JP3024455B2 - Audio encoding device and audio decoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech encoding / decoding apparatus used for digitally transmitting or storing speech, and for performing vector quantization of speech source signal information.

[0002]

2. Description of the Related Art FIG. 8 shows a conventional speech encoding / decoding apparatus for separating speech into spectrum envelope information and excitation signal information and vector-quantizing the excitation signal information. Figure 8 shows WBKleijn, DJKrasinski, and RHKet
by chum, “Improved Speech Quality and Efficient Vec
tor Quantization in SELP ”(ICASSP'88, pp.155-158,
1988). In the figure, 1 is an encoding unit, 2 is a decoding unit, 3 is a transmission path,
4 is an input voice and 5 is an output voice. 6 is a spectrum analysis means, 7 is a spectrum parameter coding means,
8 is an adaptive excitation coding means, and 9 and 14 are adaptive excitation codebooks. Reference numeral 10 denotes a driving excitation coding unit, and reference numerals 11 and 16 denote driving excitation codebooks. 12, 17 are sound source vector generating means;
13 is an adaptive excitation decoding means, 15 is a driving excitation decoding means, 18 is a spectrum parameter decoding means, 19
Is a synthesis filter.

[0003] The operation of the conventional speech encoding / decoding device will be described below. First, the operation of the encoding unit 1 will be described. The spectrum analysis means 6 analyzes the input speech 4 and extracts spectrum parameters. The spectrum parameter encoding means 7 quantizes the spectrum parameters, outputs a code corresponding to the quantized spectrum parameters to the decoding unit 2 via the transmission path 3, and outputs the quantized spectrum parameters to the adaptive excitation encoding means 8 and the driving excitation code. Means 1
Output to 0. The adaptive excitation codebook 9 stores the excitation signal obtained in the preceding frame, and outputs an adaptive excitation vector obtained by repeating the excitation signal in the current frame at a pitch cycle using the pitch information as a code.
FIG. 9 shows an example of the excitation signal in the preceding frame and the adaptive excitation vector obtained at that time. The adaptive sound source vector is
When the pitch period is shorter than the frame length, FIG.
As shown in (b), the sound source signal is repeated at a pitch cycle in the current frame.

[0004] Adaptive excitation coding means 8 includes M adaptive excitation vectors a _i (i
= 1,..., M) and the spectrum parameters inputted from the spectrum parameter encoding means 7,
The M synthesized speech vectors A _i (i = 1,..., M) are synthesized. Then, the inter-vector distance D _i between the input speech vector X cut out for each frame from the input speech 4, for example, determined according to equation (1). Here, the gain β _i is determined according to, for example, equation (2) so that the distance D _i is minimized. (Note that t indicates transposition.)

[0005]

(Equation 1)

Next, the vector A _{I in} which the distance between the vectors is minimized is searched, and the code I and the code of the gain β _{qI obtained} by quantizing the gain β _I are output to the decoding unit 2 via the transmission line 3. Also, the selected adaptive _excitation vector a _I and its gain β _qI are output to the _excitation vector generation means 12, and the error vector X ′ X ′ = X−β _qI A _I is output to the driving _excitation coding means 10. I do.

[0007] The driving excitation codebook 11 stores, for example, N driving excitation vectors generated from random noise. The driving excitation coding means 10 generates a driving excitation vector c _j (j = 1,...) Inputted from the driving excitation codebook 11.
, N) and the spectrum parameters input from the spectrum parameter encoding means, and synthesizes N synthesized speech vectors C _j (j = 1,..., N). Then, the inter-vector distance D _j with respect to the error vector X ′ input from the adaptive excitation coding means 8 is calculated by, for example, the equation (3).
Ask according to. Here, the gain γ _j is determined according to, for example, equation (4) so that the distance D _j is minimized.

[0008]

(Equation 2)

Next, a search is made for a vector C _J that minimizes the inter-vector distance, and the code of the code J and the gain γ _{qJ obtained} by quantizing the gain γ _J are output to the decoding unit 2 via the transmission path 3. At the same time, the selected drive _excitation vector c _J and its gain γ _qJ are output to the _excitation vector generation means 12.

_Excitation vector generation means 12 generates adaptive _excitation vector a _I and gain β _qI inputted from adaptive _excitation coding means 8 and driving _excitation vector c _J and driving _excitation vector c _J inputted from driving _excitation encoding means 10. its gain gamma generate excitation vector _{β qI a I + γ qJ c} J than _QJ, and outputs it to the adaptive excitation codebook 9.

Next, the operation of the decoding unit 2 will be described. The adaptive excitation decoding means 13 receives an adaptive excitation vector a from the adaptive excitation codebook 14 storing the same adaptive excitation vector as the adaptive excitation codebook 9 based on the code I of the adaptive excitation vector input from the encoding unit 1. _I is read out, and the gain β _qI is decoded from the sign of the gain for the adaptive _excitation vector input from the encoding unit 1, and the adaptive _excitation vector a _I and its gain β _qI are output to the _excitation vector generation means 17. . Further, the driving excitation decoding means 15 includes:
Based on the code J of the driving excitation vector inputted from the encoding unit 1, the driving excitation vector C _J is read from the driving excitation codebook 16 storing the same driving excitation vector as the driving excitation codebook 11, and the code is read. The gain γ _qJ is decoded from the sign of the gain for the drive _excitation vector input from the _converting unit 1, and the drive _excitation vector c _J and its gain γ _qJ are output to the _excitation vector generation unit 17.

The excitation vector generation means 17 outputs the adaptive excitation vector a _I inputted from the adaptive excitation decoding means 13.
And its gain β _qI , the driving _excitation vector c _J inputted from the driving _excitation decoding means 15, and its gain γ
generate excitation vector _{β qI a I + γ qJ c} J than _QJ, and outputs it to the adaptive excitation codebook 14 and synthesis filter 19.

The spectrum parameter decoding means 18 comprises:
The spectrum parameter is decoded based on the sign of the spectrum parameter input from the encoding unit 1 and output to the synthesis filter 19. The synthesis filter 19 synthesizes the output speech 5 using the sound source vector input from the sound source vector generation means 17 and the spectrum parameter input from the spectrum parameter decoding means 18.

[0014]

The voice signal includes a voiced sound and an unvoiced sound, each of which is characterized by whether the sound source signal has a pulse component of a pitch period or white noise having no periodicity. The reproducibility of the pitch periodicity of voiced sounds has a large effect on the quality of synthesized speech. In the conventional speech encoding / decoding device as described above, the pulse-like component of the pitch period of the sound source signal in the voiced sound is generated by using the adaptive sound source vector created by the sound source signal of the preceding frame. However, since the adaptive sound source vector is composed of a plurality of samples, a pulse-like component necessary for voiced sound cannot be actively generated. For this reason, there has been a problem that an appropriate sound source cannot be generated in a transition portion from unvoiced sound to voiced sound, and the sound quality is deteriorated. In addition, even if a pulse-like component is generated, since a sound source is generated including other samples, an optimal pulse sequence is not always obtained, and this leads to a lack of pitch periodicity and a deterioration in sound quality. Was.

Further, since the excitation signal is generated by the linear sum of the adaptive excitation vector and the driving excitation vector not considering the pitch periodicity, even if the adaptive excitation vector generates the pitch periodicity, the amount of transmitted information is reduced. In order to reduce it, it is necessary to increase the frame length, that is, the vector length, or to reduce the codebook size of the driving excitation codebook. In this case, the pitch periodicity of the generated excitation signal is disturbed and the sound quality is reduced. There was also a problem of deterioration.

The present invention has been made to solve such a problem, and in addition to an adaptive excitation vector, generates an excitation vector consisting of a periodic pulse train that can be uniquely determined for the adaptive excitation vector, By using this to generate a sound source signal and repeatedly using the drive sound source vector in synchronization with the pitch period, the pitch periodicity of the sound source signal is less disturbed even with a small amount of transmitted information, thereby synthesizing high-quality decoded speech. It is intended to be.

[0017]

A speech encoding apparatus according to the present invention comprises a spectrum analysis encoding means, an adaptive excitation codebook, a periodic pulse excitation generation means, a multi-vector adaptive excitation encoding means, and an excitation vector generation means. A speech encoding apparatus for encoding input speech for each frame divided at a fixed interval, wherein the spectrum analysis encoding means extracts and encodes a spectrum parameter which is spectrum envelope information of the input speech, and generates an adaptive excitation code. The book generates and stores a plurality of adaptive excitation vectors from the excitation signal output by the excitation vector generation means in the preceding frame, and the periodic pulse excitation generation means generates a main adaptive excitation vector output from the adaptive excitation codebook. A periodic pulse excitation vector determined as a component is output, and the multi-vector adaptive excitation encoding means outputs the encoded A decoded speech generated from a spectral parameter, one of the plurality of adaptive excitation vectors, a first gain applied to the adaptive excitation vector, the periodic pulse excitation vector, and a second gain applied to the periodic pulse excitation vector; The adaptive excitation vector, the first gain, and the second gain are determined and encoded so that distortion from the input speech is minimized, and the excitation vector generation unit determines at least the determined and encoded adaptive excitation vector and the first and second gains.
A sound source signal is generated and output from a gain, a second gain, and the periodic pulse sound source vector.

According to a second aspect of the present invention, there is provided a decoding apparatus comprising: a spectrum parameter decoding unit; a multi-vector adaptive excitation decoding unit; an adaptive excitation codebook;
An audio decoding apparatus comprising: an excitation vector generating unit; a synthesis filter; and generating a decoded signal from a code of a spectrum parameter encoded for each frame divided at a fixed interval, a code of an adaptive excitation vector, and a code of a gain. , The spectrum parameter decoding means is input with the code of the spectrum parameter, decodes the spectrum parameter from the code of the spectrum parameter, the multi-vector adaptive excitation decoding means is input with the code of the adaptive excitation vector and the code of the gain, The adaptive excitation vector code is output to the adaptive excitation codebook, and the first
The adaptive excitation codebook generates and stores a plurality of adaptive excitation vectors from the excitation signal output by the excitation vector generation means in the preceding frame, and decodes the stored adaptive excitation code. Outputting an adaptive excitation vector corresponding to the code of the adaptive excitation vector output from the multi-vector adaptive excitation decoding means from among the vectors, the periodic pulse excitation generating means outputting the adaptive excitation vector output from the adaptive excitation codebook A periodic pulse excitation vector determined as a main component of the above is output, and the excitation vector generating means adds at least the adaptive excitation vector to which the first gain has been added and the periodic pulse excitation vector to which the second gain has been added, and outputs as an excitation signal The synthesis filter uses the decoded spectrum parameters to generate the sound source. It is obtained so as to generate a decoded signal from the item.

According to a third aspect of the present invention, there is provided a speech decoding apparatus comprising:
A spectrum analysis coding unit, a pitch position extraction unit, an adaptive excitation codebook, a driving excitation codebook, a pitch synchronization unit, an adaptive excitation coding unit, a driving excitation coding unit, and an excitation vector generation unit are provided, and are divided into a fixed interval. A speech encoding apparatus for encoding an input speech for each frame, wherein a spectrum analysis encoding means extracts and encodes a spectrum parameter which is spectrum envelope information of the input speech, and a pitch position extraction means, The adaptive excitation codebook generates and stores a plurality of adaptive excitation vectors from the excitation signal output by the excitation vector generation means in the preceding frame, by extracting a plurality of feature points arranged at a pitch cycle interval as pitch positions and encoding them. The driving excitation codebook stores a plurality of driving excitation vectors having a predetermined pitch synchronization position, Means for adjusting a pitch synchronization position of each of the drive excitation vectors to the encoded pitch position, generating a pitch-synchronous drive excitation vector in which the drive excitation vector is repeated at a pitch cycle, and wherein the adaptive excitation encoding means includes A synthesized speech vector of the adaptive sound source vector is generated based on the converted spectral parameters. Determining the gain, outputting the code of the determined adaptive excitation vector, encoding the first gain, and driving excitation encoding means based on the encoded spectral parameters, And the determined adaptive sound source vector to which the determined first gain has been added. When the sum of the synthesized speech vector of the pitch-synchronous drive excitation vector to which the second gain is added and the synthesized speech vector of the pitch-synchronized excitation vector is used as the decoded signal vector, the pitch-synchronous drive is performed so that the distortion of the input speech and the decoded signal vector is minimized. Determining a synthesized speech vector and a second gain of the sound source vector, outputting a code of the determined pitch synchronous drive sound source vector, and encoding the determined second gain; An excitation signal is generated and output from the adaptive excitation vector, the first gain, the second gain, and the pitch-synchronized excitation vector that have been encoded.

According to a fourth aspect of the present invention, there is provided a speech encoding apparatus comprising:
The apparatus includes a spectrum analysis encoding unit, an adaptive excitation codebook, a driving excitation codebook, a pitch synchronization unit, an adaptive excitation encoding unit, a driving excitation encoding unit, and an excitation vector generation unit, and receives an input speech for each frame divided at regular intervals. Wherein the spectrum analysis encoding means extracts and encodes a spectrum parameter which is spectrum envelope information of the input speech, and the adaptive excitation codebook includes the excitation vector generation means in a preceding frame. A plurality of adaptive excitation vectors are generated and stored from the excitation signal output by the above, the driving excitation codebook stores a plurality of driving excitation vectors having a predetermined pitch synchronization position, and the pitch synchronization means The pitch of the driving sound source vector is adjusted to the pitch positions, which are a plurality of points arranged at pitch interval intervals, and the driving sound source vector An adaptive excitation encoding means generates a pitch-synchronous drive excitation vector repeated at a pitch cycle, generates a synthesized speech vector of the adaptive excitation vector based on the encoded spectrum parameter, and calculates an input speech and a first gain. The adaptive excitation vector and the first gain are determined so that the distortion of the synthesized speech vector of the applied adaptive excitation vector is minimized, and the sign of the determined adaptive excitation vector is output. The driving excitation encoding means generates a synthesized speech vector of the pitch-synchronized excitation vector based on the encoded spectral parameters, and applies the determined first gain to the determined adaptive gain. A synthesized speech vector of the pitch-synchronous drive excitation vector to which the synthesized speech vector of the sound source vector and the second gain are added. Is determined as a decoded signal vector, a synthesized voice vector, a pitch position, and a second gain of the pitch synchronous drive excitation vector are determined so that distortion of the input voice and the decoded signal vector is minimized, and the determined pitch is determined. A code of the synchronous drive excitation vector is output, and the determined pitch position and the second gain are encoded. The excitation vector generating means includes at least the determined and encoded adaptive excitation vector, the first gain, and the second gain. A sound source signal is generated and output from a gain and the pitch synchronous drive sound source vector.

A speech decoding apparatus according to a fifth aspect of the present invention
Each frame is provided with a spectrum parameter decoding unit, an adaptive excitation decoding unit, an adaptive excitation codebook, a driving excitation codebook, a pitch synchronization unit, a driving excitation decoding unit, an excitation vector generation unit, and a synthesis filter. A speech decoding apparatus for generating a decoded signal from a spectrum parameter code, an adaptive excitation vector code, a driving excitation vector code, a gain code, and a pitch position code encoded in Receives the code of the spectrum parameter, decodes the spectrum parameter from the code of the spectrum parameter, the adaptive excitation decoding means receives the code of the adaptive excitation vector and the code of the first gain to be given to the adaptive excitation vector, Outputting the code of the adaptive excitation vector to the adaptive excitation codebook, The adaptive excitation codebook generates a plurality of adaptive excitation vectors from the excitation signal output by the excitation vector generation means in the preceding frame and stores the adaptive excitation codebook. An adaptive excitation vector corresponding to the code of the adaptive excitation vector output by the adaptive excitation decoding means is output from the excitation vectors, and the driving excitation codebook includes a plurality of driving excitation vectors having a predetermined pitch synchronization position. Storing the driving excitation vector corresponding to the code of the driving excitation vector output by the driving excitation vector decoding means from the stored driving excitation vectors to the pitch synchronization means, The means receives a code of the pitch position, decodes the pitch position from the code, and decodes the decoded pitch position. The pitch synchronous position is adjusted, a pitch synchronous driving excitation vector is generated by repeating the driving excitation vector input from the driving excitation codebook at a pitch cycle, and the driving excitation decoding means determines the code of the driving excitation vector and the driving excitation vector. The number given to the vector
The two-gain code is input, the code of the excitation vector is output to the excitation codebook, and the second gain is decoded from the code of the second gain. And a pitch-synchronous drive excitation vector to which the second gain has been applied, and outputs a excitation signal, and the synthesis filter uses the decoded spectral parameters to convert a decoded signal from the excitation signal It is generated.

A speech encoding apparatus according to a sixth aspect of the present invention
A spectrum analysis coding unit, an adaptive excitation codebook, a pitch position extracting unit, a driving excitation codebook, a pitch synchronization unit, an adaptive excitation coding unit, a driving excitation coding unit, and an excitation vector generation unit are provided. A speech encoding apparatus for encoding an input speech for each frame, wherein a spectrum analysis encoding means extracts and encodes a spectrum parameter which is spectrum envelope information of an input speech signal, and an adaptive excitation codebook includes A plurality of adaptive excitation vectors are generated and stored in the frame from the excitation signal output by the excitation vector generation unit, and the pitch position extraction unit extracts a plurality of feature points arranged at a pitch cycle interval from the adaptive excitation vector as a pitch position. The driving excitation codebook stores a plurality of driving excitation vectors having a predetermined pitch synchronization position, and The synchronization unit adjusts the pitch synchronization position of each of the driving excitation vectors to the extracted pitch position, generates a pitch synchronization driving excitation vector in which the driving excitation vector is repeated at a pitch cycle, and the adaptive excitation encoding unit includes A synthesized speech vector of the adaptive excitation vector is generated based on the encoded spectral parameters, and the adaptive excitation vector and the synthesized speech vector of the adaptive excitation vector to which the first gain has been applied are minimized. 1 is determined, the code of the determined adaptive excitation vector is output, and the first gain is encoded, and the driving excitation encoding means performs the pitch synchronous driving excitation based on the encoded spectral parameters. The determined adaptive sound obtained by generating a synthesized speech vector of the vector and adding the determined first gain. When the sum of the synthesized voice vector of the vector and the synthesized voice vector of the pitch-synchronized drive excitation vector to which the second gain is added is set as the decoded signal vector, the pitch-synchronized drive is performed so that the distortion between the input voice and the decoded signal vector is minimized. Determining a synthesized speech vector and a second gain of the sound source vector, outputting a code of the determined pitch synchronous drive sound source vector, and encoding the determined second gain; An excitation signal is generated and output from the adaptive excitation vector, the first gain, the second gain, and the pitch-synchronized excitation vector that have been encoded.

A speech decoding apparatus according to a seventh aspect of the present invention
A spectral parameter decoding unit, an adaptive excitation decoding unit, an adaptive excitation codebook, a pitch position extraction unit, a driving excitation codebook, a pitch synchronization unit, a driving excitation decoding unit, an excitation vector generation unit, a synthesis filter, and a fixed interval. A speech decoding apparatus for generating a decoded signal from a code of a spectral parameter encoded for each frame divided into a code of an adaptive excitation vector, a code of a drive excitation vector, and a code of a gain, comprising: Is input the sign of the spectral parameter, decodes the spectral parameter from the sign of the spectral parameter,
The adaptive excitation decoding means receives the code of the adaptive excitation vector and the code of the first gain to be given to the adaptive excitation vector, outputs the code of the adaptive excitation vector to the adaptive excitation codebook, and outputs the first gain. , The adaptive excitation codebook generates and stores a plurality of adaptive excitation vectors from the excitation signal output by the excitation vector generation means in the preceding frame,
An adaptive excitation vector corresponding to the code of the adaptive excitation vector output by the adaptive excitation decoding means is output from the stored adaptive excitation vectors, and the pitch position extracting means outputs the adaptive excitation vector at a pitch cycle interval from the adaptive excitation vector. A plurality of aligned feature points are extracted as pitch positions, the driving excitation codebook stores a plurality of driving excitation vectors having a predetermined pitch synchronization position, and the driving excitation vector is selected from among the stored driving excitation vectors. A driving excitation vector corresponding to the code of the driving excitation vector output by the decoding unit is output to the pitch synchronization unit, and the pitch synchronization unit sets the pitch synchronization of each of the driving excitation vectors to the extracted pitch position. Adjust the pitch and repeat the driving excitation vector input from the driving excitation codebook at the pitch cycle. A driving excitation vector is generated, and the driving excitation decoding means receives the code of the driving excitation vector and the code of the second gain added to the driving excitation vector, and outputs the code of the driving excitation vector to the driving excitation codebook. At the same time, the second gain is decoded from the code of the second gain, and the excitation vector generating means is the sum of the adaptive excitation vector to which the first gain is applied and the pitch synchronous excitation excitation vector to which the second gain is applied. An excitation signal is output, and the synthesis filter generates a decoded signal from the excitation signal using the decoded spectrum parameter.

[0024]

According to the input speech encoding apparatus of the present invention, a periodic pulse excitation vector is newly generated corresponding to the adaptive excitation vector, and each of the periodic pulse excitation vectors is multiplied by a gain to be synthesized. Selected and encoded. In the speech decoding apparatus according to the second aspect of the present invention, an excitation vector having periodicity is generated by the transmitted code and the gain, and the excitation vector is generated by the generated excitation vector and the adaptive excitation vector.
The sound source vector is decoded. In the speech coding apparatus according to the third and fourth aspects of the present invention, the driving excitation vector is periodically repeated, and the vector and the gain that minimize the distortion from the input speech are selected and coded. Further, in the speech decoding apparatus according to the fifth aspect of the present invention, the driving excitation vector is periodically generated by the transmitted code, and the excitation vector is decoded by the generated excitation vector and the adaptive excitation vector. Claim 6
In the speech encoding apparatus according to the invention, a feature point determined by a pitch period in an adaptive excitation vector is extracted as a pitch position, whereby a driving excitation vector is periodically repeated, and a vector and a gain that minimize distortion from input speech are obtained. Is selected and encoded. In the speech decoding apparatus according to the seventh aspect of the present invention, the pitch position is extracted by the transmitted code, the received spectrum parameter and the gain, whereby the driving excitation vector is periodically generated. , The excitation vector is decoded.

[0025]

[Embodiment 1] FIG. 1 is a block diagram of a first embodiment of the present invention.
In FIG. 1, the same portions as those in FIG. 8 are denoted by the same reference numerals, and description thereof will be omitted. In FIG. 1, the new part is
20 multi-vector adaptive excitation coding means, 21 and 23
And a multi-vector adaptive excitation decoding means 22. A vocal tract comprising spectrum analysis means 6, spectrum parameter coding means 7, multiple vector adaptive excitation coding means 20, driving excitation coding means 1
0, the sharing of the vocal chords composed of the sound source vector generation means 12 is the same as in the conventional example, but it is intended to improve the pitch periodicity of the vocal chords in particular and to improve the voice quality after synthesis.

The operation of one embodiment of the present invention will be described below. First, the operation of the encoding unit 1 will be described. The periodic pulse excitation generating means 21 generates a periodic pulse excitation vector composed of a periodic unit impulse train in which only the dimension having the maximum amplitude is 1 and the other dimensions are 0 from the adaptive excitation vector a _i input from the adaptive excitation codebook 9. generate p _i ,
And outputs the p _i into multiple vectors encoding means 20. Originally, the adaptive excitation codebook 9 stores the sampling sequence values of the input up to the previous frame, and from this, the multi-vector adaptive excitation coding means 20 selects the vector that generates the synthesized speech closest to the input speech of the current frame. However, the stored vector of adaptive excitation codebook 9 is not sufficient for maintaining the periodicity. The periodicity is reinforced by newly creating the synchronous pulse sound source vector p _i .

FIG. 2 shows an example of the adaptive excitation vector (a) and the periodic pulse excitation vector (b) generated therefrom in this embodiment. In other words, the vector having the largest amplitude among many stored vectors is unitized and output. This makes it possible to emphasize periodicity that is important for sound reproduction.

The multi-vector adaptive excitation coding means 20 comprises:
The adaptive excitation vector a _i input from the adaptive excitation codebook 9 and the periodic pulse excitation vector p _i input from the periodic pulse excitation generating means 21 are used, using the spectrum parameters input from the spectrum parameter encoding means 7.
Then, the synthesized speech vectors A _i and P _i are synthesized. Then, the inter-vector distance D _i between the input speech vector X cut out for each frame from the input speech 4, for example, the formula (5)
Ask according to. Here, the gains β _Ai and β _Pi are the distances _Di
Is determined according to, for example, Expressions (6) and (7) so that is minimized.

[0029]

(Equation 3)

Next, a set of vectors A _I and P _I that minimizes the inter-vector distance is searched for, and the sign I and the signs of the gains β _qAI and β _{qPI obtained} by quantizing the gains β _AI and β _PI are transmitted through the transmission line. 3 to the decoding unit 2. Also, the selected adaptive _excitation vector a _I and its gain β _qAI and the periodic pulse _excitation vector p _I and its gain β _qPI are output to the excitation vector generation means 12 and X ′ = X as an error vector X ′. −β _qAI A _I −β _qPI P _I is output to driving _excitation coding means 10. Thus, as compared with the conventional, it is novel that the vector of A _I and the set of vectors of A _I and P _I. The operation of the driving _excitation encoding means 10 for further quantizing the vector difference and encoding and transmitting the same is the same as that of the conventional example in the present embodiment. Send.

Next, the operation of the decoding unit 2 will be described. The multi-vector adaptive excitation decoding means 22 outputs the code I of the adaptive excitation vector input from the encoding unit 1 to the adaptive excitation codebook 14. The adaptive excitation codebook 14 converts the adaptive excitation vector a _I corresponding to the code I into a multi-vector adaptive excitation decoding means 22 and a periodic pulse excitation generating means 23.
Output to The periodic pulse excitation generating means 23 generates a periodic pulse excitation vector p _I from the adaptive excitation vector a _I input from the adaptive excitation codebook 14, and outputs it to the multi-vector adaptive excitation decoding means 22.

The multi-vector adaptive excitation decoding means 22 comprises:
From the adaptive excitation vector a _I input from the adaptive excitation codebook 14, the periodic pulse excitation vector p _I input from the periodic pulse excitation generating means 23, and the sign of each gain input from the encoding unit 1. The decoded gains β _qAI and β _qPI are output to the _excitation vector generation means 17.
That is, a set of β _qAI , a _I , β _qPI , and p _I is given to the sound source vector generation means.

The excitation vector generation means 17 receives the adaptive _excitation vector a _I , the periodic pulse _excitation vector p _I and the respective gains β _qAI and β _qPI input from the multi-vector adaptive _excitation decoding means 22 and the other input. The following _excitation vector is generated from the driving _excitation vector c _J and the gain γ _qJ input from a certain driving _excitation decoding means 15. β _qAI a _I + β _qPI p _I + γ _qJ c _{J The} generated _excitation vector is output to adaptive _excitation codebook 14 and synthesis filter 19. The output of the synthesis filter 19 is an audio output that is synthesized and restored.

Embodiment 2 FIG. In the first embodiment, the periodic pulse excitation vector is generated in consideration of only the pitch periodicity in the frame to be encoded. On the other hand, the pitch period of the adaptive excitation vector that is cut out and selected / encoded takes a value that is twice the actual pitch period of the input speech, and the pulse train generated by the periodic pulse excitation vector is not necessarily periodic between frames. It does not become. At this time, using the information of the preceding frame, for example, an average pitch period is obtained, and the periodic pulse excitation vector is generated from a pulse train having the average pitch period.
The periodic pulse excitation vector may be generated in consideration of the pitch periodicity between frames.

Embodiment 3 FIG. In the first embodiment, the amplitude of each pulse of the periodic pulse sound source vector is assumed to be all equal. By changing this, for example, according to the power of the sound, the amplitude is sequentially increased or decreased from the previous frame, such as linearly changing, such that each periodic pulse sound source vector is set to have a different amplitude. Is also good.

Embodiment 4 FIG. In the first embodiment, the pulse position of the periodic pulse sound source vector is obtained from the maximum amplitude point of the adaptive sound source vector. However, in a synthesized speech vector obtained by synthesizing the two, the inner product of both vectors becomes maximum. A periodic pulse sound source vector may be set.

Embodiment 5 FIG. In the first embodiment, a synthesized speech vector is generated using the adaptive excitation vector and the periodic pulse excitation vector as they are,
As a result, the inter-vector distance between the vector and the input speech vector is obtained. However, for example, if the amplitude value of the adaptive excitation vector at the pulse position of the periodic pulse excitation vector is 0,
Then, the adaptive excitation vector and the periodic pulse excitation vector may be orthogonalized to remove the correlation between the vectors, and for the periodicity, a synthetic speech vector using only the periodic pulse excitation vector may be generated. Then, the inter-vector distance between the result vector and the input speech vector may be obtained.

Embodiment 6 FIG. In the fifth embodiment, the adaptive excitation vector and the periodic pulse excitation vector are orthogonalized to remove their correlation. The correlation between vectors may be removed by orthogonalizing synthesized speech vectors obtained by synthesizing them.

Embodiment 7 FIG. In the first embodiment, the periodic pulse sound source vector is composed of one vector. In this case, when the gain of each pulse of the periodic pulse sound source vector has two pulses in FIG. 2B, the gain before and after the pulse may be set to different values in terms of time. .

Embodiment 8 FIG. FIG. 3 is a block diagram showing one embodiment of the third and fifth aspects of the present invention. 3, the same parts as those in FIG. 1 are denoted by the same reference numerals, and description thereof will be omitted. FIG.
, 24 is an adaptive excitation coding means, 25 is a pitch position extraction means, and 26 is a driving excitation coding means. 27,
28 is a pitch synchronization means, and 29 is a sound source vector generation means.

The operation of the embodiment of the present invention shown in FIG. 3 will be described below. First, the operation of the encoding unit 1 will be described. Adaptive excitation coding means 24 synthesizes synthesized speech vector A _i using adaptive excitation vector a _i input from adaptive excitation codebook 9 and spectrum parameters input from spectrum parameter coding means 7. Then, a search is made for the synthesized speech vector A _I and its gain β _I that minimize the distortion with respect to the input speech vector X cut out for each frame from the input speech 4, and its sign I and gain β _I are searched for.
_Is output to the decoding unit 2 via the transmission path 3 and the selected adaptive _excitation vector a _I and its gain β _qI are output to the excitation vector generation unit 12.
Further, the error vector X ′ is set to X ′ = X−β _qI A _I , and the X ′ and the adaptive _excitation vector a _I are output to the driving excitation coding unit 26 and the pitch corresponding to the adaptive excitation vector a _I Cycle is pitch position extraction means 2
5 is output.

The pitch position extracting means 25 creates a pulse train of the pitch cycle using the pitch cycle input from the adaptive excitation coding means 24 and converts the spectrum parameter input from the spectrum parameter coding means 7 into a pulse train using this as a sound source. Is used to generate a synthesized speech vector, an input speech vector X cut out from the input speech 4 for each frame.
Search for a pulse train that minimizes distortion. Then, the pulse synchronizing means 27 sets the pulse position as a pitch position.
Output to

FIG. 4 is an operation block diagram at the time of searching for a pitch position. In this operation, the start position of the pulse of the pitch cycle is divided from the start position of the frame by, for example, the cutout time of one frame by 128 and shifted by 1/128 hours. Then, a combination of a pulse train with a pitch cycle shifted from the head position is selected as the one closest to the input voice. By doing so, a point that is uniquely determined with respect to the pitch structure of the voice is determined regardless of the framing of the input voice, so that the periodicity can be clarified.

In the driving excitation codebook 11, for example, as a result of learning by the Linde, Buzo, Gray method (LBG method) from a short-period prediction residual signal cut out at a pitch length synchronized with the pitch position of speech, N Are stored. In addition, a pitch position obtained by cutting out each drive sound source vector is set as a pitch synchronization position of the drive sound source vector. The pitch synchronization unit 27 adjusts the pitch synchronization position of the driving excitation vector c _j input from the driving excitation codebook 11 to the pitch position input from the pitch position extraction unit 25, and repeats the pitch synchronization driving at a pitch cycle. Create a sound source vector c _j ′. Then, this is output to driving excitation coding means 26. FIG. 5 shows an example of the pitch synchronous drive sound source vector in the present embodiment. The driving excitation coding means 26 converts the pitch-synchronized driving excitation vector c _j ′ input from the pitch synchronization means 27 into the adaptive excitation vector a _I input from the adaptive excitation coding means 24.
Then, for example, an orthogonalized pitch synchronous drive excitation vector c _j ″ that is orthogonalized according to equation (8) is created.

[0045]

(Equation 4)

Next, a synthesized speech vector C _j ″ is synthesized from the orthogonalized pitch synchronous drive excitation vector c _j ″. The inter-vector distance D _j with the error vector X ′ input from the adaptive excitation coding means 24 is calculated, for example, by using equation (9).
Ask according to. Here, the gain γ _j is determined according to, for example, equation (10) so as to minimize the inter-vector distance D _j .

[0047]

(Equation 5)

Next, a vector C _J ″ in which the inter-vector distance D _j is minimized is searched for, and the sign J and the gain γ _J are searched for.
_Is output to the decoding unit 2 via the transmission path 3. At the same time, the selected orthogonalized pitch synchronous drive _excitation vector c _J ″ and its gain γ _qJ are output to the _excitation vector generation means 12. This means that the _period of the _excitation signal generated because it was conventionally asynchronous with the pitch. The periodicity of the sound source signal is improved by synchronizing the driving sound source vector, which has been disturbed, with the pitch.

Next, the operation of the decoding unit 2 will be described. The pitch synchronization means 28 creates a pitch-synchronous driving excitation vector c _J ′ that is repeated in synchronization with the pitch position input from the encoding unit 1 from the driving excitation vector c _J input from the driving excitation codebook 16. , To the driving excitation decoding means 15.

The excitation vector generating means 29 first orthogonalizes the pitch synchronous driving excitation vector c _J ′ input from the driving excitation decoding means 15 with the adaptive excitation vector a _I input from the adaptive excitation decoding means 13. To generate the orthogonalized pitch synchronous drive _excitation vector c _J ″. Next, the adaptive _excitation vector a _I and the gain β _qI input from the adaptive _excitation decoder 13 and the orthogonalized pitch synchronous drive _excitation vector c c "than the gain gamma _QJ input from and excitation decoding means 15, excitation vector _{β qI a I + γ qJ c} J of the formula" _J, and outputs it to the adaptive excitation codebook 14 and synthesis filter 19 The output of the synthesis filter 19 is an audio output that is synthesized and restored.

Embodiment 9 FIG. 6 is a block diagram showing an embodiment of the invention according to claims 4 and 5 of the present invention. 6, the same parts as those in FIG. 3 are denoted by the same reference numerals, and description thereof will be omitted. FIG.
In the above, reference numeral 30 denotes a driving excitation coding means.

Hereinafter, an embodiment of the present invention shown in FIG. 6 will be described. Adaptive excitation coding means 24 synthesizes synthesized speech vector A _i using adaptive excitation vector a _i input from adaptive excitation codebook 9 and spectrum parameters input from spectrum parameter coding means 7.
Then, a search is made for a vector A _I and its gain β _I that minimize the distortion with respect to the input speech vector X cut out from the input speech 4 for each frame, and the sign I and the gain β _{qI obtained} by quantizing the gain β _I are _obtained . The code is output to the decoding unit 2 via the transmission path 3, the selected adaptive _excitation vector a _I and its gain β _qI are output to the excitation vector generation unit 12, and the error vector X ′ is _calculated as _follows : X ′ = X −β _qI A _I, and _outputs X ′ and the pitch period corresponding to the adaptive _excitation vectors a _I and a _I to the driving excitation encoding means 30.

The driving excitation coding means 30 first outputs the pitch position k to the pitch synchronization means 27 using the pitch period input from the adaptive excitation coding means 24. This pitch position k is a position determined by the pitch cycle starting from the k-th point from the head position of the frame. Pitch synchronization means 2
Reference numeral 7 denotes a pitch-synchronous driving excitation vector that repeats at a pitch cycle by adjusting the pitch synchronization position of the driving excitation vector c _j input from the driving excitation codebook 11 to the pitch position k input from the driving excitation encoding means 30. Create c _jk '. Then, this is output to driving excitation coding means 30. The driving excitation encoding means 30 outputs the pitch synchronous driving _excitation vector c _jk ′ input from the pitch synchronization means 27.
_Is generated with respect to the adaptive excitation vector a _I input from the adaptive excitation encoding means 24, for example, in accordance with equation (11) to create an orthogonalized pitch synchronous drive _excitation vector c _jk ″.

[0054]

(Equation 6)

Next, a synthesized speech vector C _jk ″ is synthesized from the orthogonalized pitch synchronous drive _excitation vector c _jk ″. Then, the inter-vector distance D _jk with respect to the error vector X ′ input from the adaptive excitation coding means 24 is calculated by, for example, the equation (1)
Request according to 2). Here, the gain γ _jk is determined according to, for example, Expression (13) so as to minimize the inter-vector distance D _jk .

[0056]

(Equation 7)

Next, the inter-vector distance _{Djk is determined} by the code j (j = 1,..., N) of the driving sound source vector and the pitch position k.
(K = 1,..., L), a vector C _JK ″ that minimizes the inter-vector distance D _jk is searched for, and this code J, pitch position K, and gain γ _JK are determined. The quantized γ _qJK code is transmitted to the decoding unit 2 via the transmission path 3.
Output to At the same time, the selected orthogonalized pitch synchronous drive _excitation vector c _JK ″ and its gain γ _qJK are output to the _excitation vector generation unit 12. This is _achieved by extracting the pitch position as the one that minimizes the distortion of the synthesized speech. Thus, the sound quality of the synthesized voice is improved.

Embodiment 10 FIG. FIG. 7 is a block diagram showing an embodiment of the invention according to claim 6 of the present invention. 7, the same parts as those in FIG. 6 are denoted by the same reference numerals, and the description will be omitted. In FIG. 7, 3
1 is an adaptive excitation coding means, and 32 and 33 are pitch position extracting means.

The operation of the embodiment of the present invention shown in FIG. 7 will be described below. First, the operation of the encoding unit 1 will be described. The pitch position extracting means 32 synthesizes the synthesized speech vector A _i using the adaptive excitation vector a _i inputted from the adaptive excitation codebook 9 and the spectrum parameter inputted from the spectrum parameter encoding means 7. Next, when a synthetic speech vector is generated using the spectral parameters by using a pulse train having a pitch cycle corresponding to the adaptive sound source vector a _i as a sound source, a pulse train that minimizes distortion from the synthetic speech vector A _i is searched for. I do. Then, the pulse position is output to adaptive excitation encoding means 31 as a pitch position.

The adaptive excitation coding means 31 first uses the pulse sequence of the pitch period corresponding to the adaptive excitation vector a _i input from the adaptive excitation codebook 9 as the excitation, and uses the spectrum input from the spectrum parameter coding means 7 as the excitation. When a synthesized speech vector is generated using the parameters, a pulse train that minimizes distortion from the input speech vector X cut out from the input speech 4 for each frame is searched. Then, the pulse position is set as a pitch position in the input voice. The following processing is performed only when the pitch position in the adaptive sound source vector a _i input from the pitch position extraction means 32 is within a certain distance, for example, 5 samples, from the pitch position in the input voice. In this case, the adaptive excitation vector a _i is excluded from the encoding target.

Next, adaptive excitation coding means 31 generates a pitch pulse vector p _{i in} which only the pitch position in adaptive excitation vector a _i is 1 and the others are 0. Then, using the spectral parameters, synthesized speech vectors A _i and P _i are synthesized from the adaptive sound source vector a _i and the pitch pulse vector p _i . Also, determine the inter-vector distance D _i between the input speech vector X, for example, in accordance with equation (14). Here, min (x, y) is a function for selecting the smaller value of x and y, and the selection result is output.
The gain beta _Ai, beta _Pi is such that the distance D _i is minimized, for example, formula (15), determined in accordance with equation (16).

[0062]

(Equation 8)

Next, the vector A _{I in} which the distance between the vectors is minimized is searched, and the code I and the code of the gain β _{qAI obtained} by quantizing the gain β _AI are transmitted to the decoding unit 2 via the transmission path 3. Output. The adaptive _excitation vector a _I and its gain β _qAI are output to the excitation vector generation means 12. Further, the pitch position in the adaptive sound source vector a _I input from the pitch position extracting means 32 is output to the pitch synchronizing means 27. Also, the error vector X '
_{Is set} to X ′ = X−β _qAI A _I, and this X ′ and the adaptive _excitation vector a _I are also output to the driving excitation encoding means 26.

Next, the operation of the decoding unit 2 will be described. The pitch position extracting means 33 extracts a pitch position from the adaptive excitation vector a _I input from the adaptive excitation codebook 14 and the spectrum parameter input from the spectrum parameter decoding means 18. Then, this pitch position is output to the pitch synchronization means 28. This means that the pitch position can be uniquely obtained even in the decoding unit by extracting the pitch position from the adaptive excitation vector and the spectrum parameter, which are obtained by quantizing the input voice, without transmitting the pitch position. This reduces the amount of transmitted information without deteriorating the sound quality of the synthesized speech.

Embodiment 11 FIG. In the tenth embodiment, in the adaptive excitation coding means, the distance reference between the pitch position extracted from the adaptive excitation vector and the pitch position obtained from the input speech is fixed.
For example, it may be variable according to the pitch length.

Embodiment 12 FIG. In the tenth embodiment, the adaptive excitation encoding means encodes only the difference between the pitch position extracted from the adaptive excitation vector and the pitch position obtained from the input speech within a certain distance. For frames whose vectors do not meet this criterion, this distance criterion may be relaxed or abolished.

[0067]

As described above, according to the present invention, a speech encoding apparatus extracts and encodes a spectrum parameter and encodes a spectrum parameter, and an adaptive excitation source for generating and storing a plurality of adaptive excitation vectors. A codebook, a periodic pulse excitation generating means for generating a periodic pulse excitation vector, an adaptive excitation vector that minimizes distortion with the input speech, and a gain to be applied to the adaptive excitation vector and the periodic pulse excitation vector. Since the multi-vector adaptive excitation encoding means for encoding and the excitation vector generation means for generating and outputting an excitation signal are provided, and the speech decoding apparatus is provided with means corresponding to the encoding apparatus, By simply adding the transmission information, it is possible to synthesize a high-quality voice with little disturbance of the pitch periodicity.

According to the third to fifth aspects of the present invention, the speech encoding device includes a spectrum parameter analysis encoding means for extracting and encoding a spectrum parameter.
Pitch excitation extracting means, an adaptive excitation codebook, a driving excitation codebook storing a plurality of driving excitation vectors having a predetermined pitch synchronization position, a pitch synchronization means for repeating the driving excitation vector at a pitch cycle, Encoding means, driving excitation excitation means for determining and encoding a pitch-synchronous excitation vector and a second gain for minimizing distortion between an input speech and a decoded signal vector, and excitation vector generation for generating and outputting an excitation signal Means, and the speech decoding apparatus has means corresponding to the encoding apparatus,
By adding only a small amount of transmission information, there is an effect that high-quality speech with little disturbance of pitch periodicity can be synthesized.

According to the sixth and seventh aspects of the present invention, the speech coding apparatus includes a spectrum parameter analysis coding means for extracting and coding a spectrum parameter;
Pitch position extracting means for extracting a pitch position from an adaptive excitation vector, an adaptive excitation codebook, a driving excitation codebook storing a plurality of driving excitation vectors having a predetermined pitch synchronization position, and a driving excitation vector at a pitch period. Pitch synchronizing means to be repeated, adaptive excitation coding means, driving excitation coding means for determining and coding a pitch-synchronized driving excitation vector and a second gain for minimizing distortion of input speech and a decoded signal vector, and excitation signal Sound source vector generating means for generating and outputting
Since means corresponding to the encoding device is provided, there is an effect that high-quality speech with little disturbance of pitch periodicity can be synthesized without increasing transmission information.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing a first embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of a periodic pulse sound source vector according to the present invention.

FIG. 3 is a configuration diagram showing an eighth embodiment of the present invention.

FIG. 4 is an explanatory diagram showing pitch position extraction according to the present invention.

FIG. 5 is an explanatory diagram showing an example of a pitch synchronous drive sound source vector according to the present invention.

FIG. 6 is a configuration diagram showing a ninth embodiment of the present invention.

FIG. 7 is a configuration diagram showing a tenth embodiment of the present invention.

FIG. 8 is a configuration diagram showing a conventional speech encoding / decoding device.

FIG. 9 is an explanatory diagram showing an example of an adaptive sound source vector.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Encoding part 2 Decoding part 3 Transmission path 4 Input speech 5 Output speech 6 Spectrum analysis means 7 Spectrum parameter encoding means 8 Adaptive excitation encoding means 9, 14 Adaptive excitation codebook 10 Driving excitation encoding means 11, 16 Driving Excitation codebook 12, 17 Excitation vector generation means 13 Adaptive excitation decoding means 15 Driving excitation decoding means 18 Spectral parameter decoding means 19 Synthetic filter 20 Multiple vector adaptive excitation coding means 21, 23 Periodic pulse excitation generation means 22 Multiple vectors Adaptive excitation decoding means 24 Adaptive excitation coding means 25 Pitch position extraction means 26 Driving excitation coding means 27, 28 Pitch synchronization means 29 Excitation vector generation means 30 Driving excitation coding means 31 Adaptive excitation coding means 32, 33 Pitch Position extraction means

Claims (7)

(57) [Claims] 1. An apparatus comprising spectrum analysis coding means, adaptive excitation codebook, periodic pulse excitation generation means, multi-vector adaptive excitation encoding means, excitation vector generation means, and encodes input speech for each frame divided at regular intervals. Wherein the spectrum analysis encoding means extracts and encodes a spectrum parameter which is spectrum envelope information of the input speech, and the adaptive excitation codebook is output by the excitation vector generation means in the preceding frame. A plurality of adaptive excitation vectors are generated from the generated excitation signal and stored, and the periodic pulse excitation generating means outputs a periodic pulse excitation vector determined as a main component of the adaptive excitation vector output from the adaptive excitation codebook, and Excitation encoding means includes the encoded spectral parameters and the plurality of adaptive excitation The distortion between the input speech and the decoded speech generated from any one of the tors and the first gain given to the adaptive excitation vector, the periodic pulse excitation vector and the second gain given to the periodic pulse excitation vector is minimized. The adaptive excitation vector, the first gain and the second gain are determined and encoded so that the adaptive excitation vector, the determined and encoded adaptive excitation vector, the first gain and the second gain, and A speech encoding device that generates and outputs a sound source signal from a periodic pulse sound source vector. 2. An apparatus comprising a spectrum parameter decoding means, a multi-vector adaptive excitation decoding means, an adaptive excitation codebook, a periodic pulse excitation generation means, an excitation vector generation means, and a synthesis filter, and encodes each frame divided at regular intervals. A speech decoding apparatus for generating a decoded signal from the obtained code of the spectrum parameter, the code of the adaptive excitation vector, and the code of the gain, wherein the spectrum parameter decoding means receives the code of the spectrum parameter, The multi-vector adaptive excitation decoding means receives the code of the adaptive excitation vector and the code of the gain, and outputs the code of the adaptive excitation vector to the adaptive excitation codebook. Decoding the first gain and the second gain from the code, Are generated and stored in the preceding frame from the excitation signal output by the excitation vector generation means, and output from the stored adaptive excitation vector by the multi-vector adaptive excitation decoding means. The periodic pulse excitation generating means outputs a periodic pulse excitation vector determined as a main component of the adaptive excitation vector output from the adaptive excitation codebook, and outputs an excitation vector corresponding to the code of the adaptive excitation vector. The generating means adds at least the adaptive excitation vector to which the first gain has been added and the periodic pulse excitation vector to which the second gain has been added, and outputs the sum as an excitation signal. The synthesis filter uses the decoded spectral parameters. And a speech decoding apparatus for generating a decoded signal from the sound source signal. 3. A spectrum analysis encoding means, a pitch position extracting means, an adaptive excitation codebook, a driving excitation codebook, a pitch synchronization means, an adaptive excitation encoding means, a driving excitation encoding means,
A speech encoding apparatus comprising a sound source vector generation unit and encoding an input speech for each frame divided at a fixed interval, wherein the spectrum analysis encoding unit extracts a spectrum parameter which is spectrum envelope information of the input speech. The pitch position extracting means extracts a plurality of feature points arranged at a pitch period interval from the input speech as pitch positions and encodes them, and the adaptive excitation codebook is output by the excitation vector generating means in the preceding frame. A plurality of adaptive excitation vectors are generated and stored from the excitation signal, the driving excitation codebook stores a plurality of driving excitation vectors having a predetermined pitch synchronization position, and the pitch synchronization unit controls the encoded pitch position. To the pitch synchronization position of each drive sound source vector, and repeat the drive sound source vector at the pitch cycle. The adaptive excitation encoding means generates a synthesized speech vector of the adaptive excitation vector based on the encoded spectral parameters, and applies the input speech and the first gain to the adaptive excitation vector. The adaptive excitation vector and the first gain are determined so as to minimize the distortion of the synthesized speech vector, and the code of the determined adaptive excitation vector is output.
The driving excitation encoding means generates a synthesized speech vector of the pitch-synchronized driving excitation vector based on the encoded spectral parameters, and applies the determined first gain. When the sum of the synthesized speech vector of the adaptive excitation vector and the synthesized speech vector of the pitch-synchronized excitation vector to which the second gain is applied is defined as a decoded signal vector, the input speech and the decoded signal vector are minimized in distortion. Determining a synthesized speech vector and a second gain of the pitch-synchronous drive excitation vector, outputting a code of the determined pitch-synchronous drive excitation vector, and encoding the determined second gain; At least the determined and encoded adaptive excitation vector, the first gain and the second gain, and the pitch synchronization Speech encoding apparatus for generating and outputting a sound source signal from Doongen vector. 4. An apparatus comprising spectrum analysis coding means, adaptive excitation codebook, driving excitation codebook, pitch synchronization means, adaptive excitation coding means, driving excitation coding means, excitation vector generation means, and divided at regular intervals. A speech encoding apparatus for encoding an input speech for each frame, wherein the spectrum analysis encoding means extracts and encodes a spectrum parameter which is spectrum envelope information of the input speech, A plurality of adaptive excitation vectors are generated and stored from the excitation signal output by the excitation vector generation means, and the driving excitation codebook stores a plurality of driving excitation vectors having a predetermined pitch synchronization position. Adjusts the pitch synchronization position of the driving sound source vector to the pitch position, which is a plurality of points arranged at a pitch cycle interval on the time axis, Generating a pitch-synchronous drive excitation vector obtained by repeating the dynamic excitation vector at a pitch cycle, the adaptive excitation encoding means generates a synthesized speech vector of the adaptive excitation vector based on the encoded spectral parameters, and The adaptive excitation vector and the first gain are determined so that the distortion of the synthesized speech vector of the adaptive excitation vector to which the first gain is added is minimized, and the code of the determined adaptive excitation vector is output, and the determined adaptive excitation vector is output. The first excitation is encoded, and the driving excitation encoding means generates a synthesized speech vector of the pitch synchronous excitation vector based on the encoded spectral parameters, and provides the determined first gain. The synthesized speech vector of the determined adaptive sound source vector and the pitch synchronous drive sound source vector to which the second gain is added. When the sum of the synthesized speech vectors is used as the decoded signal vector, the synthesized speech vector, pitch position, and second gain of the pitch synchronous drive excitation vector are determined so that the distortion of the input speech and the decoded signal vector is minimized. Outputting the code of the determined pitch-synchronous drive excitation vector and encoding the determined pitch position and the second gain, wherein the excitation vector generating means includes at least the determined and encoded adaptive excitation vector and the first A speech encoding device for generating and outputting a sound source signal from a gain, a second gain, and the pitch synchronized drive sound source vector. 5. An apparatus comprising spectrum parameter decoding means, adaptive excitation decoding means, adaptive excitation codebook, driving excitation codebook, pitch synchronization means, driving excitation decoding means, excitation vector generation means, synthesis filter, and a fixed interval. An audio decoding device that generates a decoded signal from a code of a spectrum parameter encoded for each frame divided into a code of an adaptive excitation vector, a code of a drive excitation vector, a code of a gain, and a code of a pitch position, The spectrum parameter decoding means receives a code of the spectrum parameter, decodes the spectrum parameter from the code of the spectrum parameter, and the adaptive excitation decoding means determines the code of the adaptive excitation vector and the first gain to be given to the adaptive excitation vector. When a code is input and the code of the adaptive excitation vector is output to the adaptive excitation codebook, In addition, the first gain is decoded from the code of the first gain, and the adaptive excitation codebook generates and stores a plurality of adaptive excitation vectors from the excitation signal output by the excitation vector generation means in the preceding frame, An adaptive excitation vector corresponding to the code of the adaptive excitation vector output by the adaptive excitation decoding means is output from the stored adaptive excitation vectors, and the driving excitation codebook has a predetermined pitch synchronization position. A plurality of drive excitation vectors are stored, and a drive excitation vector corresponding to the code of the drive excitation vector output by the drive excitation vector decoding unit is output from the stored drive excitation vectors to the pitch synchronization unit. The pitch synchronization means receives the code of the pitch position, decodes the pitch position from the code, and The pitch synchronization position is adjusted to the calculated pitch position, and a pitch synchronization driving excitation vector is generated by repeating the driving excitation vector input from the driving excitation codebook at a pitch cycle. A code and a code of a second gain to be given to the driving excitation vector are input, and the code of the driving excitation vector is output to the driving excitation codebook, and a second gain is decoded from the code of the second gain, The excitation vector generation means outputs an excitation signal which is the sum of the adaptive excitation vector to which the first gain is applied and the pitch-synchronous drive excitation vector to which the second gain is applied, and the synthesis filter converts the decoded spectrum parameter to A speech decoding apparatus for generating a decoded signal from the sound source signal using the speech decoding apparatus. 6. A spectrum analysis coding means, an adaptive excitation codebook, a pitch position extracting means, a driving excitation codebook, a pitch synchronization means, an adaptive excitation coding means, a driving excitation coding means,
An audio encoding apparatus comprising an excitation vector generation unit and encoding input audio for each frame divided at a fixed interval, wherein the spectrum analysis encoding unit extracts a spectrum parameter which is spectrum envelope information of the input audio signal. The adaptive excitation codebook generates and stores a plurality of adaptive excitation vectors from the excitation signal output by the excitation vector generating means in the preceding frame, and the pitch position extracting means calculates a pitch period from the adaptive excitation vector. A plurality of feature points arranged at intervals are extracted as a pitch position, the driving excitation codebook stores a plurality of driving excitation vectors having a predetermined pitch synchronization position, and the pitch synchronization unit stores the plurality of feature points in the extracted pitch position. The pitch synchronized position of each drive sound source vector is adjusted, and the drive sound source vector is repeated at the pitch cycle. The adaptive excitation encoding means generates a synthesized speech vector of the adaptive excitation vector based on the encoded spectral parameters, and applies the input speech and the first gain to the adaptive excitation. The adaptive excitation vector and the first gain are determined so that the distortion of the synthesized speech vector of the vector is minimized, the code of the determined adaptive excitation vector is output, and the first gain is encoded. Generates a synthesized speech vector of the pitch-synchronous drive excitation vector based on the encoded spectral parameters, and generates a synthesized speech vector of the determined adaptive excitation vector to which the determined first gain is added and a second speech vector. When the sum of the synthesized voice vector of the pitch synchronous drive excitation vector with the gain given is a decoded signal vector, Determining a synthesized speech vector and a second gain of the pitch-synchronous drive excitation vector so as to minimize distortion of the power speech and the decoded signal vector; outputting a code of the determined pitch-synchronous drive excitation vector; The excitation vector generating means generates an excitation signal from at least the determined and encoded adaptive excitation vector, the first gain and the second gain, and the pitch-synchronized excitation excitation vector. An audio encoding device to output. 7. Spectral parameter decoding means, adaptive excitation decoding means, adaptive excitation codebook, pitch position extraction means, driving excitation codebook, pitch synchronization means, driving excitation decoding means, excitation vector generation means, synthesis filter With
An audio decoding apparatus for generating a decoded signal from a code of a spectral parameter encoded for each frame divided at a fixed interval, a code of an adaptive excitation vector, a code of a driving excitation vector, and a code of a gain, comprising: The coding means receives the code of the spectrum parameter and decodes the spectrum parameter from the code of the spectrum parameter. The adaptive excitation decoding means sets the code of the adaptive excitation vector and the code of the first gain to be given to the adaptive excitation vector. The adaptive excitation codebook is input and outputs the code of the adaptive excitation vector to the adaptive excitation codebook, and decodes the first gain from the code of the first gain. While generating and storing a plurality of adaptive sound source vectors from the output sound source signal, An adaptive excitation vector corresponding to the code of the adaptive excitation vector output by the adaptive excitation decoding means is output from the stored adaptive excitation vectors, and the pitch position extracting means is arranged at a pitch cycle interval from the adaptive excitation vector. A plurality of feature points are extracted as pitch positions. The driving excitation codebook stores a plurality of driving excitation vectors having a predetermined pitch synchronization position, and decodes the driving excitation vector from among the stored driving excitation vectors. A drive excitation vector corresponding to the sign of the drive excitation vector output by the synchronization means is output to the pitch synchronization means, and the pitch synchronization means adjusts the pitch synchronization position of each of the drive excitation vectors to the extracted pitch position. And the pitch obtained by repeating the driving excitation vector input from the driving excitation codebook at the pitch cycle. A driving excitation vector is generated, and the driving excitation decoding unit receives the code of the driving excitation vector and the code of the second gain to be given to the driving excitation vector, and outputs the code of the driving excitation vector to the driving excitation codebook. At the same time, the second gain is decoded from the code of the second gain, and the excitation vector generation means is the sum of the adaptive excitation vector to which the first gain is applied and the pitch synchronous excitation excitation vector to which the second gain is applied. A speech decoding device that outputs a sound source signal, and a synthesis filter generates a decoded signal from the sound source signal using the decoded spectrum parameter.