value of zero) and a fixed waveform called a spreading pattern. Comprising, said pulse spread codebook,
An audio encoding device having a configuration different from the configuration of the pulse spread codebook on the audio decoding device side. 2. A spread pattern storage unit, which is a constituent part of the pulse spread codebook, stores a spread pattern different from the spread pattern stored in the spread pattern storage unit of the speech decoding device. The speech encoding device according to claim 1, wherein 3. The spread pattern storage unit stores a spread pattern obtained by simplifying the spread pattern stored in the spread pattern storage unit of the speech decoding device. 3. The speech encoding device according to 2. 4. A diffusion pattern storage unit stores a diffusion pattern obtained by replacing constituent elements of the diffusion pattern stored in the diffusion pattern storage unit of the speech decoding device with zeros at appropriate intervals. 4. The speech encoding device according to claim 2, wherein 5. The spreading pattern storage unit obtains a spreading pattern obtained by replacing the constituent elements of the spreading pattern stored in the spreading pattern storage unit of the speech decoding device with zeros for every N samples (N is a natural number). The speech encoding device according to any one of claims 2 to 4, wherein is stored. 6. A diffusion pattern storage unit stores a diffusion pattern obtained by replacing the constituent elements of the diffusion pattern stored in the diffusion pattern storage unit of the speech decoding device with zero for each sample. The speech encoding device according to claim 5, wherein: 7. The diffusion pattern storage unit stores a diffusion pattern obtained by truncating the constituent elements of the diffusion pattern stored in the diffusion pattern storage unit of the speech decoding device at an appropriate length. 4. The speech encoding device according to claim 2, wherein: 8. A spreading pattern storage unit, wherein a spreading pattern obtained by truncating the constituent elements of the spreading pattern stored in the spreading pattern storage unit of the speech decoding device at a length of N samples (N is a natural number). 8. The speech encoding device according to claim 2, wherein 9. The diffusion pattern storage unit stores a diffusion pattern obtained by truncating a component of the diffusion pattern stored in the diffusion pattern storage unit of the speech decoding device to half its length. The speech encoding device according to any one of claims 2, 3, and 7, wherein 10. A speech decoding device for decoding a speech signal having a speech code generated by the speech encoding device according to any one of claims 1 to 9. 11. A signal processing processor in which a software program for realizing the speech encoding device according to claim 1 is described. 12. A signal processing processor in which a software program for realizing the speech decoding device according to claim 10 is described. 13. A speech encoding / decoding system characterized in that the configuration of the pulse spread codebook of the speech encoding device is different from the configuration of the pulse spread codebook of the speech decoding device. 14. The difference between the configuration of the pulse spread codebook of the speech encoding device and the configuration of the pulse spread codebook of the speech decoding device is the shape of the diffusion pattern provided in each pulse spread codebook. The speech encoding / decoding system according to claim 13, wherein: 15. The speech coding / decoding according to claim 14, wherein the shape of the diffusion pattern on the speech encoding device side is a simplified version of the shape of the diffusion pattern on the speech decoding device side. system. 16. The spread pattern on the speech encoding device side is a shape obtained by replacing the components of the spread pattern on the speech decoding device side with zeros at appropriate intervals. Item 16. An audio encoding / decoding system according to any one of Items 13 to 15. 17. The shape of the diffusion pattern on the side of the audio encoding device is a shape obtained by replacing the components of the diffusion pattern on the side of the audio decoding device with zero for every N samples (N is a natural number). 17. The speech encoding / decoding system according to claim 13, wherein 18. The spread pattern on the speech encoding device side is a shape obtained by replacing the components of the spread pattern on the speech decoding device side with zero for each sample. 18. The speech encoding / decoding system according to 17. 19. The spread pattern on the audio encoding device side is a shape obtained by truncating a component of the diffusion pattern on the audio decoding device side to an appropriate length. 16. The speech encoding / decoding system according to any one of items 1 to 15. 20. The shape of the diffusion pattern on the speech encoding device side is a shape obtained by truncating the components of the diffusion pattern on the speech decoding device side by a length of N samples (N is a natural number). Claims 13, 14, 15, 19 characterized by the above-mentioned.
A speech encoding / decoding system according to any one of the above. 21. The spread pattern on the speech encoding device side is a shape obtained by truncating a component of the diffusion pattern on the speech decoding device side to half its length. , 14, 15, 19. 22. A communication base station comprising the signal processing processor according to claim 11. 23. A communication terminal comprising the signal processing processor according to claim 11. 24. A wireless communication system in which the communication base station according to claim 22 and the communication terminal according to claim 23 are connected by a wireless network.